CN104010676A - Image processing, frequency estimation, mechanical control and illumination for an automatic IV monitoring and controlling system - Google Patents

Abstract

A device, a mechanical apparatus and an illumination system for intravenous (IV) monitoring system are provided. The device extracts a periodical signal from IV dripping process by using any video or image processing techniques and measures the speed of the dripping by using any frequency estimation techniques. The mechanical apparatus controls the speed of the dripping by changing the thickness or diameter of the IV tube according to the IV dripping speed measured by the video or image processing. The illumination system illuminates the drip chamber so that clear image can be captured for the video or image processing.

Description

Automatically IV monitoring is controlled and illumination with image processing, frequence estimation, the machinery of control system

The cross reference of related application

U. S. application 12825368: carry out IV monitoring by Digital Image Processing, same inventor submits to

U. S. application 12804163: processed and carried out IV monitoring by video and image, same inventor submits to

U. S. application 13019698: the electro-mechanical system of controlling for IV, same inventor submits to

U. S. application 13356632: IV monitoring is controlled and illumination with image processing, frequence estimation, the machinery of control system automatically, and same inventor submits to

Federal government subsidizes research

Inapplicable

Each party name of other of joint study agreement

Inapplicable

Sequence table or program

Inapplicable

Background technology-technical field

The present invention relates to IV monitoring and control system, by video and image finish dealing with the monitoring of this system, dropping speed that frequency of utilization estimating techniques are measured this system, with mechanical component, control the drop speed of this system and with optical module, complete the illumination of this system.

Background technology-prior art

First, in infusion pump, be widely used as automatic IV control appliance.Most of infusion pump are not monitored IV speed but are used plant equipment (modal is peristaltic pump) to carry out control rate.

Secondly, also there is the trial of much monitoring IV speed via optics (modal is infrared or other sensors), image processing or other modes.

Summary of the invention

We have described further method and apparatus in the disclosure.

Process video/image for extracting the image processing techniques of the periodic measurement of IV drop process, disclosed method comprises:

1. image enhancement technique, this further comprises that greyscale transformation, frequency domain process and wavelet technique.

2. threshold process technology, this further comprises alternative manner, arbitrarily/constant or manual allocation/definite threshold size and average/intermediate value or other simple threshold process methods.

Frequence estimation will be estimated drop frequency in the periodic signal from extracting in image sequence, disclosed method comprises:

1. nonparametric technique, this further comprises simple time domain approach, Time-domain Statistics method, Fourier and Fourier correlation method and wavelet transformation.

2. parametric technique, this further comprises automatic returning or automatic returning average-average (mean-average) frequency spectrum evaluation method and characteristic vector/subspace method.

Mechanical system is controlled dropping speed by pressing pipe.Device comprises pipe depressor and supporter, uses the embodiment of driving screw and difference driving screw, use lever, use nonlinear motion guiding element, rotary push device and cam.

Illuminator is guaranteed caught video/image (a plurality ofs') quality.Principle, method and apparatus comprise principle that reflection/luminance contrast reduces, multiple light courcess, from a plurality of sources of secondary source, from the light source of direct reflection, from the light source of the amplification of lens, use any smoothness reflecting surface, avoid taking this reflection/luminance contrast and use light guide/stopper.

Please refer to the corresponding chapters and sections of each discussed method, device.

Accompanying drawing-Tu

Image and Fig. 1 .1-1B that Fig. 1 .1-1A illustrates present drip chamber illustrate the region of carrying out graphical analysis in Fig. 1 .1-1A thereon.

Fig. 1 .1-2A illustrates vertical Sobel gradient.

Fig. 1 .1-2B illustrates vertical Prewitt gradient.

Fig. 1 .1-2B illustrates Laplce's () operator.

Fig. 1 .1-3 illustrates image and Sobel, Prewitt and Laplce's result.

Fig. 1 .1-4A is illustrated in the analytical procedure of carrying out on a series of caught images to Fig. 1 .1-4D.The result that each figure comprises its upper left original image, upper right Sobel gradient operator, the threshold process result of lower-left Soebl gradient, erosion (erosion) result that bottom right is lower-left.

Fig. 1 .1-5 is illustrated in Fig. 1 .1-4A to the erosion kernel using in D.

Fig. 1 .1-6A illustrates the drop height from speed II ≈ 13 cycle drop video.

Fig. 1 .1-6B illustrates the DFT of Fig. 1 .1-6A.

Fig. 1 .1-7A illustrates the drop size (size) from speed II ≈ 13 cycle drop video.

Fig. 1 .1-7B illustrates the DFT of Fig. 1 .1-7A.

Fig. 1 .1-8A illustrates the average gray from the region of speed II ≈ 13 cycle drop video.

Fig. 1 .1-8B illustrates the DFT of Fig. 1 .1-8A.

Fig. 1 .2.1-1 illustrates the comparison between image gradient, power law and exponential transform result.

Fig. 1 .2.1-2 illustrates the result of power law and then Otsu threshold process and erosion.

Fig. 1 .2.2-3 is illustrated in the drop sized data obtaining after greyscale transformation and then Otsu threshold process, erosion and largest connected component.Figure is DFT.

Fig. 1 .2.3-1 is that power law conversion is divided into the segmentation interpolation of five sections.

Fig. 1 .2.3-2 illustrates the image segmentation conversion of comparing with original function.

Fig. 1 .2.3-3 is by the segmented conversion signal that then Otsu threshold process, erosion and largest component obtain.Top be drop size and height and corresponding DFT below.

Fig. 1 .2.4-1 illustrates the Thresholding Algorithm of histogram equalization effect.

Fig. 1 .2.4-2 illustrates the Histogram Matching result of figure image intensifying.

Fig. 1 .2.4-3 illustrates by the Histogram Matching signal that then Otsu threshold process, erosion and largest component obtain.Top be drop size and height and corresponding DFT below.

Fig. 1 .3-1 illustrates the frequency filtering of how carrying out with the equivalence of spatial domain wave filter.

Fig. 1 .3-2 illustrates how vertical Sobel mask is converted to convolution kernels.

Fig. 1 .3-3 illustrates frequency domain high pass filter effect.

Fig. 1 .3-4 illustrates another example of frequency domain filtering.

Fig. 1 .4-1 illustrates the result of wavelet filtering.

Fig. 1 .4-2 illustrate by the wavelet filtering signal that then Otsu threshold process, erosion and largest connected component obtain with and DFT.

Fig. 1 .5.1-1 has compared alternative manner and Otsu method.

Fig. 1 .5.1-2 illustrates by alternative manner threshold process, Sobel gradient before this before, and be then to corrode and largest connected component and the signal that obtains, in addition its DFT.

Fig. 1 .5.2-1 has compared constant size threshold value (constant level threshold) and Otsu and alternative manner.

Fig. 1 .5.2-2 illustrates by constant size threshold value and processes, Sobel gradient before this before, and be then to corrode and largest connected component and the signal that obtains, in addition its DFT.

Fig. 1 .5.3-1 has compared Otsu, average and intermediate value threshold process.

Fig. 1 .5.3-2 illustrates by average threshold process, Sobel gradient before this before, and be then to corrode and largest connected component and the signal that obtains, in addition its DFT.

Fig. 1 .5.3-3 illustrates by intermediate value threshold process, Sobel gradient before this before, and be then to corrode and largest connected component and the signal that obtains, in addition its DFT.

Fig. 2 .2.1.1-1 illustrates and can how to use cross the border (crossing) to carry out cycle count.

Fig. 2 .2.1.2-1 illustrates and can how by local maximum, to carry out cycle count.

Fig. 2 .2.2.1-1 illustrates the self correlation of drop altitude signal, speed I, II and III.

Fig. 2 .2.2.1-2 illustrates the cycle estimation from Fig. 2 .2.2.1-1.

Fig. 2 .2.2.1-3 illustrates the self correlation of drop size signal, speed I, II and III.

Fig. 2 .2.2.1-4 illustrates the cycle estimation from Fig. 2 .2.2.1-3.

Fig. 2 .2.2.2-1 illustrates the auto-covariance of drop altitude signal, speed I, II and III.

Fig. 2 .2.2.2-2 illustrates the auto-covariance of the amplification of drop altitude signal, speed I, II and III.

Fig. 2 .2.2.2-3 illustrates the auto-covariance of drop size signal, speed I, II and III.

Fig. 2 .2.2.2-4 illustrates the auto-covariance of the amplification of drop size signal, speed I, II and III.

Fig. 2 .2.2.3-1 illustrate drop height, speed I, II and III without inclined to one side AMDF.

Fig. 2 .2.2.3-2 illustrate drop height, speed I, II and III amplification without inclined to one side AMDF.

Fig. 2 .2.2.3-3 illustrate drop size, speed I, II and III without inclined to one side AMDF.

Fig. 2 .2.2.3-4 illustrate drop size, speed II and III amplification without inclined to one side AMDF.

What speed I cycle count Fig. 2 .2.2.3-5 is depicted as is incorrect.

Fig. 2 .2.2.3-6 illustrate speed I, II and III zone leveling gray scale without inclined to one side AMDF.

Fig. 2 .2.2.3-7 illustrate speed I, II and III zone leveling gray scale amplification without inclined to one side AMDF.

Fig. 2 .2.3.1-1 illustrates periodogram, DTFT and the DFT of speed II drop altitude signal.

Fig. 2 .2.3.1-2 illustrates periodogram, DTFT and the DFT of the amplification of drop size, speed II signal.

Fig. 2 .2.3.2-1 illustrates speed II drop altitude signal to L=90,60,30,15 Bartlett periodogram.

Fig. 2 .2.3.3-1 illustrates self correlation and correlogram and the DFT of drop height speed II signal.

Fig. 2 .2.3.4-1 illustrates the DTFT of the auto-covariance of drop altitude signal.

Fig. 2 .2.3.4-2 illustrates the DTFT of the auto-covariance of drop size signal.

Fig. 2 .2.3.5-1 illustrates the DCT-II expansion of speed I drop altitude signal.

Fig. 2 .2.3.5-2 illustrates the amplitude of the DCT-II coefficient of drop altitude signal.

Fig. 2 .2.3.5-3 illustrates the amplitude of the DCT-II coefficient of drop size signal.

Fig. 2 .2.3.5-4 illustrates the DST-II expansion of incorrect and correct way.

Fig. 2 .2.3.5-5 illustrates the amplitude of the DST-II coefficient of drop altitude signal.

Fig. 2 .2.3.5-6 illustrates the amplitude of the DST-II coefficient of drop size signal.

Fig. 2 .2.4-1 illustrates the small echo cycle count of speed I drop altitude signal.

Fig. 2 .2.4-2 illustrates the small echo cycle count of speed II drop altitude signal.

Fig. 2 .2.4-3 illustrates the small echo cycle count of speed III drop altitude signal.

Fig. 2 .3.1.1-1 illustrates the Yule-Walker method of speed I drop height.

Fig. 2 .3.1.1-2 illustrates the Yule-Walker method of speed II drop height.

Fig. 2 .3.1.1-3 illustrates the Yule-Walker method of speed III drop height.

Fig. 2 .3.1.1-4 illustrates the Yule-Walker method of speed I drop size.

Fig. 2 .3.1.1-5 illustrates the Yule-Walker method of speed II drop size.

Fig. 2 .3.1.1-6 illustrates the Yule-Walker method of speed III drop size.

Fig. 2 .3.2.1-1 illustrates the pseudo-spectrum of MUSIC method of speed I drop altitude signal.

Fig. 2 .3.2.1-2 illustrates the pseudo-spectrum of MUSIC method of speed II drop altitude signal.

Fig. 2 .3.2.1-3 illustrates the pseudo-spectrum of MUSIC method of speed III drop altitude signal.

Fig. 2 .3.2.1-4 illustrates the pseudo-spectrum of MUSIC method of speed I drop size signal.

Fig. 2 .3.2.1-5 illustrates the pseudo-spectrum of MUSIC method of speed II drop size signal.

Fig. 2 .3.2.1-6 illustrates the pseudo-spectrum of MUSIC method of speed III drop size signal.

Fig. 3-1 illustrates the general synoptic diagram of Machinery Control System.

Fig. 3 .1-1 illustrates the IV speed regulator for manual adjustments.

Fig. 3 .1-2 illustrates side view or the front view of the possible shape of IV pipe depressor/supporter.

Fig. 3 .1-3 illustrate IV pipe depressor/supporter possible shape axial/top/bottom view.

Fig. 3 .1-4 illustrates the way of contact between shape, edge, angle and the three of IV pipe, depressor and supporter.

Fig. 3 .1-5 is illustrated in the concrete example of the depressor with sharp edges in top, a left side, side and right view.

Fig. 3 .2-1 illustrates side view and the axial view of driving screw.

Fig. 3 .2.1-1 illustrates the combination of difference driving screw, and wherein precision can be enhanced 10 times.. size is pure schematic.

Fig. 3 .3-1 illustrates the schematic diagram moving from axle.

Fig. 3 .3-2 illustrates key/keyway and combines to control from axle and move.

Fig. 3 .3-3 illustrates tooth bar (spline)/groove and combines to control from axle and move.

Fig. 3 .3-4 illustrates bearing (a plurality of) and controls from axle and move.

Fig. 3 .4-1 is illustrated in and in translational motion, uses lever.

Fig. 3 .5.1-1 illustrates pivotable " utcracker ".

Fig. 3 .5.1-2 illustrates the principle moving from axle that can be absorbed by " utcracker " of pivotable.

Fig. 3 .5.1-3 illustrates the leverage of pivotable " utcracker ".

The linearly moving parts that Fig. 3 .5.1-4 is illustrated in any position, can contacts with rotary part in any geometric configuration.

Fig. 3 .5.2-1 illustrates " utcracker " of the pivotable of rotation.

Fig. 3 .6-1 illustrates the use of pincers.

Fig. 4 .1-1 illustrates the example of good illumination.

Fig. 4 .1-2 illustrates the example of poor lighting.

Fig. 4 .1-3 illustrates the reason of reflection/luminance contrast.

Fig. 4 .2-1 illustrates by increasing the distance between light source and present drip chamber, can reduce reflection/luminance contrast.

The brightness disproportionation that Fig. 4 .2-2 illustrates by a plurality of source reflection/luminance contrasts of common elimination weighs, and can reduce reflection/luminance contrast.

How Fig. 4 .3-1 can use a plurality of light sources if illustrating.

Fig. 4 .4-1 illustrates and can how by photoconduction/light pipe/light tunnel/integrated/optical fiber, to guide single source from a plurality of position illumination present drip chamber.

Fig. 4 .4-2 illustrates the multifarious principle of image that photoconduction/light pipe/light tunnel/integrated/optical fiber creates point light source of single.

Fig. 4 .5-1 illustrates the mirror of the minute surface combination that can be used to direct light.

Thereby Fig. 4 .6-1 illustrates and can how to amplify light source and eliminate the unbalanced of each point source.

Fig. 4 .7-1 illustrates and can how with reflecting surface, to reduce reflection/luminance contrast.

Fig. 4 .7-2 illustrates formation and the shape that reflecting surface can be in different ways.

Fig. 4 .8-1 illustrates and can how with rough surface, to make light random scatter.

Fig. 4 .10-1 illustrates the light guide/stopper that extends to object from light source.

As long as Fig. 4 .10-2 illustrates the reflection/luminance contrast that can reduce in the image-capturing apparatus visual field, the light guide/stopper and the picture catching that can any relative position be configured.

Fig. 4 .10-3 illustrates and can place light guide/stopper at diverse location.

Accompanying drawing-reference marker

Do not use

The specific embodiment

0. foreword

In the disclosure, provided the method and apparatus aspect complete automatic IV monitoring and control system all, this has expanded the scope of application before me and has made my application before more complete:

1. U. S. application 12825368: the IV by Digital Image Processing monitors

2. U. S. application 12804163: the IV processing by video and image monitors

3. the electro-mechanical system that U. S. application 13019698:IV controls

The schematic diagram of whole system, shown in figure 0.1-1, comprises illumination subsystems, picture catching, processing and frequence estimation subsystem and mechanical control subsystem.

In first segment image is processed, we will disclose the great amount of images treatment technology of our application.

In second section frequence estimation, we will disclose a large amount of frequence estimation technology, for from the periodic measurement count cycle.By providing at least one example at each apoplexy due to endogenous wind, covered the known technology of all categories.

In the 3rd joint machinery is controlled, we disclose for effectively controlling various mechanism and the principle of drop speed.

In the 4th joint illumination, we disclose the technology of the present drip chamber of throwing light on suitably, so that reflection and other defect are by the object that can not disturb us observing.

In figure 0.1-2, provide the succinct flow chart of monitoring and control procedure.

0.1 data set

This part is expanded on US12804163 in application, and the test in this is open can extract (may exist due to the ND variation of experimental enviroment some slightly different) from be used to apply for the same video Monitoring Data of IV drop process of US12804163.At length shown in following table, frame rate, dropping speed and signal type.

Table 0.1-1, data set is described

Therefore, have 3 * 3=9 different pieces of information group altogether, and they are all listed in table 0.1-1 and are plotted in figure 0.1-1 in figure 0.1-9.

0.2 symbol

Due to the multiformity of the algorithm of describing in the disclosure and the multiformity of data set, each that enumerate one by one its combination by be waste time and energy and will cause long description.Therefore we be partial to use criterion group symbol such as:

{ }: comprise

*: cartesian product

empty set

And other.

For example, method A, method B} * data I, data I I} is applied to each in two methods in each in two data sets meaning, this also can be represented as { (method A, data I), (method A, data I I), (method B, data I), (method B, data I I) }.

This is middle school mathematics knowledge, and any person skilled in the art will be considered to be familiar with in this.

1. image is processed

Technology disclosed herein is the expansion of disclosed technology in application US12804163.We will prove, for catching for monitoring the video/image of IV process, also have a lot of other technologies that can be used to process image.

1.1 look back the image processing techniques in application US12804163

First, we recall disclosed image processing techniques in application US12804163.Fig. 1 .1-1A illustrates the image of IV present drip chamber, and in Fig. 1 .1-1B, we specify wherein and will carry out the region of image processing with the rectangle of points of proximity drip.Select the object in the region of points of proximity drip to be mainly to process with low resolution, low frame rate image-capturing apparatus and low speed processor.If can extracting cycle signal, actual realization can be monitored any region.

Fig. 1 .1-2A illustrates vertical Sobel gradient operator, and Fig. 1 .1-2B illustrates vertical Prewitt gradient operator, and Fig. 1 .1-2C illustrates Laplace operator.

The image of drop process shown in Fig. 1 .1-3 with and Sobel, Prewitt and Laplce's gradient.Note, for Sobel and Prewitt gradient, we get its absolute value sum vertical and level result:

| gradient |=| G _x|+| G _y|,

This is standard practices in the art.

The object of applying these image enhancement techniques is to emphasize that key character is for subsequent treatment, and gradient operation is emphasized out from its background by drop in this particular example.

Fig. 1 .1-4A illustrates the image processing step of four images of order to D.In each subimage, Fig. 1 .1-4A for example, upper left is the original image having at the index shown in title, and we illustrate the example that applies Sobel gradient (vertical+level) in upper right.Lower-left applies Otsu threshold process method in Sobel gradient result, and is binary picture by greyscale image transitions.In bottom right, first we corroded threshold value result and calculated it and be communicated with component.

From be communicated with component, extract important information.At Fig. 1 .1-4A in each title of the bottom-right graph picture of D, first digit illustrates the quantity that is communicated with component, second digit is illustrated in the size of the largest connected component recording in a plurality of pixels, and third digit illustrates the average height of the y coordinate of largest connected component, and wherein y increases from image top to bottom.

If we check that 1.1-4A is to D, we can confirm that each treatment step has improved picture quality and result is always consistent with our visual examination at once.Whole 180 images of processing from low-resolution camera in the frame rate with 15/ second reached after 12 seconds, at the figure of largest connected component y coordinate shown in Fig. 1 .1-6A, and at its DFT shown in Fig. 1 .1-6B.At the image of the size of largest connected component shown in Fig. 1 .1-7A, and at its DFT shown in Fig. 1 .1-7B.

We see for the height of largest connected component and drop signal, and by having the index (index) of the non-DC component of maximum DFT amplitude, DFT has all identified correct amount of cycles,

If we are comparison diagram 1.1-6A and Fig. 1 .1-7A, although we see the signal of the height of comparing largest connected component, the signal of the size of largest connected component is very not obvious, and DFT result has still been identified the cycle of equal number exactly from both.

In Fig. 1 .1-8A, for identical image sequence, we illustrate by very coarse, the simplest and may be that the most possible mode of extracting significant signal is extracted signal: get simply the average of all grey scale pixel values.Do not apply an image processing techniques, but the signal in Fig. 1 .1-8 still illustrates regular periodic patterns, and its DFT in Fig. 1 .1-8B has also identified correct cycle count.

For the experiment of the image sequence of the drop process from other speed, still please refer to application US12804163.

By the success of cycle count, disclose following item:

1. drop altitude signal, is the signal of best and maximum characteristics in drop cycle demonstrably

2. drop size signal, the signal of appropriate characteristic (or some people can think that this is the poorest, as Fig. 1 .1-7A can be impliedly)

3. (or some people can be classified as the signal of appropriate characteristic to the average gray signal in region, as Fig. 1 .1-8 can be impliedly), although in good periodicity shown in our example, from lacking the fact of all processing steps, it can prove this signal and be minimum characteristic/difference signal under other shooting environmental.

By [best, appropriateness, the poorest], by the extreme and centre with whole frequency spectrum, provide example, we have covered the whole bag of tricks of extracting cycle signal from IV drop process, and use accurate result from example to prove and can detect with any cyclical signal the amount of cycles in IV drop cycle.

Although the technology of introducing in application US12804163 is strong and sufficient, we will be at other optional embodiment shown in the disclosure.

1.2 greyscale transformation

Greyscale transformation is the simplest in all image enhancement techniques.Please refer to [Prentice Hall, the Gonzalez of 2002, R.C., Woods, the Digital Image Processing second edition of R.E.] for discussing.

Definition:

By the value representation of conversion preceding pixel be x, transforming function transformation function to be expressed as value representation after T (), conversion be y, by being related to that y=T (x) carries out greyscale transformation from x to y transfer.

1.2.1 power law conversion

The primitive form of power law conversion is

y=T(x)=cx ^p

Wherein, c and p generally get on the occasion of, but do not get rid of the probability of negative value.

Sometimes be also written as

y=T(x)=c(x+ε) ^p

Wherein ε is the biasing that is added to x.

X is generally normalized to [0,1], and c (x+ ε) ideally ^pto be mapped as [0,1] to [0,1], but in practice, this does not need strictly to observe yet.

In order to obtain power law conversion, how to be applied to the concept of the image that catches in IV process, referring to Fig. 1 .2.1-1.The upper left corner illustrates untreated image (21 mean the 21st image in sequence).Second, third and the 4th row shown in Sobel, Prewitt and Laplce's gradient result.We apply power law conversion y=5x ⁴, and result is in the first row, the 5th row.Brighter part (due to luminous reflectance, being drop in image) is enhanced, and background is suppressed.This makes following treatment step more easy.

Note, before applying power law conversion, we need

1. pixel is normalized to [0,1].

2. apply conversion.

And in the end in step, be converted to original scope (in our example, we only block).

Fig. 1 .2.1-2 shows whole results of the power law conversion that adopts the threshold process discussed in application US12804163, erosion.The image of lower-left illustrates threshold process result, and the image of bottom right is the result of after erosion (thereby having removed less part).From bottom-right graph picture, by find largest connected component in image, extract upright position and the height of drop, and be stored as vector.We can be clear that, in treatment step, how the information of extracting mates well with our visual interpretation.

Fig. 1 .2.1-3 illustrates how from drop height, to record dropping speed.Top illustrates uses y=5x ⁴the variation of the drop height that conversion (for figure image intensifying) is extracted, therefrom we count 13 cycles; Bottom illustrates its DFT conversion, and maximum non-DC component is at X=13 place, thereby can determine 13 cycles.

Fig. 1 .2.1-4 illustrates how from drop size, to record dropping speed.Top illustrates uses y=5x ⁴the variation of the drop size that conversion (for figure image intensifying) is extracted, therefrom we count 13 cycles; Bottom illustrates its DFT conversion, and maximum non-DC component is at X=13 place, thereby can determine 13 cycles.

We have also verified, use y=5x ⁴in conjunction with Otsu threshold process, erosion and largest connected component, for all speed I, II and III, produce correct periodicity drop height and size signal.From gained drop height and size signal, DFT has identified correct cycle count.

1.2.2 exponential transform

General synonym: transformation transformation

The simple greyscale transformation of another type is exponential transform, and this is also referred to as transformation transformation, because logarithm and index are reciprocal.

The primitive form of exponential transform is

y=T(x)=c·[a ^x-ε]

Wherein a is for just and being generally greater than one.C is also generally positive number.

X is generally normalized to [0,1], and c[a ideally ^x-ε] also will be mapped as [0,1] to [0,1], but in practice, this does not need strictly to observe.

Referring to Fig. 1 .2.1-1, in the 6th image of the first row, we use exponential transform y=5[3 ^x-2], the result different from power law conversion and Sobel, Prewitt and Laplce's gradient result is shown.Compare the enhancing of other types, picture contrast is stronger, and the brightest part position and the location matches of drop.

In Fig. 1 .2.2-1, the top right plot in the group of each 4 width subimage looks like to illustrate y=5[3 ^x-2] result, and lower-left is that its threshold process result, bottom right are to use largest connected component to extract the erosion results of drop height and size thereon.Obviously, result is correct.

Fig. 1 .2.2-2 top is the drop altitude information on 180 samples, and bottom is its DFT result.The non-DC component of its maximum is at X=13 place, represents 13 cycles; Fig. 1 .2.2-3 top is the drop sized data on 180 samples, and bottom is its DFT result.DFT provides the cycle count as drop altitude signal.

After tested, for all speed I, II and III, when with Otsu threshold process, erosion and largest connected component combination, gray scale y=5 (3 ^x-2) provide correct periodicity drop altitude signal.From gained drop altitude signal, DFT provides correct cycle count.

1.2.3 piecewise linear transform

General synonym:

Piecewise linear function transformation of variables, as [Gonzalez, R.C., Woods, R.E. Digital Image Processing (Digital Image Processing) (second edition) 3.2.4 joint] described in, be supplementing of other greyscale transformation technology, it is advantageous that it can be similar to the function of any complexity.By using this technology, all technology before can simulating.In hardware is realized, it is equivalent to conventionally alleged " question blank ".

We illustrate, y=5x in power law conversion ⁴work.We carry out this function of interpolation with five linearity ranges:

X

Y

0.0	0.0
		0.2	0.008
0.4	0.128
		0.6	0.648
0.8	2.048
		1	5.0

Table 1.2.3-1, at [0,1] neutral line interpolation y=5x ⁴end points and intermediate point.

First pixel value is normalized to [0,1], and then in the section at its place, is interpolated.Functional arrangement is illustrated in Fig. 1 .2.3-1, and the result of the sharpening image of five images is illustrated in Fig. 1 .2.3-2.Before converting back original pixels scope, block simply the value outside [0,1].In each image, we have also placed original y=5x ⁴how closely result after result and interpolation and we can see both couplings.

Compared to power law, exponential sum, to number conversion, the advantage that stagewise conversion has is its available query table realization, compares real-time calculating faster.Therefore, recommended the actual realization as these class methods on product.

We have also verified, use this five sections of stagewise linear transformation y=5x ⁴in conjunction with Otsu threshold process, erosion and largest connected component, for all speed I, II and III, produce correct periodicity drop height and size signal.From gained drop height and size signal, DFT provides correct cycle count.

1.2.4 based on histogrammic greyscale transformation

We have also shown and can use histogram information to carry out gradation conversion.This class technology be commonly use and [Prentice Hall,, Gonzalez in 2002, R.C., Woods, the Digital Image Processing(Digital Image Processing of R.E.), second edition, 3.3 joint] in can find detailed description.

Owing to finding the detailed concrete realization of histogram technology in lot of documents (book that comprises above-mentioned Gonzalez) and software kit, we only convert to illustrate this principle with continuous function.

First, the discrete brightness (intensity) of gray level image value can extended interval [0, L-1] and the probability of each discrete luminance values be n _i/ N, wherein N is total number of pixels and n _iit is the pixel quantity with brightness i.

Span [0, L-1] can be normalized to [0,1] and we are expressed as r by the stochastic variable corresponding to normalization interval.Easy in order to discuss, our hypothetical probabilities density (density) function is continuous and monotone increasing strictly, so its inverse always exists.

Probability density function is represented as

P _r(r),0≤r≤1

And will there is cumulative distribution function, for

$C_r\left(r\right)={\∫}_0^r{P}_r\left(r\right)\mathrm{dr},0\≤C_r\left(r\right)\≤1$

May exist and there is cumulative distribution function Z _r(r) another rectangular histogram, we wish also have its shape through the image of conversion, and the mode that realizes this is via following conversion

$r\→T\left(r\right)Z_r^{-1}\left(C_r\left(r\right)\right)$

And it is provable, the image through changing will have just and Z _r(r) the new rectangular histogram of coupling.

The rectangular histogram of coupling/appointment (specification) image is called as Histogram Matching with the histogrammic process of distributing arbitrarily or rectangular histogram is specified.When target rectangular histogram is while being uniformly distributed simply, be called as histogram equalization.

When heel is used the Thresholding Algorithm of histogram transformation, equilibrium is inappropriate, and in an example shown in Fig. 1 .2.4-1.Use histogrammic Thresholding Algorithm generally to make great efforts to find " trough " in rectangular histogram, and histogram equalization make histogram equalization and filled existing trough.We see in Fig. 1 .2.4-1, very bad by the threshold value result of Otsu method.

Yet, also may be combined with from some other algorithms or now video/image there is the characteristic different with our example.In any case histogram equalization can be assumed to be the particular case of Histogram Matching, and we do not get rid of the probability of using it.

Fig. 1 .2.4-2 illustrates the Histogram Matching of several images, and wherein target is Sobel gradient (horizontal+vertical) result of the image 21 in speed II ≈ 13 sequences.Obtained appropriate result and we also can see how the rectangular histogram of result closely mates with target.

We test { drop height, drop size } * speed I, II, III} signal, described signal is extracted by Histogram Matching, is then Otdu threshold process, erosion and largest connected component.For all these signals, result is all accurately, because DFT provides correct cycle count.

We have illustrated application Histogram Matching/appointment (comprising equilibrium) thus in the disclosure.Can also have other distortion based on histogrammic technology or use the technology of other information of rectangular histogram and figure image intensifying, but the method that neither one can be disclosed herein with us there is remarkable difference.

The general introduction of greyscale transformation

By the disclosure is combined with application US12804163, we have shown that image enhancement technique can thus convincingly

1. use separately the value of each pixel itself.

2. use Histogram Matching/balancing technique.

3. use the spatial domain technique, particularly various gradients of other types, to calculate new pixel value according to itself and adjacent value thereof.

There are some distortion, but there is no an essential distinction with we are disclosed:

1. for power law conversion y=T (x)=c (x+ ε) ^p, p<1 is also possible, this provides the value mapping different from p>1.If the image catching by different images capture device and shooting environmental has characteristic comparatively suitable for p<1, obviously can use p<1.

2. based on above-mentioned (1) identical reason, also can use and refer to inverse of a number, i.e. logarithmic transformation, it is often defined as y=clog (x+1).

Therefore we reach a conclusion, and can be used in the treatment step of the video/image processing based on IV monitoring system as the greyscale transformation of general category technology.In this classification, other technologies all can be with our disclosed method without any essential distinction.

1.3 frequency domains are processed

The frequent frequency domain technique of using in image is processed.Frequency domain technique works like this: first image is converted to its frequency domain representation, application treatment technology and result inversion is changed in spatial domain.

Please refer to [Prentice Hall,, Gonzalez in 2002, R.C., Woods, the Digital Image Processing second edition of R.E., chapter 4] for discussing.

Definition:

Digital picture is defined as to P _{(a, b)}, 0≤a≤M-1,0≤b≤N-1, and make periodically to expand to the coordinate dropping on outside [0, M-1] * [0, N-1], thereby

$P_{(a+k_{1}M,b+k_{2}N)}=P_{(a,b)}$

K wherein ₁and k ₂it is all integer.

Periodically propagation energy makes us by periodicity Defined, be

${(P*G)}_{(a,b)}=\underset{i=0}{\overset{M-1}{\mathrm{\&Sigma;}}}\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}(P_{(i,j)}\&CenterDot;G_{(a-i,b-j)})=\underset{(i,j)=\left(\mathrm{0,0}\right)}{\overset{(M-1,N-1)}{\mathrm{\&Sigma;}}}{P}_{(i,j)}\&CenterDot;G_{(a,b)-(i,j)}$

This is important, because only have periodic to expand us, just can prove theoretically following most important theories:

Theoretical I:

The 2D-DFT of the convolution of image P and G is long-pending identical with their 2D-DFT's separately.This proof is as follows:

∴

P*G=F ^-1(F(P*G))=F ^-1(F(G)·F(P))

In practice, the size of normally used spatial domain wave filter is generally less.For example, standard Sobel, Prewitt and the Laplce's gradient magnitude in application US12804163, used are only 3 * 3.Therefore, whether we use the periodicity expansion of digital picture and Defined afterwards only to affect and to be positioned at the very pixel at edge, and we do not have much interest to this.

Above-mentioned theory means, can realize via frequency domain multiformity the impact of each spatial domain wave filter of periodic expanded definition.

Shown in Fig. 1 .3-1, in frequency, carrying out the process of Sobel vertical filtering:

1. get the DFT of original image.In Fig. 1 .3-1 lower-left, illustrate Shi Xiang center (center) moving coordinate [0,0] through shifted version, this is the convention of this area.

2. spatial domain wave filter is generally configured to (2k+1) * (2k+1) matrix, and corresponding to the weight of each image pixel itself at center (k+1, k+1).In spatial domain filtering, spatial domain wave filter by directly " mask " on original image, be multiplied by pixel below and get it and; Yet, due to convolution only, there is direct frequency domain pairing (counterpart) but there is no mask, first we need to be transformed to identical convolution:

I. exchange the pixel of luv space territory wave filter, around center (k+1, k+1) exchange first up and down symmetrically, then left and right exchange or exchange equivalently.This is shown in Fig. 1 .3-2.Note, in Fig. 1 .3-2, left and right exchange does not have effect, because vertical Sobel gradient has been horizontal symmetry.

II. creating and the new images of original image formed objects, is zero by all pixel value assignment, be then shifted (I) thus result moves to initial point by center (k+1, k+1).For dropping on the pixel outside border after displacement, record their position and value and use the periodicity expansion in following III.

III. with periodically expanding pixel is arranged in effective image-region as defined above.

IV. get the DFT of (III).

3. the result of (2.IV) is multiplied by the result of I.

By get its inverse Fourier transform by results conversion (3) to spatial domain.

Note, if third and fourth image of the first row of we comparison diagram 1.3-1, we will be clear that the periodically impact of expansion on the coboundary of original image and lower boundary.Because this impact occurs over just boundary, in most of the cases, by the value that the arbitrary value such as 0 or be set to is derived by the value of its neighbor by this position of border pixel values, can cast out simply this impact.

Also note that the DFT conversion due in a big way, first the second row of Fig. 1 .3-1 is illustrated as to the absolute value of DFT and adds constant (10) scale of then taking the logarithm.This is also standard practices in the art.Because each logarithmic image has different range, the gray scale in different images should not compare mutually.

We have set up each the spatial domain wave filter that can realize via frequency domain filter now.In practice, we need to be when each use the DFT image of calculating filter, but can only do so once and event memory for using in the future.

We also can use the wave filter being directly dispensed in frequency domain, and this is also the basic general knowledge in this area.We illustrate the example that strengthens image with high pass filter.

The high pass filter that we select is

$H(I,J)=\left\{\begin{array}{cc}1,& \left|I\right|>1\mathrm{or}\left|J\right|1\\ 0,& \left|I\right|\&le;1\mathrm{and}\left|J\right|\&le;1\end{array}\right.$

To remove 3 * 3=9 frequency coefficient altogether.The example of its impact shown in Fig. 1 .3-3.Note, in image name, we use shorthand (x, y) to be illustrated in the y number of sub images of x in capable.

The 1st the 2nd of row image illustrates the logarithmic scale 2D DFT of original image.Shown in the 3rd image through the frequency spectrum of filtering and displacement, it clearly has dark/Kong center due to wave filter impact.At the image through rebuilding shown in the 4th image, and in demonstration dark shown in the 5th image of the 1st row and that amplify, illustrate and how to have strengthened contrast.

Various threshold process methods such as Otsu and alternative manner still can be carried out threshold process to image well, adopt the constant value threshold process of suitable constant value (be 35, see Fig. 1 .3-3) also like this herein.

In the result on another image (in sequence the 31st) shown in Fig. 1 .3-4, and we are clear that in this image, following methods { Otsu, the constant threshold of alternative manner and employing threshold value 35 is processed } also identifies just size and the position of drop.

We test and find, with the frequency domain that is equivalent to Sobel gradient, carrying out filtering is useful for the dropping speed of the video Monitoring Data in 5.3 cycles of speed I ≈, 13 cycles of speed II ≈, 23 cycles of speed III ≈, and can make together with additive method for extracting drop height and drop size signal.

We have also verified, use 3 * 3 high pass frequency domain filters for all speed I, II and III, to produce correct periodicity drop height and size signal in conjunction with Otsu threshold process, erosion and largest connected component.From gained drop height and size signal, DFT has identified correct cycle count.

Please refer to Fig. 1 .3-5, the speed II drop height and the size signal that for the frequency domain high-pass filtering by heel Otsu threshold process, erosion and largest connected component, obtain.

We illustrate thus

1. can in frequency domain, obtain spatial domain filtering.

2. also can strengthen image with the wave filter purely designing in frequency domain.

Therefore complete for the displaying of the application of frequency domain technique.

Depend on video and other factors, certainly can apply the different wave filter using from us, but these do not form any essential distinction.

1.4 wavelet method

We have also shown can realize figure image intensifying with wavelet transformation.The basic concept of wavelet transformation is multiresolution.The in the situation that of 1-D, in each rank, it calculates one group of approximation coefficient and one group of difference coefficient; In 2-D situation, still use the theory of identical approximate and difference.

We use the simplest small echo: Haar small echo, illustrates.For the image of big or small 2M * 2N, other conversion of each level and down-sampled after, new large young pathbreaker is original 1/4, i.e. M * N.The information of 2 * 2 frames of each in original image will be comprised in four coefficients now, illustrate as follows:

$\left[\begin{array}{cc}a& b\\ c& d\end{array}\right]\&DoubleRightArrow;\begin{array}{cc}\frac{a+b+c+d}{2}& c_{\mathrm{average}}\\ \frac{a+b-c-d}{2}& c_{\mathrm{horizontal}}\\ \frac{a+c-b-d}{2}& c_{\mathrm{vertical}}\\ \frac{a+d-b-c}{2}& c_{\mathrm{diagonal}}\end{array}$

Observe c _horizontal(C _level) be the vertical gradient of obtaining together with two lower pixel places on two in essence, and c _vertical(C _vertically) be the horizontal gradient obtaining together with two lower pixel places on two in essence, then this similarity is implying that they can replaced general spatial territory gradient and use.

In the example of Fig. 1 .4-1, in each figure, the 2nd image of upper row is c _horizontaland c _verticalabsolute value sum,

|c _horizontal|+|c _vertical|

If we are by itself and the comparison of Sobel gradient result, a little less than they it seems, this is due to 1/2 coefficient, and this coefficient is much smaller than Sobel gradient factor.In the 1st row of each group, the 3rd illustrates and the identical result of the proportional demonstration of its scope (minimum (min) → 0, maximum (max) → 255).Yet ensuing processing, such as threshold process, still based on original small echo gradient result.

Although there are some obvious noises, we see that the brightness in drop location obviously strengthens.The the 1st and the 2nd image in the 2nd row in each figure illustrates Otsu threshold process and erosion results, and the 3rd image be the threshold value being provided by alternative manner (seeing 1.5.1 joint), and uses largest connected component can therefrom extract drop size/highly.

Although the variation of the drop height shown in the 2nd image of the 2nd row of each subimage and size and not as with before method in chapters and sections and application US12804163 extract obvious like that, for shown in speed II ≈ 13 cycle video, still can from Fig. 1 .4-2, see the periodicity in result.DFT determines the 13 correct cycle counts for drop height and size signal.

?	Drop height	Drop size
			Speed I	√	?
Speed II	√	√
			Speed III	?	√

Table 1.4-1, mark √ means that the small echo gradient result of heel Otsu threshold process, erosion and largest connected component is accurately, wherein DFT has provided correct cycle count.

Therefore, we shown we application image processing step in use wavelet transformation.The dissimilar small echo with different length and value that also has One's name is legion, but neither one has essence different from the example of ours herein.

1.5 threshold process

Threshold process is one of the most basic image processing techniques.In application US12804163, we illustrate and can carry out automatic detection threshold level by Otsu method.In the disclosure, it is also feasible that we will illustrate additive method.

1.5.1 alternative manner

General synonym:

Alternative manner is found out threshold process size L in iterative process.Its realization is simple, need to for the concrete knowledge of image and relative noise, be not sane.

Step:

1. select the initial value of L, for example, the meansigma methods of image.Also can select or produce even at random other to be reasonably worth.

2. pixel is divided into two groups:

S ₁={P _i,j:P _i,j>L}

S ₂={P _i,j:P _i,j≤L}

3. calculate the average of two groups

m ₁=mean(S ₁)

m ₂=mean(S ₂)

4. produce new threshold process size conjecture (guess)

$L_{\mathrm{new}}=\frac{m_1+m_2}{2}$

5. if L _new(L _newly) equal L, L=L is set _new, exit;

Otherwise, L=L is still set _new, then skip to step 2.

Note that if L is not used to integer precision, need to regulate etc. and unisonly come testing differentia whether to be less than particular value, such as 1.

Experimental result on three images in the sequence of the speed II ≈ 13 cycle video in Fig. 1 .5.1-1 illustrates, and the method provides the result about the same with Otsu method on all images.First with Sobel vertical+level thresholds processes image.

Frame number	Otsu method	Alternative manner
			30	98	99
33	89	89
			37	88	85
40	30	32
			44	86	91

Table 1.5.1-1, relatively alternative manner and Otsu method

We empirical tests after Sobel gradient, by alternative manner, then corrode and largest connected component, the result producing, provides drop height and size signal, therefrom by DFT, for speed I, II and III video, correctly obtains cycle count.Result for 13 cycles of speed II ≈ is illustrated in Fig. 1 .5.1-2.

Otsu method has represented the Thresholding Algorithm classification of using histogram information; Use alternative manner, we have been illustrated in the situation of using ambiguously histogram information also can complete automatic threshold.

The threshold size of 1.5.2 arbitrarily/constant threshold size, artificial assignment

General synonym:

We also can use fixed threshold to process size rather than use any automatic algorithms.In the experiment shown in Fig. 1 .5.2-1, use the size 91 of constant, artificial assignment.Except in image 40, in most of images of sequence, this is close to Otsu and alternative manner result.

Following form illustrates except minority image, and constant threshold is mated Otsu and alternative manner well.Difference in minority image does not change overall cycle count.We empirical tests after Sobel gradient, by constant 91 threshold values, produced and and then corrodes the result with largest connected component, provide drop height and size signal, therefrom by DFT, for speed I, II and III video, correctly obtain cycle count.Result for 13 cycles of speed II ≈ is illustrated in Fig. 1 .5.2-2.

Image number	Otsu	Alternative manner	Constant
				30	98	99	91
33	89	89	91
				37	88	85	91
40	30	32	91
				44	86	91	91

Table 1.5.2-1 is constant size threshold value and Otsu and alternative manner relatively

In reality realizes, we control illumination, reflection, camera exposure time and other technologies.Due to construction standby on these parameters all make to fix, can rational expectation, we can find the constant threshold working well for application to process size always.

1.5.3 average/intermediate value or other simple threshold process methods

General synonym:

In Fig. 1 .5.3-1, in each quadrant, using the image pixel average as threshold process size and intermediate value and Otsu threshold process comparison (also corroding subsequently and largest connected component).Although visible, they do not have the very good information of transmission aspect the size of drop and position, and the final signal extracting after erosion and largest connected component, it is found that:

1. for average threshold value, the drop height of speed I, II and III all provides correct cycle count via DFT.The result of speed II is illustrated in Fig. 1 .5.3-2.

2. for intermediate value threshold value, the drop size in 13 cycles of speed I ≈ only, rather than other, via DFT, provide correct cycle count.The result of speed I is illustrated in Fig. 1 .5.3-3.

The shortcoming of average and intermediate value threshold process is that they are to noise-sensitive.Yet, as we are in application shown in US12804163, even without the meansigma methods of any pretreated original image, show and can correctly carry out the periodicity of cycle count, as long as the signal producing by average or intermediate value threshold process step keeps the periodicity identical with primary signal, it can be used as possible selection.

In addition, whether method works and depends on the characteristic of data set.Use good capture apparatus, environment, parameter setting and other to improve, can significantly reduce the noise in image, and can rational expectation, average or intermediate value threshold process will work for data set.

General introduction about threshold process

The most basic operation in processing as image, exists a lot of distinct methods to do threshold process.

If we are divided into two wide in range classifications by them:

(1), in non-automatic threshold process method classification, constant has been shown for we or manually definite threshold value is feasible.

(2) in the classification of automatic threshold processing method,

A. for based on histogrammic method, we have shown that Otsu method is feasible.Can rational expectation, additive method is also feasible.

B. for non-based on histogrammic method, it is feasible that we have illustrated alternative manner, and

I. depend on signal quality, even simple as feasible in the method for the average of image and intermediate value.

Although we do not have all methods of limit, every other method can be classified as { (1), (2) .a, (2) .b, (2) .b.i} classification, and we have shown example (a plurality of) in each classification.Every other method and our method do not have essential distinction.

2. frequence estimation

In application US12804163 before me, I have been illustrated in for obtaining the degree of freedom of the measurement selection aspect of cyclical signal.The disclosure will be illustrated in the degree of freedom of the algorithm selection aspect of frequence estimation.

Depend on technical field, term " frequence estimation " has a lot of synonyms." frequency spectrum/frequency spectrum " can be used for substituting " frequency ", and " analyze/detect " also can substitute " estimation " in a lot of occasions.The term of picture " cycle/periodically " and so on and " counting " is also normally used.Believe, the selection of term, if obviously use in the disclosure from me different, can't cause the application's claim inapplicable, because what accurately define protection domain is basic method, rather than for the specific selection of method name.

In application US12804163, I use counting of discrete Fourier transform (DFT) to form/drop on the amount of cycles in IV drop process.Dropping speed and three dissimilar periodic signals measurements (average gray of specific region in drop height, drop size and image) for wider range have provided experimental result, and DFT provides the cycle count accurately for all these.By the experiment as periodic signal by drop size, we show, even when carrying out direct eyes observation have any problem aspect definite cycle accurate counting on signal, DFT will still provide and the same count number accurately of itself observing from drop process.Accuracy by the experiment front shown in the disclosure, with and theoretical steadiness and realize simplicity, make it become the ideal chose of the frequence estimation method of the IV monitoring based on video/image.

Yet, in this problem, also exist reason to provide the optional method of DFT.One is because DFT itself still needs to improve.Because DFT only exists centrifugal pump is calculated at place, if the actual cycle of drop is the mark during specified time interval T, fractional part will be one of two integers close to this fractional value by loss and result.This is " resolution " problem in frequence estimation term in essence.This can by under show that the whole bag of tricks solves, and by for using the more Fast Convergent of IV dropping speed control mechanism to have contribution.

Another reason will be to guarantee complete and FR protection of the present invention.Frequency spectrum estimation has had as the subject of having set up centenary the history ([Stoica of surpassing, Petre & Moses, the Spectral Analysis of Signals(signal Spectrum Analysis of Randolph L.) preamble, XV page] and [Marple, the Digital Spectral Analysis with Applications(of L. has the digital spectrum analysis of application)]).The theory of discrete and continuous Fourier transformation can be regarded as to a large amount of offsprings' the older generation.After these, the method source of invention arises from the application-specific that can not be solved by method satisfaction before.For example, multiple emitter location and signal parameter estimation (MUSIC) method are used to determine in measurement that the signal from receiving at array element makes the parameter [Schmidt1986] of a plurality of wave surfaces of arrival aerial array originally.Each method has its specific theory hypothesis and one in the characteristic of signal and noise may be infeasible in another environment; This is only because signal quality that the video/image Processing Algorithm by us provides is very or fully good; so that a lot of methods are wherein found can be applicable to this signal, rather than they really provide on this problem improvement or have lightened new radiance.In any case because legal definition tends to use literal restriction, inventor will comprise the optional embodiment that is used to realize identical object wider and limit as possible as far as possible.A large amount of frequence estimation method lists are provided for this purpose.

2.1 well-formedness standards

For concordance, we use identical data group, thereby the difference between algorithm can be easy to comparison.Yet the data of our three types reduce aspect quality:

Quality(quality) (drop height) > Quality(drop size) > Quality(zone leveling gray scale)

If Quality () distributes numerical value, the signal quality that larger representative is higher.For drop altitude signal, this is best in three types, and all following aspects therefrom provide correct estimation; For zone leveling gray scale, certain methods can be applied; The situation of drop size between two extreme between.

Yet, as we have repeated to emphasize in US12804163 low in application, can use any periodic measurement.Self-evidently, it is good cyclical signal that drop height is compared zone leveling gray scale, but zone leveling gray scale can not be mainly the mode of signal of extracting due to us by some the test in following algorithm.As we described in US12804163, complete video/image processing in application herein in the very little video window near drop accent (drop forms and start).In fact, as Fig. 2 E from application US12804163 can find to Fig. 3 D, window size is less than 20 (width) * 50 (highly), is less than 1000 pixels.Select the object of so little window to be to guarantee accurately and scheme cheaply.On the other hand, mathematical algorithm itself does not propose any restriction for window size, and in fact true realization can be used the frame rate (>15) of more high-resolution photographing unit and Geng Gao, this will be that zone leveling gray scale is supported as first of useful signal.The average gray that signal is chosen as to specific region is actually the system of using infrared ray to detect for drop looking back, the drop wherein at every turn falling produces pulse while covering the infrared ray path between transmitter and receptor, this has provided second of this signal effectiveness and has supported; The 3rd of signal effectiveness supports that wherein, for (II:12 is to 13 drops) data, DFT provides correct estimation from zone leveling grey scale signal from Fig. 4 I and Fig. 4 J of application US12804163.

Based on this 3 point, we can put letter and declare that zone leveling gray scale is also the useful signal of frequency detecting; But in frequency detecting algorithm, use this signal, may need correspondingly to adjust image resolution ratio, frame rate and other parameters.Therefore,, if together with reading zone leveling gray scale in following example the test by special algorithm or I not being listed in them with special algorithm, this is only such for the specific data sets of listed I in the disclosure.If I adjust camera environment, parameter and video/image Processing Algorithm the good average gray data that are applicable to algorithms of different are provided, the data consistency in the disclosure between the concordance of data and the disclosure and application US12804163 will be lost.For each special algorithm, environment, acquisition parameters and the video/image Processing Algorithm through adjusting still can provide for its enough good average gray data.

Same reasons also extends to the drop size signal of listed I in the disclosure, this and not as drop altitude signal has like that, so high degree of guaranteeing can be by the test of all frequence estimation algorithms.Yet, suppose correspondingly to have adjusted environment, acquisition parameters and video/image Processing Algorithm, it can be used together with each special algorithm.

Reader should also be noted that different frequence estimation method uses capable of being combined is with complimentary to one another.One is inspired example will be to use DFT and AMDF(average amplitude difference function), this discusses institute in [hybrid algorithm (I) of 2.2.2.4 joint].As the method that has firm mathematical theory and confirmed by test in application US12804163, DFT provides cycle count accurately at gray level resolution place.Other algorithms of some of AMDF, have the ability of count cycle length (will determine fractional cycle counting) exactly.Use not enough data of some types of good (following defined by Suttability (S, E)) of its quality, possible AMDF identifies incorrect Cycle Length, because its finds the local minimum of locating in out of position.As the reparation to this, first DFT can be used to find the quantity of number of cycles P, and estimates from DFT result acquisition Cycle Length by division N/P, and this will be close to actual cycle length.We will use N/P as about estimation, and adopt AMDF to be positioned near the more meticulous actual cycle length of N/P.This will show in the chapters and sections of AMDF, and if other algorithms can combine with DFT or not comprise that other group algorithms of DFT can be combined to realize better result (comparing the single algorithm of independent use), will repeat to mention.

In the disclosure, we are no less than ten algorithms for frequence estimation by discussion.The combination of two algorithms will have N _algorithm* N _algorithm>=10 ²=100 types, and also may be by more algorithm combination together.Therefore there is no need also can not list all combinations.

Helpful is herein to provide well-formednessformal definition, well-formedness is the relation between special algorithm and specific data sets:

Well-formedness: for the specific data sets S of length N, by the viewed frequence estimation symbol E () of the clog-free people of vision and amount of cycles P,

Suitability (well-formedness) $(S,E)\stackrel{\mathrm{def}}{=}\left\{\begin{array}{cc}1,& |P-E\left(S\right)|<\mathrm{\&epsiv;}\left(N\right)\\ 0,& |P-E\left(S\right)|\&GreaterEqual;\mathrm{\&epsiv;}\left(N\right)\end{array}\right.$

ε (N) can be constant in decimal in N of haveing nothing to do, such as 0.5 or 1; Or it can be the function of the monotone increasing of N.Intuitively, if you only have the signal of length 10, the error of one-period is exactly intolerable certainly; If but you have the signal of length 1000, same error drops in tolerance certainly.

With this new function, express our above-mentioned discussion exactly:

And actually determined is quality and the selected ε (N) of S.

2.2 nonparametric technique

Definition:

So-called nonparametric technique, we are intended to express: hypothesis does not produce the knowledge of the actual physics model of signal (and noise), and do not make a try the parameter of appraising model.Directly from the estimation frequency of signal own.

2.2.1 simple method

These methods are directly moved substantially in the situation of not making any conversion (broadly, being not limited to time-domain and frequency-domain conversion) in time domain, and it checks periodic visual cues in signal curve, and imitates eye-observation by automatic step.

2.2.1.1

General synonym: zero passage, zero value detection, threshold process

If signal is assumed to be quasi periodic, it generally can swing within the cycle between spikes/low-points.Therefore the action of its cross the border (crossing) specific middle sizes values can be used to sense cycle.

In specific implementation, can there are some distortion:

(1) detect rising edge

A) size of must > crossing the border

B) necessary >=size of crossing the border

(2) detect trailing edge

A) size of must < crossing the border

B) necessary≤size of crossing the border

We are in Fig. 2 .2.1.1-1, and for the drop altitude information in 13 cycles of speed II ≈, this method provides counting accurately.Threshold value is chosen as 13 pixels tall, and this should be unable to obscure with cycle count, although two values are the same.For 5.3 cycles of speed II ≈, at n _{radix (base) 1}the less spike at=146 places causes inaccurate cycle count; For 23 cycles of speed III ≈, at close n _{radix 1}" spike separating " at=80 places also increased inaccurate counting.In addition, for 23 cycles of speed III ≈, at n _{radix 1}=50,51 places, whether the cycle that should be counted accurately depend in (1) a), b) between selection, only selecting b) time counting be accurately.

13 threshold value is not automatically to calculate, but manually choose.Can envision natch some special algorithms and come artificially to look for threshold value, but the counter-example that such algorithm can be easy to be built intentionally institute is invalid.Use other selections of threshold value, still can occur incorrect counting.

Useful straightforward procedure is alleviated this problem.For example, all indexes (index) that cross the border a little can be stored in array, and use by another of array and scan to find for it obviously index of " too approaching " of the former or the latter.This can be formally described below:

Mean be " by " average headway between the chain index of point, and if two chain indexs to compare ε mean more approaching, this means that one of them can be corresponding to little spike or separated spike.ε is 0.5, has been tested as the desired value of the drop altitude signal of speed I, II and III.

2.2.1.2 maximum/minima

General synonym: (signal) derivative, (part) maximum/minimum

Another method is to find local maximum/minima.This is for similarly being feasible well for the ideal signal of pure sinusoid and so on.The data drop altitude information of collecting for us, it is found that, must make restriction so that not every maximum/minimum can be selected, and must abandon the adjacent maximum/minimum of " too approaching ".

Following false code has been tested as correctly identifies maximum to the drop altitude information in 13 cycles of speed II ≈.

We need to compare S[i] and Threshold(threshold value) reason be to remove the maximum that is positioned at low clearance place, and use the reason of ε mean to be to remove the approaching index corresponding to little spike or separated spike.

The peaked result in 13 cycles of recognition speed II ≈, at drop altitude information shown in Fig. 2 .2.1.2-1.Small circle labelling the maximum of identifying.Cross in figure is illustrated in n _{radix 1}the maximum at=88 places has abandoned because it too approaches the local peaking at n radix 1=86 place, and therefore due to ε mean standard, it is rejected.

Except the signal of good quality especially, find ε and Threshold for all speed and all types of cyclical signal (drop height, drop size, zone leveling gray scale etc.), difficulty can always exist.

2.2.2 Time-domain Statistics method

We directly use statistical method for three methods of cycle estimation by illustrating in time domain.These methods are detecting destroyed or are even being submerged in aspect the periodicity of the signal in noise and have very strong ability.

2.2.2.1 self correlation

General synonym:

Sequence of real numbers S is defined as without partial autocorrelation

$R_{\mathrm{xx}}(S,m)=\frac{1}{N-\left|m\right|}\left\{\begin{array}{cc}\underset{n=1}{\overset{N-M}{\mathrm{\&Sigma;}}}S[n+m]S\left[n\right]& m\&GreaterEqual;0\\ R_{\mathrm{xx}}(-m)& m<0\end{array}\right.$

It also has and has inclined to one side version

${\stackrel{\&OverBar;}{R}}_{\mathrm{xx}}(S,m)=\frac{1}{N}\left\{\begin{array}{cc}\underset{n=1}{\overset{N-M}{\mathrm{\&Sigma;}}}S[n+m]S\left[n\right]& m\&GreaterEqual;0\\ R_{\mathrm{xx}}(-m)& m<0\end{array}\right.$

Wherein we are placed on R top to show difference by upper line.We only discuss without inclined to one side version hereinafter, and also can use inclined to one side version in practice.

The sequence R deriving from S _xx(S, m) shows the periodicity of identical with S (because S is not pure periodic sequence, " identical " should not be interpreted as ideal situation).

When | m| becomes close to N, and N-|m| will diminish, and on average will be N-|m|, and corresponding N-|m| value will not be accurate estimation.This can find out in Fig. 2 .2.2.1-1 and Fig. 2 .2.2.1-3.Therefore, in practice, we only check less conventionally | the situation of m|.

In addition, for real number S, R _xx(S, m) is with respect to 0 symmetry.

R _xx(S,m)=R _xx(S,-m)

Therefore for the situation of calculating positive m, be enough.

In our application, we can select m to guarantee can cover at least three cycles in m frame of video, and certainly, this depends on frame rate and the dropping speed of image capture device.

By counting R _xxspacing between (S, 0) and next local maximum is carried out determining of execution cycle.Can give mathematical proof, but be also significantly instinctively, to realize local maximum, wherein T _ethe estimation of the T of indication cycle and k are integers.

For the drop altitude information in 5.3 cycles of speed I ≈, we can find at R _xx(S, 0) next peak value afterwards will be when m=34.Fig. 2 .2.2.1-1 draws symmetrically around N, so we have the R that is positioned at n=180 place _xx(S, 0) and be positioned at the R at n=214 place _xx(S, 34).

We identify this method advantage of DFT used in application US12804163 relatively at once.Because DFT only quantizes, for this restriction of resolution, make to find the true score cycle count between 5 and 6.Yet self correlation, gives us actual cycle length estimation more accurately.In order to determine amount of cycles,

N _periods=180/34=5.29412

This is not only more accurate than DFT, and attainable more more accurate than very wholwe-hearted observer, because be also difficult to via perusal estimation fractional cycle for us.

Provide the ability of fractional cycle counting, in the time of in being incorporated into IV monitoring and control system (seeing application US13019698), there is very important advantage:

Example:

If resolution is restricted to number of cycles, if doctor formula 62 drops/minute dropping speed, this is by 6.2 drops corresponding in 6 seconds.Only can identify monitoring and the control appliance of number of cycles can not just determine whether speed has reached 6.2 drops afterwards in the monitoring in 6 second cycle, but only can know after the adjustment repeating and monitoring, and speed is now in [6,7] scope.For approximate as far as possible this speed, be necessary observation period expansion, for longer, in this situation, to be at least 30 seconds, because 62 only have two approximate numbers that are less than self: 2 and 31.Even if observed the drop counting of 31 drops in the time of 30 seconds, actual speed is still in (30,32) _{30 seconds}=(60,64) _{60 seconds}between.In order to obtain accurate speed controlling, must from shorter extending to, reach 1 minute or even longer by observing interval.Can divide the equipment of counting number completely unaffected, and each observation cycle can be as far as possible short.

In clinical practice, nurse seldom waits for one minute or longlyer adjusts or observe transfusion speed.If this equipment is restrained too slow, for nurse and patient, be all inconvenient.If lift his/her situation pointing such as patient, change, this also needs to take a long time again to restrain.

I emphasize that at this this is relatively to apply for one of most important improvement of DFT cycle count in US12804163.Inventor confirms, the embodiment that provides the algorithm of fractional cycle counting by use tests, and causes convergence rate faster.

Attention:

" convergence " is mathematical term and herein for meaning that the actual speed of drop finally drops in the range of tolerable variance of predetermined value after the adjusting-monitoring feedback loop repeating.

Therefore, recommend it as one of recommend method in reality realization herein.

It is also feasible for the drop size of speed I, II and III that Fig. 2 .2.2.1-3 and Fig. 2 .2.2.1-4 illustrate self correlation.

Self correlation	Drop height	Drop size
			5.3 cycles of speed I ≈	34	35
13 cycles of speed II ≈	14	15
			23 cycles of speed III ≈	8	7

Table 2.2.2.1-1, compares drop height and drop size results for the estimation of self correlation cycle

If we compare different pieces of information size { drop height, drop size }, we will find that drop height is obviously the good signal of cycle estimation.Drop size signal can cause 1 Cycle Length error.Yet if can improve signal quality and can make other and adjust, such as increasing frame rate, existence can be eliminated the very large variation of difference.

2.2.2.2 auto-covariance

General synonym:

The inclined to one side auto-covariance that has of sequence of real numbers S is defined as

${\stackrel{\&OverBar;}{V}}_{\mathrm{xx}}(S,m)=\frac{1}{N}\left\{\begin{array}{cc}\underset{n=1}{\overset{N-m}{\mathrm{\&Sigma;}}}\left(S\right[n+m]-\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}S[i\left]\right)\left(S\right[n]-\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}S[i\left]\right)& m\&GreaterEqual;0\\ V_{\mathrm{xx}}(-m)& m<0\end{array}\right.$

And at the top of V, use line and nothing version difference partially.We only discuss without inclined to one side version hereinafter, but also can use inclined to one side version in practice.

Sequence of real numbers S is defined as without inclined to one side auto-covariance

$V_{\mathrm{xx}}(S,m)=\frac{1}{N-\left|m\right|}\left\{\begin{array}{cc}\underset{n=1}{\overset{N-m}{\mathrm{\&Sigma;}}}\left(S\right[n+m]-\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}S[i\left]\right)\left(S\right[n]-\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}S[i\left]\right)& m\&GreaterEqual;0\\ V_{\mathrm{xx}}(-m)& m<0\end{array}\right.$

Wherein we notice be actually S[n] average.V _xx(m) be actually the self correlation that removes average sequence,

$V_{\mathrm{xx}}(S,m)=R_{\mathrm{xx}}(S-{\mathrm{\&mu;}}_s,m),{\mathrm{\&mu;}}_s=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}S\left[i\right]$

V _xx(S, m) is also about 0 symmetry:

V _xx(S,m)=V _xx(S,-m)

Therefore for the situation of calculating positive m, be only enough.

Only for R _xx(S, m), we can select m to guarantee can cover at least three cycles in m frame of video, however this depends on frame rate and the dropping speed of image capture device.

By counting V _xxspacing between (S, 0) and next local maximum is carried out determining of execution cycle.With R _xx(S, m) is similar, V _xx(S, kT _e) will obtain local maximum, wherein T _ethe estimation of the T of indication cycle and k are integers.

The ability that auto-covariance also has can be detected actual cycle length, therefore can calculate fractional cycle counting.

Our experiment has illustrated auto-covariance, and for drop height and drop size signal, the two all has the ability of sense cycle.Its result is illustrated in Fig. 2 .2.2.2-1 in Fig. 2 .2.2.2-4.

Self correlation	Drop height	Drop size
			5.3 cycles of speed I ≈	35	35
13 cycles of speed II ≈	14	15
			23 cycles of speed III ≈	8	8

Table 2.2.2.2-1, compares drop height and drop size results for the estimation of auto-covariance cycle

Similar with self correlation, auto-covariance is also one of recommend method in actual realization.

2.2.2.3 average amplitude difference function (AMDF)

General synonym: comb filter, the pectination method of optimization

For the inclined to one side average amplitude difference function of having of sequence of real numbers S (AMDF), be defined as

${\stackrel{\&OverBar;}{D}}_{\mathrm{xx}}(S,m)=\frac{1}{N}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left|S\right[n+m]-S[n\left]\right|}^k,m>0$

Wherein we use line and nothing version difference partially at the top of D.We have only discussed inclined to one side version hereinafter, but also can use inclined to one side version in practice.

For being defined as without inclined to one side average amplitude difference function (AMDF) of sequence of real numbers S

$D_{\mathrm{xx}}(S,m)=\frac{1}{N-m}\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left|S\right[n+m]-S[n\left]\right|}^k,m>0$

Due to the symmetry about 0, only for m>0, define.Can select k be any on the occasion of, but conventionally use similar 1,2 and so on integer.

And character is

$D(S,m)=\frac{1}{N}{\left|S\right[n]-S[n\left]\right|}^k=\frac{1}{N}{\left|0\right|}^k=0$

For its Mathematical Discussion, refer to [M.J.Ross, H.L.Shaffer, A.Cohen, R.Freudberg and H.J.Manley's " Average Magnitude Difference Function Pitch Extractor(average amplitude difference function spacing extractor); ' ' IEEE Trans.on Acoustics; Speech and Signal Processing(is about the IEEE collection of thesis of acoustics, voice and signal processing), 22 volumes, No. 5; 353-362 page, 1974].

If we compare formula and the autocorrelative formula R of AMDF _xx(S, m), identifies D at once _xx(S, m) is actually R _xxthe remainder item (complement) of (S, m).R _xx(S, m) get displacement to the upper sum of products, and D _xxthe right difference sum that (S, m) gets displacement adds super power k.For R _xx(S, m) works as m=kT _etime obtain local maximum, wherein T _ethe estimation of the T of indication cycle and k are integers; For D _xx(S, m), works as m=kT _eshi Ze is local minimum on the contrary.

Our experiment (except the * in row header) has illustrated drop height and the zone leveling gray scale for all speed I, II and III, and AMDF is feasible.

AMDF	Drop height	Drop size *	Zone leveling gray scale
				5.3 cycles of speed I ≈	35	35	35
13 cycles of speed II ≈	14	15	14
				23 cycles of speed III ≈	8	8	8

Table 2.2.2.3-1 estimates relatively drop height and drop size results for the AMDF cycle

Please refer to Fig. 2 .2.2.3-1 to Fig. 2 .2.2.3-7.For first data in array, axle basis is since 1, thereby actual cycle length is X-1.

The experiment that has inclined to one side version of AMDF does not illustrate for succinct reason, by being multiplied by (N-m)/N without AMDF item partially and can obtaining each value corresponding.This provides the correct result of drop height and zone leveling gradation data.For drop sized data, its result stand all if any version partially for the same problem of speed I ≈ 5.3 periodic signals, and can the same way as shown in next joint revise.

In the next section we to prepare that AMDF is shown also feasible for drop size signal.

* mean and use DFT or other algorithms

2.2.2.4 hybrid algorithm (I) *

We notice for AMDF, and its identification ability seems to be inferior to self correlation and auto-covariance, because it seems to detect exactly the drop size signal cycle.But the excellent in shape of the AMDF curve of drop altitude signal seems in fact for the signal of even poorer quality, to have recommended its oneself, because it has been feasible for zone leveling grey scale signal.So infeasible for drop size signal?

Careful inspection can illustrate (seeing that Fig. 2 .2.2.3-1 is to Fig. 2 .2.2.3-7):

1. for the drop size signal of 13 cycles of speed II ≈ and speed III ≈ 23, detection is correct.

2. for the drop size signal in 5.3 cycles of speed II ≈, because causing wrong Cycle Length, the less local minimum of locating at m=20 (X=21) detects.Obviously at m=35 (X=36), locate there is no darker value, but (deceive) program of still swindling is treated as m=kT _edue to minima.

This can be easy to solve by several corrections:

1. the research of local minimum is restricted to less AMDF value D (S, m).In order formally to realize, threshold value TH can be set so that can only check D (S, m) <TH for local minimum.Also can be for signal series S(or sequence D (S, m) preferably, m ∈ 1 ..., function N-1) solves TH.For example, find the average of local minimum, and requirement wherein on average operation, add line, and ε will be constant value, such as, 1.5.Yet, by firm theoretical basis, find ε or other functions determine TH and are not easy.

2. still there is simple settling mode: DFT+AMDF.Theory is very simple:

A) DFT only has integer resolution, but very accurate;

B) AMDF provides cycle count accurately, but may detect wrong local minimum.

Therefore we combine this two algorithms.Because DFT identifies 5 cycles and only has integer resolution, actual cycle length

If we study in this interval of the AMDF result in 5.3 cycles of speed II ≈, we have clearly avoided m=20 (X=21), and due to j will only obtain m=35 (X=36).In this way, we have obtained both the bests.

Use coarse resolution and then be that the theory of more fine-resolution can find in a series of application, for example, in differential or polynomial numerical solution, wherein this is called as " successively approaching ".Basic concept is the same.

The object of this section of " hybrid algorithm (I) * " by name is, for showing the wisdom combination based on to the algorithm of the understanding of the merits and demerits of each algorithm, can obtain than the better result of independent application.As we are the 2nd joint frequence estimation beginning discusses, will in the disclosure, discuss and be no less than 10 frequency algorithms, and wherein 2,3 and more combinations, One's name is legion.Herein, our statement also discloses the principle of the hybrid algorithm/algorithm combination of the IV detection application based on video/image, and desired protection will be clearer in the claims.

2.2.3 the method for Fourier and Fourier correlation

Fourier transformation has distortion and derivant and the impossible limit of One's name is legion.Method before Fourier family method and the before obvious difference between method are is directly used time-domain signal, and Fourier methods will estimate that at different frequency place it forms component.E ^{j ω}the desirable discrete or continuous ω of t.We have presented the method for using discrete Fourier transform in application US12804163, and in the disclosure, we will describe its continuous method for it.

2.2.3.1 periodogram and discrete time Fourier transform

General synonym: DTFT: discrete time Fourier transform

The periodogram of sequence of real numbers S is defined as

$P\left(\mathrm{\&omega;}\right)=\frac{1}{N}{\left|\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}S\right[n\left]e^{-\mathrm{j\&omega;n}}\right|}^2$

To this definition, please refer to [Prentice-Hall, 1997, Stoica, P., and opinion is drawn in the Introduction to Spectral Analysis(spectrum analysis of R.L.Moses), 24-26 page].

The discrete time Fourier transform of sequence of real numbers S (DTFT) is defined as

$D\left(\mathrm{\&omega;}\right)=\underset{n=1}{\overset{N}{\mathrm{\&Sigma;}}}S\left[n\right]e^{-\mathrm{j\&omega;n}}$

It has inverse transformation

$S\left[n\right]=\frac{1}{2\mathrm{\&pi;}}{\&Integral;}_{-\mathrm{\&pi;}}^{\mathrm{\&pi;}}D\left(\mathrm{\&omega;}\right)e^{\mathrm{j\&omega;n}}\&CenterDot;\mathrm{d\&omega;}$

Periodogram is following associated with DTFT quilt

$P\left(\mathrm{\&omega;}\right)=\frac{1}{N}{\left|D\left(\mathrm{\&omega;}\right)\right|}^2$

This means periodogram be exactly DTFT size square divided by N.Both all can be used to the signal frequency of estimating that we apply.

And periodogram and DTFT are their estimation fractional frequencies with respect to the advantage of DFT.This also can be realized by self correlation, auto-covariance and AMDF and a lot of other algorithms described below, but has different principles.In clinical practice, this will cause convergence rate faster, and this is important improvement for DFT speed counting.For discussion, please refer to [2.2.2.1 saves self correlation].

Note, when computing cycle figure and DTFT, recommendation be the version that removes average that uses signal.This be due to

In other textbooks and document, can appear in the analysis mode of DFT and not in synthesis type.Also there are other conventions, such as all using in analysis and synthesis type for derivation, please refer to [the Discrete-Time Signal Processing(discrete-time signal of Oppenheim and Schafer is processed) (second edition), the 8th chapter].

Discrete in the situation that, by S[n] be decomposed into and have a just in time n complex exponential, and in whole [π, π] closed set, estimate DTFT(and periodogram) for a lot of signals, the major part of its energy is positioned at DC component place, so in DFT can be greater than a lot of other coefficients, but we can be simply from coefficient magnitude relatively eliminating c ₀.For DTFT(and periodogram), the DC energy of signal is by the low-frequency band being distributed in centered by 0, so get rid of 0, will leave the ω value much with less △ ω increment, and at the D at these ω places (ω) and P (ω) by larger than the D (ω) and the P (ω) that are positioned at corresponding to the higher ω place of signal AC component, and if we compare when D (ω) or P (ω) size are carried out estimating signal frequency and can cause problem.The simplest scheme is the average that removes signal.

Can only be similar to closed set, so we are subtly by each be divided into m piece m can choose arbitrarily.In Fig. 2 .2.3.1-1, m=100, so resolution is very meticulous.Give DFT for comparing.We see that DTFT and DFT are proportional; Periodogram can have the smaller value through amplifying due to square higher value of eliminating on DTFT, and relatively, this has implied that the contrast between each value is strengthened basically.

By scanning simply DTFT/ periodogram sequence, will complete frequence estimation.DTFT/ peak value of periodogram for the drop altitude signal in 13 cycles of speed II ≈ is positioned at place, at S[n] on amount of cycles can be calculated as

$P=\frac{\frac{2\mathrm{\&pi;}}{m\&CenterDot;N}\&CenterDot;1287\&CenterDot;N}{2\mathrm{\&pi;}}=\frac{1287}{m}=\frac{1287}{100}=12.87$

This obviously has higher resolution than 13 cycle counts of DFT.

If we compare the result of { self correlation, auto-covariance, AMDF, DTFT/ periodogram } of the drop altitude signal in 13 cycles of speed II ≈:

?	Self correlation	Auto-covariance	AMDF	DTFT/ periodogram
					Cycle Length	14	14	14	?
Cycle count	180/14=12	180/14=12.857	180/14=12.857	12.87

Concordance is very approaching.Therefore confirmed the accuracy of these four methods simultaneously.

Attention is aspect calculating:

In order to improve efficiency, we can use and be similar to " successively the approaching " of describing in [2.2.2.4 saves hybrid algorithm (I)], wherein after DFT, carry out AMDF.First we can use the approximately estimation of DFT location, and then only near the discrete ω being determined by DFT, calculate the ω of meticulous enhancing.In this above-mentioned situation, 12.857 two integers around, i.e. [12,13], are converted into [1201,1301] of DTFT/ periodogram index.Then only exist m=100 ω value to need to calculate.This is all feasible for any m.

Also verified that DTFT/ periodogram is for { drop height, drop size } * { any other signal in III} is feasible for speed I, II.The result of the drop size signal of speed II is illustrated in Fig. 2 .2.3.1-2.

2.2.3.2Bartlett periodogram equalization

General synonym: Bartlett method, the periodogram of equalization

Bartlett periodogram is on average a kind of distortion of periodogram, and is defined as

$B\left(\mathrm{\&omega;}\right)=\frac{1}{N}\underset{i=0}{\overset{K-1}{\mathrm{\&Sigma;}}}{\left|\underset{n=0}{\overset{L-1}{\mathrm{\&Sigma;}}}S\right[\mathrm{iL}+n\left]e^{-\mathrm{j\&omega;n}}\right|}^2,\mathrm{KL}=N$

It is upper that equalization occurs in each section of length L, and through square large young pathbreaker in K such section, add up, and finally divided by N.The local average of isomorphism on each section of length L, spectral resolution will reduce K, and change, can expect to be reduced.

For whole discussion, please refer to the 8.2.4 joint of [Hayes, the Statistical Digital Signal Processing statistical Digital Signal Processing of and Modeling(and the modeling of M.) (Wiley, 1996)].

We have carried out { drop height, drop size } * { speed I, II, III} * { L=90, the experiment of 60,30,15}, altogether 24 various combinations.In the drop size signal result of speed II shown in Fig. 2 .2.3.2-1, and for periodogram result relatively.In each figure, we can be clear that, that L becomes is less, change (variance) the less while resolution that becomes also reduces.For too small L-value, such as 15, near zero value can become large so cause the problem of estimation aspect.In fact, consider the extreme case of L=1:

$\begin{array}{c}L=1,K=N/1=N\\ B\left(\mathrm{\&omega;}\right)=\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{\left|\underset{n=0}{\overset{1-1}{\mathrm{\&Sigma;}}}S\right[i+n\left]e^{-\mathrm{j\&omega;n}}\right|}^2\\ =\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{\left|S\right[i+0\left]e^{-\mathrm{j\&omega;n}}\right|}^2\\ =\frac{1}{N}\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{\left|S\right[i\left]\right|}^2\end{array}$

That is, B (ω) can be changed into the constant haveing nothing to do in ω.Therefore be noted that to avoid L-value too little.

Also will in Fig. 2 .2.3.2-1, be used to divide subtly the value of the m of △ ω increment to be chosen as 100, but actual realization can be used any value.Only, for periodogram and DTFT, in order effectively calculating and estimation more accurately, can first with DFT, to locate interval and then near this interval, calculate Bartlett periodogram.

2.2.3.3 the periodogram of correlogram/auto-covariance

General synonym:

The correlogram of sequence of real numbers S is defined as

${\mathrm{\&phi;}}_c\left(\mathrm{\&omega;}\right)=\underset{k=-(N-1)}{\overset{N-1}{\mathrm{\&Sigma;}}}{R}_{\mathrm{xx}}(S,k)e^{-\mathrm{j\&omega;k}}$

Wherein Rxx (S, k) is the autocorrelation sequence of definition in [2.2.2.1 saves self correlation].About its Mathematical Discussion, please refer to [Blackman and Tukey1959, Stoica, Petre and Moses, the Spectral Analysis of Signals(signal Spectrum Analysis of Randolph L.), the 2.2nd joint].

Hybrid algorithm (II) *

Definition comparison by this with [2.2.3.1 joint periodogram and DTFT] middle DTFT, in fact it can be defined as autocorrelative DTFT.Because this is derived sequence (that is, the R from S _xx) DTFT, it also can be classified as in [2.2.2.4 saves hybrid algorithm (I)] hybrid algorithm of first introducing.

Why we can carry out estimation frequency with correlogram, mainly contain two reasons:

1. intuitively, as we are seen in [2.2.2.1 saves self correlation], self correlation shows the periodicity the same with signal S itself, be reasonably therefore its DTFT by us with direct from the same periodicity of the DTFT of S.

2. tight evidence will require restriction, so that autocorrelation sequence declines fast enough, and will use the Convolution Properties of DTFT.Detailed content please refer to [Stoica, Petre and Moses, the Spectral Analysis of Signals(signal Spectrum Analysis of RandolphL.) chapter 1 and chapter 2].

Also note that for periodogram/DTFT, in Fig. 2 .2.3.3-1, before calculating correlogram, remove autocorrelative average.

Note, due to symmetry, for correlogram, we have only calculated the first half of length.

Also note that the fact of having explained due in [2.2.3.1 joint periodogram and DTFT], when calculating continuous DTFT, we have removed R _xxthe average of sequence.Otherwise at low frequency place, also have high value, this can cause the difficulty of cycle estimation.

In Fig. 2 .2.3.3-1, the m=100 of DTFT and correlogram peak value are found to be positioned at X=2571 place, φ _c(ω) length of sequence is 2N-1=359.Then, the amount of cycles in original series S can be calculated as

$P=\frac{\frac{(X-1)\&CenterDot;2\mathrm{\&pi;}}{(2N-1)\&CenterDot;m}\&CenterDot;N}{2\mathrm{\&pi;}}=\frac{X-1}{m}\&CenterDot;\frac{N}{2N-1}=\frac{2570}{100}\&CenterDot;\frac{180}{359}=12.886$

Table 2.2.3.3-1 is correlogram and additive method relatively

When with above-mentioned otherwise result relatively time, correlogram result is found to be with them and approaches very much.Therefore confirmed the accuracy of correlogram.

Also for { its remainder in III} * { drop height, drop size } signal is tested for speed I, II, has all obtained measuring accurately.

Because we are comparing DTFT size, they are only different from periodogram with square operation, obviously known in constant ratio (constant scale), and autocorrelative periodogram is also feasible.

2.2.3.4DTFT/ the periodogram of auto-covariance

General synonym:

As the DTFT on our definable autocorrelation sequence, we go back the DTFT in definable autocovariance sequence.Also there is another example, can the method from our algorithm instruction catalogue (repertoire) in without stint derive hybrid algorithm.

This hybrid algorithm is defined as simply

${\mathrm{DTFT}}_{V_{\mathrm{xx}}}\left(k\right)=\underset{k=-(N-1)}{\overset{N-1}{\mathrm{\&Sigma;}}}{V}_{\mathrm{xx}}(S,k)e^{-\mathrm{j\&omega;k}}$

Auto-covariance V _xxsequence can be have inclined to one side or without inclined to one side.

Due to symmetry, at V _xxin, we can only calculate V _xxthe first half or later half.{ drop height, drop size } * { result of III} data is illustrated in Fig. 2 .2.3.4-1 and Fig. 2 .2.3.4-2 for speed I, II, wherein only for the inclined to one side V of nothing _xxhalf calculating of sequence and the DTFT100 precision of getting DFT.We confirm that result is correct.

Due in experiment, what we compared is the size of the coefficient of DTFT, and they are only different from periodogram with square operation at constant, obviously known, and the periodogram of auto-covariance is also feasible.

2.2.3.5 discrete cosine transform (DCT) and discrete sine transform (DST)

General synonym:

DCT and DST are the modification of DFT, have represented the sequence of real numbers S with real coefficient.Depend on the different choice that delimiting period and symmetry are expanded, have at least 16 different modification of DCT and DST.For these, please refer to [Wang, Z.1984., the fast algorithm of the DISCRETE W TRANSFORM of Fast Algorithms for the Discrete W Transform for the Discrete Fourier Transform(discrete Fourier transform), IEEE Trans, on ASSP(is about the IEEE collection of thesis of ASSP), Vol.32 (4), 803-816 page] and [Martucci, S.A.1994., the symmetrical convolution of Symmetric Convolution and the Discrete Sine and Cosine Transforms(and discrete sine and cosine transform), IEEE Trans on Signal Processing(is about the IEEE collection of thesis of signal processing), Vol.42 (5), 1038-1051 page].Yet, do not have a kind of modification to there is internal difference, so we will use two kinds of modal version d CT-II and DST-II: its use is described.

For DCT, also please refer to [Proakis, the Digital Signal Processing(Digital Signal Processing of John G.) (the 4th edition), the 7.5th joint].DST can find in numerous other documents.

In order to represent to have the sequence of real numbers of real coefficient, must make the expansion of original series.For DCT-II, left and right sides " upset " original series first, and be then attached to after original series.Formal

$S_{\mathrm{DCT}-\mathrm{II}}=\left\{\begin{array}{cc}S\left[n\right],& 0\&le;n\&le;N-1\\ S[2N-1-n],& N\&le;n\&le;2N-1\end{array}\right.$

We are by S _dCT-IIdFT as popular example:

$\begin{array}{c}{\mathrm{DFT}}_{S_{\mathrm{DCT}-\mathrm{II}}}\left(k\right)=\underset{n=0}{\overset{2N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}\&CenterDot;\mathrm{nk}}\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}\&CenterDot;\mathrm{nk}}+S_{\mathrm{DCT}-\mathrm{II}}(2N-1-n)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}(2N-1-n)\&CenterDot;k}\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;\mathrm{nk}}+e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(2N-1-n)\&CenterDot;k}]\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k}+e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}-n)\&CenterDot;k}]}\\ =e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k}+e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}-n)\&CenterDot;k}]\\ =e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;2\&CenterDot;\cos (\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k)\end{array}$

∴

${\mathrm{DCT}-\mathrm{II}}_s\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DCT}-\mathrm{II}}\left(n\right)\&CenterDot;2\&CenterDot;\cos (\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k)$

Due to symmetry, top n DCT-II coefficient has comprised the full detail of sequence spreading, and can be by first using each be multiplied by DCT-II sequence, then get contrary DFT, carry out reconstruct S _dCT-II.The top n result of contrary DFT will be original series S.

Yet what we were interested is the amount of cycles of counting original series S.Strict, " upset " operation owing to setting up sequence spreading, very applicable this object of DCT.But if the periodicity in primary signal S is really very strong, similar with measuring period from DFT, this is also still feasible.

Fig. 2 .2.3.5-1 illustrates the DCT-II expansion with the original drop height of comparing, speed II ≈ 5.36 periodic signals.Attention is aspect periodicity, and how symmetric extension has increased indefinite property to signal.

Even if having the built in problem of cycle explanation, Fig. 2 .2.3.5-2 still illustrates other accurate cycle count of correct integer level of the drop altitude signal of speed I, II and III.For example, in second figure, because the non-DC component of largest amount has index 26(radix 1), the cycle should be (26-1)/2=12.5, consistent with additive method.

Fig. 2 .2.3.5-3 illustrates the DCT-II result for the drop sized data of all speed.Result for speed I and II is correct.For 23 cycles of speed III ≈, the value at X=48 place is 613, substantially equals at 615 of X=2 place.This has implied, in order to use safely DCT-II, we may need to limit the signal of largest amount coefficient hunting zone or the better quality of needs.

DST-II and DCT-II do not exist together and are only its not only " upset " original series, also put upside down (getting negative) sequence so that the new sequence elimination of odd symmetry cosine coefficient retained sinusoidal sequence simultaneously.

$S_{\mathrm{DST}-\mathrm{II}}=\left\{\begin{array}{cc}S\left[n\right],& 0\&le;n\&le;N-1\\ -S[2N-1-n],& N\&le;n\&le;2N-1\end{array}\right.$

We are by S _dST-IIdFT as popular example:

$\begin{array}{c}{\mathrm{DFT}}_{S_{\mathrm{DST}-\mathrm{II}}}\left(k\right)=\underset{n=0}{\overset{2N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}\&CenterDot;\mathrm{nk}}\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}\&CenterDot;\mathrm{nk}}-S_{\mathrm{DST}-\mathrm{II}}(2N-1-n)\&CenterDot;e^{-j\frac{2\mathrm{\&pi;}}{2N}\&CenterDot;(2N-1-n)\&CenterDot;k}\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;\mathrm{nk}}-e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(2N-1-n)\&CenterDot;k}]\\ =\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k}-e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}-n)\&CenterDot;k}]\\ =e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;[e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2}k)}+e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}n)\&CenterDot;k}]\\ =e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;2j\&CenterDot;\sin (\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k)\\ =j\&CenterDot;e^{-j\frac{\mathrm{\&pi;}}{N}\&CenterDot;(-\frac{1}{2}k)}\&CenterDot;\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;2\&CenterDot;\sin (\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k)\end{array}$

∴

$\mathrm{DST}-{\mathrm{II}}_S\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\&Sigma;}}}{S}_{\mathrm{DST}-\mathrm{II}}\left(n\right)\&CenterDot;2\&CenterDot;\sin (\frac{\mathrm{\&pi;}}{N}\&CenterDot;(n+\frac{1}{2})k)$

Yet, must warn, unless original series is symmetrical roughly around X-axis, before making expansion, need to from original, deduct the average of sequence.

Fig. 2 .2.3.5-4 illustrates the data for drop height speed I 5.3 cycles of ≈, the expansion of directly overturning+putting upside down make the first half for just and later half for negative.If there is any possible periodicity in sequence, immediate numeral will be one.Can from the 3rd subimage of Fig. 2 .2.3.5-4, find out, amplitude peak coefficient is positioned at X=2(radix 1) locate, 2-1=1, and this means expansion sequence in only there is one-period.

The correct way of expansion is first from original series, to deduct average, then do normal DST-II expansion.

{ drop height, drop size } * { result of III} is illustrated in 2.2.3.5-5 and 2.2.3.5-6 for speed I, II.The X index from amplitude peak coefficient, deduct one (owing to having used radix 1) and divided by 2 after, we confirm that all cycle counts are correct.

Obviously filling and the expansion scheme of other One's name is legions of DCT and DST are useful, but neither one and these two formations essential distinction herein.

2.2.4 wavelet transformation

General synonym: Haar conversion, wavelet method

Basic concept after wavelet transformation is multiresolution and bank of filters.There is low pass filter and high pass filter.At signal, by low pass filter and be down-sampled after (being labeled as ↓), result has represented the low frequency local component of signal; On the other hand, through down-sampled high pass filter signal, represented the high frequency partial component of signal.At the wavelet transformation of simple form, i.e., in Haar conversion, low frequency local component is exactly the local average of adjacent component, and high frequency partial component is two differences between adjacent component.

For discussing in detail of wavelet method, please refer to [the Wavelets and Filter Banks(wavelet and filter banks of Strang and Nguyen)] and [Daubechies, the Ten Lectures on Wavelets(of Ingrid is about ten lecturees of small echo)].

[Application of the Wavelet Transform for Pitch Detection of Speech Signals(application wavelet transformation detects for the spacing of voice signal), IEEE Transaction on Information Theory(is about the IEEE opinion problem of information theory), 38 volumes, the 2nd phase, in March, 1992], ShubhaKadambe and G.Faye Boudreaux-Bartels have described the method that detects spacing in voice signal.They have carried out the wavelet transform type that is known as second order wavelet transformation in varing proportions on voice signal, and have located in whole maximums the local maximum that surpasses 80%.If between two ratios, local maximum has any position consistency (agree), and this position is by the maximum being identified as corresponding to the transition being caused by glottis closure.Then the time that they can measure between adjacent local maximum is determined Cycle Length.

The advantage of this method stems from the on-fixed essence of voice signal.Voice signal converts in time, and comprises a lot of dissimilar sound such as consonant and vowel.It will be static hypothesis that this Kadambe method is not made signal, and reality has detected local maximum:

t(M ₁),t(M ₂),...t(M _k)

For this t (M _i+1)-t (M _i) alter a great deal.

On the other hand, IV drop process, except moving patient in the moment or the moment because of other activities of its arm, can be assumed to be the signal of static type substantially.Although be not attractive especially, still can detect with wavelet transformation the cycle of this type signal, and this principle is similar to time domain approach to a certain extent.

Use the theory identical with Kadambe, to this application, use a possible realization of small echo to be described to:

1) select small echo type

2) calculate iteratively extreme narrowing approximation coefficient c _awith difference coefficient c _d.After iteration each time, abandon c _d(or being set to complete zero) is also purely from c _amiddle reconstruction approximate signal, then by c _a(or rebuilt approximate signal) is sent to next iteration.After the iteration of specific quantity, stop this process.

3) (designer generally needs the quantity of own specified level) rank in the end, find its value surpass the particular percentile ε of global maximum local maximum (> or >=two consecutive values, or other combinations).For each local maximum X, be upwards scanned up to original S: if in all levels the adjacent local maximum that comprises of [X-K, X+K], by this location recognition, be signal peak.

4) counting number of peaks is as amount of cycles.

Note, we are to guarantee that approximate reconstruction illustrates peak value in the position identical with primary signal at the reason of search in [X-K, X+K], therefore need the region of search.Also please refer to [the Wavelets and Filter Banks(wavelet and filter banks of Strang and Nguyen)] and [Daubechies, the Ten Lectures on Wavelets(of Ingrid is about ten lecturees of small echo)].

Also note that if the length of selected small echo (being Daubechies D8 in our example) is greater than 2, need to guarantee proper operation at starting and ending place filling (expansion) point of S filling.Wavelet transformation itself does not propose any restriction for how making filling, and common selection comprises periodic expansion (similarly being that roll extrusion is to the other end), derivative expansion (calculating the extension point that derivative interpolation at end points place have derivative), zero expansion and constant expansion.In our example, we have selected constant expansion, because this cycle count algorithm for us can not cause problem.For Daubechies D8 small echo, by by original S filling being as getting off to complete

If use other filling schemes in reality realizes, must carefully for algorithm, not create any " pseudoperiodicity ".

For the drop altitude signal of all speed I, II and III, use following parameter to make experiment:

Table 2.2.4-1, the parameter of using in small echo cycle count

Result is illustrated in Fig. 2 .2.4-1 to Fig. 2 .2.4-3.The algorithm with above-mentioned parameter correctly identifies the peak value at all speed I, II and III place.With the warning triangle peak upwards referring to.Third and fourth grade of other approximate reconstruction is also illustrated in the second approximate reconstruction below, thus therefrom we can find out excess level approximate can make signal too planarization cause carrying out peak value detection.

Certainly, actual realization can adjust above-mentioned parameter, and above-mentioned detailed algorithm and parameter should only be regarded as example and unrestricted.

The Daubechies family of small echo has different length, and also has the small echo of a lot of other types, such as { biorthogonal, cubic spline (cubic spline), Haar, sombrero (Mexican hat), Morlet, Meyer, the type that symlets} and customization are set up.But these and wes' algorithm does not all form essential distinction.

Daubechies D2 small echo equals Haar small echo, and uses its wavelet transformation to be also referred to as Haar conversion.

Depend on signal quality, can rational expectation, the small echo of other types and parameter can be used to peak value (peak value) counting in this application.

2.3 parametric technique

Definition:

Parametric technique hypothesis signal meets the generation model with known function form, and the parameter in the model of then supposing by estimation continues.Then the interested spectral characteristic of sending out signals from the model of estimation.

[Stoica, Petre and Moses, the Spectral Analysis of Signals(signal Spectrum Analysis of Randolph L.), the 3.1st joint, 90 pages]

There is very eurypalynous parametric technique, comprise based on following these

1. autoregression model (AR)

2. moving average model(MA model) (MA)

3. autoregression model-moving average model(MA model) (ARMA)

4. subspace/eigenvector method

As mentioned in [2 joint frequence estimation] in us, frequency spectrum estimation as the subject of having set up has had over centenary history ([Stoica, Petre and Moses, the Spectral Analysis of Signals(signal Spectrum Analysis of Randolph L.) preamble, XV page], [Marple, L. Digital Spectral Analysis with Applications(has the digital spectrum analysis of application)]), and for different application, created different spectral estimation, wherein signal all has the characteristic of its uniqueness.A lot of parametric techniques (just as above-mentioned a lot of nonparametric techniques) are feasible for our signal, this be only because the quality of these signals fairly good (static, there is no sound pollution/destruction, a simple procedure; Deng), not our signal certain claim is correctly estimated with these parametric techniques.In fact, as the example by { self correlation, auto-covariance, AMDF, periodogram DTFT, correlogram }, we are visible, and the accuracy of these algorithms is enough good, and in fact should admit improved space very little (it's not true if not).In practice, the realization of a lot of parametric techniques is generally comparatively difficult.Wherein a lot of methods need to solve system of linear equations or solve higher order polynomial equation (for example, finding all eigenvalues of matrix).Therefore, also can relate to numerical value item.

At the AR/MA/ARMA of algorithm apoplexy due to endogenous wind, we will describe Yule-Walker method.In subspace/eigenvector method, we will describe multiple emitter location and signal parameter estimation (MUSIC) and Pisarenko Harmonic Decomposition.These two methods are known, widely used, and have represented its correlation method class.

2.3.1 autoregressive spectrum estimation

2.3.1.1Yule-Walker method

General synonym: autocorrelation method (not obscuring with autocorrelation method before)

Parametric technique is made hypothesis, and output S can be modeled as the difference equation that relates to self and list entries w:

$S\left[n\right]=-\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{a}_{k}S[n-k]+\underset{k=0}{\overset{q}{\mathrm{\&Sigma;}}}{b}_{k}w[n-k]---(\mathrm{2.3.1}-1)$

Thereby

$H\left(z\right)=\frac{B\left(z\right)}{A\left(z\right)}=\frac{\underset{k=0}{\overset{q}{\mathrm{\&Sigma;}}}{b}_k{z}^{-k}}{1+\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{a}_k{z}^{-k}},$

and

$H\left(e^{\mathrm{j\&omega;}}\right)=\frac{B\left(e^{\mathrm{j\&omega;}}\right)}{A\left(e^{\mathrm{j\&omega;}}\right)}=\frac{\underset{k=0}{\overset{q}{\mathrm{\&Sigma;}}}{b}_k{e}^{-\mathrm{j\&omega;k}}}{1+\underset{k=1}{\overset{p}{\mathrm{\&Sigma;}}}{a}_k{e}^{-\mathrm{j\&omega;k}}}$

Referring to formula (2.3-1),

1. if q=0, S[n] depend on self before value, be therefore called as autoregression (AR) model;

2. if p=0, S[n] be w[n] the rolling average of weighting, so be called rolling average (MA) model;

3. if P ≠ 0 and Q ≠ 0, S[n] depend on value and w[n before it] rolling average, so be called ARMA model (ARMA).

Yule-Walker is for estimating the method for the frequency of AR model.For sequence of real numbers S, the steps include:

1) from its autocorrelation sequence, generate Toeplitz matrix:

Herein with there being inclined to one side self correlation to guarantee that matrix one is just decided to be.Also can use without partial autocorrelation and about details and please refer to [Hayes, the Statistical Digital Signal Processing statistical Digital Signal Processing of and Modeling(and the modeling of M.) (Wiley, 1996)].

2) with Levinson-Durbin or additive method, solve

3) arrange ${\left|b_0\right|}^2=r\left(0\right)+\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_{k}r\left(k\right)$

4) estimation power frequency spectrum is $\frac{{\left|b_0\right|}^2}{{|1+\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{a}_k{e}^{-\mathrm{j\&omega;k}}|}^2}.$

Fig. 2 .3.1.1-1 illustrates Yule-Walker method for { drop height, drop size } * { III} data are feasible for speed I, II to the experiment in Fig. 2 .3.1.1-6.Thereby draw these figure with logarithmic scale, make local peaking clearly visible, otherwise they will be become too little by bi-directional scaling and cannot see.For each data set, the rank of AR model are increased to 50 from 5.

If we are (Fig. 2 .3.1.1-1 and Fig. 2 .3.1.1-3) and (Fig. 2 .3.1.1-4 and Fig. 2 .3.1.1-6) relatively, we will notice, compare the signal (23 cycles of speed III ≈) of more speed, lower rate signal (5.3 cycles of speed ≈) needs the more AR model of high-order to realize frequence estimation accurately.This is consistent with AR model: speed is lower, the cycle is longer, so each S[n] will depend on the longer sequence of value before it, so the AR model of high-order this sequence of modeling and so will provide frequence estimation more accurately better more.

Also please note that at Fig. 2 .3.1.1-1 axle label is normalized to [0, π] in Fig. 2 .3.1.1-6.In order to obtain the amount of cycles of N sample point, we are multiplied by N by the normalized X-axle value of peak value and simply then divided by 2 π.For better accuracy, below calculate and use non-normalized index.

For drop height, speed II ≈ 13 periodic signals, [0, π] is divided into 18000 parts and should from the index of maximum power spectrum value, count cycle quantity is

$\frac{\frac{(\mathrm{Index}-1)\&CenterDot;\mathrm{\&pi;}}{18000}\&CenterDot;N}{2\mathrm{\&pi;}}=\frac{(\mathrm{Index}-1)}{2}\&CenterDot;\frac{180}{18000}=\frac{(\mathrm{Index}-1)}{200}$

Table 2.3.1.1-1, is used the speed II drop altitude signal cycle estimation of Yule-Walker method

By it and the method comparison of table in 2.2.3.3-1, we find that high-order is more general always essential to provide estimation accurately, this with above-mentioned we about the discussion of Cycle Length/dropping speed and model, be also consistent.

As we above as described in, there is { AR, MA, the parameter model of ARMA} three types.With respect to two outer two kinds of modeling types, the reason that we select AR be because:

1.S is generally periodic signal, and obviously depends on the value before it.AR model is applicable to this physical basis best; Because the hypothesis of MA model is inconsistent with reality, may from MA model, there is poor estimation.

2. most of practicalness problems of frequence estimation will have the power frequency in characteristic frequency (the available utmost point (pole) carrys out modeling) center.For the identical configuration of the utmost point, may need to compare AR model more the MA model of high-order be similar to them.

3. owing to showing that from discussing 1 AR is applicable to and show that from discussing 2 MA is not suitable for, any arma modeling of modeling signal is using AR as its major part satisfactorily.Therefore, there is no essential distinction with AR model.

Although we reach a conclusion, in parameter model, AR model is applicable to our signal the most, if the quality of data is very good and/or relate to some other processing, can not get rid of the probability of carrying out modeling period IV monitor signal with MA or ARMA.Therefore, discussion before is only regarded as a kind of recommendation rather than any restriction of actual realization.

And certainly the same with the situation of additive method, we can realize more high efficiency or improved accuracy by AR modeling and additive method (such as DFT) combination.

2.3.2 characteristic vector/subspace method

2.3.2.1 multiple emitter location and signal parameter are estimated (MUSIC) and Pisarenko Harmonic Decomposition

General synonym:

Relevant: Pisarenko Harmonic Decomposition (equivalence when M=P+1)

Multiple emitter location and signal parameter estimation (MUSIC) are the improvement for Pisarenko Harmonic Decomposition.The difference of these methods is that " frequency spectrum " that we obtain is no longer the estimation of actual physics power frequency spectrum, but only for estimating complex exponential frequency.

In Pisarenko Harmonic Decomposition, S is assumed to be P complex exponential sum in white noise.The P+1 value of autocorrelation sequence is known or estimation.In (P+1) * (P+1) autocorrelation matrix, the dimension of residual noise subspace will be (P+1)-P=1 and by corresponding to minimal eigenvalue v _mincharacteristic vector scan.V _minwill with whole signal space quadrature, therefore with all signal characteristic vector quadratures, so

${e_i}^H{v}_{\min }=\underset{k=0}{\overset{P}{\mathrm{\&Sigma;}}}{v}_{\min }\left(k\right)e^{-\mathrm{jk}{\mathrm{\&omega;}}_i}=0,i=\mathrm{1,2},...,P$

It follows v _mindTFT at each ω _iplace will be zero, thus its against (with and square amplitude contrary) will be at ω _ifrequency place shows sharp-pointed peak value, and therefrom will understand the complex exponential frequency of composition.

This is contrary, and we are expressed as

$I\left(\mathrm{\&omega;}\right)=\frac{1}{{\left|\underset{k=0}{\overset{P}{\mathrm{\&Sigma;}}}{v}_{\min }\left(k\right)e^{-\mathrm{jk\&omega;}}\right|}^2}=\frac{1}{{|e^H\&CenterDot;v_{\min }|}^2}=\frac{1}{{\left|\mathrm{DTFT}\left(v_{\min }\right)\right|}^2}$

Be called as pseudo-frequency spectrum (or characteristic frequency spectrum) thus itself and the actual/physical power spectrum region occurring in additive method are separated.

For deriving and discussing, refer to 8.6 joints of [Hayes, the Statistical Digital Signal Processing statistical Digital Signal Processing of and Modeling(and the modeling of M.) (Wiley, 1996)].

In Pisarenko Harmonic Decomposition, the length that the quantity of signal characteristic vector is assumed to be autocorrelation sequence deducts one; If not so, remove this requirement and for the autocorrelation matrix of M * M, desirable other values of P are so that M>P+1, and then it can become the hypothesis that multiple emitter location and signal parameter are estimated (MUSIC) method.

Therefore be clear that, when M=P+1, these two methods are equivalent, so Pisarenko Harmonic Decomposition can be classified as the particular case of MUSIC.

In MUSIC method, then calculate pseudo-frequency spectrum I (ω) and be

$I_{\mathrm{MUSIC}}\left(\mathrm{\&omega;}\right)=\frac{1}{\underset{i=P+1}{\overset{M}{\mathrm{\&Sigma;}}}{\left|\underset{k=0}{\overset{M-1}{\mathrm{\&Sigma;}}}{v}_i\left(k\right)e^{-\mathrm{jk\&omega;}}\right|}^2}=\frac{1}{\underset{i=P+1}{\overset{M}{\mathrm{\&Sigma;}}}{\left|{\mathrm{DTFT}}_{\mathrm{\&omega;}}\left(v_i\left(k\right)\right)\right|}^2}=\frac{1}{\underset{i=P+1}{\overset{M}{\mathrm{\&Sigma;}}}{|e^H\&CenterDot;v_i|}^2}$

V wherein _i' s is the characteristic vector of noise subspace.The impact of equalization will be to have reduced false peaks.For discussing in detail, refer to the 8.6.3 joint of [Hayes, the Statistical Digital Signal Processing statistical Digital Signal Processing of and Modeling(and the modeling of M.) (Wiley, 1996)].

Use multiple emitter location and signal parameter estimation (MUSIC, we also classify as its particular case by Pisarenko Harmonic Decomposition) for the step of estimating signal frequency to be:

1., if use general MUSIC method, select P and M>P+1; If use Pisarenko Harmonic Decomposition, determine P and M=P+1.

2. calculate the autocorrelation matrix of dimension M * M.

3. calculate eigenvalue and the characteristic vector of this autocorrelation matrix.

4. calculate pseudo-frequency spectrum I _mUSIC(ω).Find its frequency corresponding with largest amount to use this frequency as frequence estimation.

Certainly, we also can locate interval with DFT, then use MUSIC or Pisarenko Harmonic Decomposition for meticulousr estimation.

By M is set, as signal S length the different value place between 2 to 50, change P, test.Cycle count in the estimation of the drop altitude signal at speed II ≈ 13 cycles places is illustrated in form.The numeral of underscore is that index and [0,2 π] of maximum pseudo-spectrum value is divided into 10000 little increments.Cycle count is calculated as

$\frac{X\&CenterDot;\frac{2\mathrm{\&pi;}}{10000}\&CenterDot;180}{2\mathrm{\&pi;}}=0.018X$

P	Cycle count
		2	( 837-1)×0.018=15.048

4	( 725-1)×0.018=13.032
		6	( 693-1)×0.018=12.456
8	( 721-1)×0.018=12.96
		10	( 713-1)×0.018=12.816
20	( 711-1)×0.018=12.78
		30	( 719-1)×0.018=12.924
50	( 720-1)×0.018=12.942

Table 2.3.2.1-1, is used different P for the cycle estimation of speed II drop height

By with table 2.2.3.3-1 comparison, we utilize the P of suitable selection as seen, compare additive method, MUSIC algorithm will cause estimating accurately.

Fig. 2 .3.2.1-1 illustrates these methods for { drop height, drop size } * { III} data are feasible for speed I, II to Fig. 2 .3.2.1-6.Depend on signal quality, for drop size signal, can need larger P(such as surpassing 80).

Also note that with logarithmic scale and illustrate Fig. 2 .3.2.1-1 to Fig. 2 .3.2.1-6.There are two reasons: (1), if illustrated with linear scale, a lot of smaller values will almost cannot be seen.(2) shown go out actual is pseudo-frequency spectrum, thus we needn't as before with linear scale, map shown in experiment.

Therefore we illustrate, not via any mode (periodogram or any other) of the original power frequency spectrum of estimation but via by the pseudo-frequency spectrum of Eigenvalues Decomposition gained, and also can estimation frequency.The conclusion that we draw is, for our IV speed monitoring application, also can use pseudo-spectral method.

3. machinery is controlled

In application US12804163, we disclose a lot of key technologies of the IV monitoring based on video/image monitoring.In chapters and sections before of the present disclosure, we have expanded their scope.

In application US13019698, we disclose the mechanism of controlling IV dropping speed, so use monitoring and control combination, can build automatic IV monitoring and control system.

In this section, we will disclose other possible mechanism that IV dropping speed is controlled.

Fig. 3-1st, the general synoptic diagram of the mechanical subsystem of our IV monitoring and control system.Compare miscellaneous part, driving screw is shown more significantly, because we think that this is vital for system.Yet we emphasize not make any restriction that actual realization must be used driving screw herein, and the function of driving screw can be substituted by miscellaneous part.

Following chapters and sections will be often referring to Fig. 3-1.

3.1 pipe depressor and supporters

Equipment such as Fig. 3 .1-1 is used in the conventional drop based on chamber, for regulating dropping speed.By pressing pipe, adjust its thickness and work.The thickness that depressor rolls and changes pipe by pressing it in groove.

For the IV system of automatic control, we need to have some object of similar functions.

First we need some things to press the sectional area that pipe passes through to change liquid.This is not difficult especially problem.For example, we can control pipe thickness with our finger.Can effectively change the long-pending any mechanism of pipe section all feasible.

For the shape of depressor and the rear support device of pipe, there is no special requirement.Loose, thereby as long as they can mate good effectively pressing tube sublist area.Fig. 3 .1-2 illustrates side (or front) diagram that some of rear support device of pipe may shapes.

From axial (top/bottom) of pipe, it seems, also exist the coupling of One's name is legion and complementary shape to form the combination of depressor/supporter.In five examples shown in Fig. 3 .1-3.

In the left side of Fig. 3 .1-4, from the top to the bottom, we provide the example of the shape of the contact point between depressor and pipe, can have angle or smooth or circular.On the right side of Fig. 3 .1-4, we see that angle can be obtuse angle, right angle or acute angle; It can attenuate or expand; At contact point place, it can have sharp edges or flat surfaces.

Formal demand is very simple:

If adjust the relative position of depressor and supporter, can make tube fluid stop, so this two can be used as depressor/supporter pair.

Depressor be what kind of more specifically example be illustrated in Fig. 3 .1-5, B.4 first this be illustrated as application 13019698 figure.In side view, we see that the angle at axle edge is only that 15 ° and we add that annotation represents that it need to become sharp-pointed.Why? in principle, more wedge angle degree is less, contact area will be less at edge, therefore has larger pressure.Experimentally, we have also carried out experiment and have in fact found that edge is sharper, angle is less with difform depressor, and it needs the power from motor still less to drive depressor.

In Fig. 3 .1-5, so depressor is plotted as and has internal whorl and it can be used as nut, be installed in driving screw, thereby the rotation of driving screw will be converted into moving axially of depressor.Also possible, depressor is directly connected with linearly moving parts, and does not need the combination of driving screw-nut.In both cases, all may have from axle and move, and the technology in available [3.3 joint Linear-moving guiding] reduces.

Although also note that we have seen variform a lot of depressor in above-mentioned accompanying drawing, also has the depressor that does not much fall into the other types of this apoplexy due to endogenous wind.Above-mentioned depressor/supporter, to being used to linear actuators, is directly pushed depressor to the direction of pipe.Except linear actuators, also have

1. " pivotable " " utcracker " type is discussed in 3.5.1 joint.

2. rotation type is discussed in 3.5.2 joint.

3. cam embodiment discusses in 3.6 joints.

General synoptic diagram for mechanical subsystem, please refer to Fig. 3-1.

In the next section, we will discuss linear actuators.

Target

What expect is that the mechanism that IV controls has following characteristic:

1. high-resolution, can be accurate thereby control.

2. self-locking, thus do not need energy for maintaining control position.

3. strong power output, thus enough pressure can on IV pipe, be applied.

After between dissimilar mechanism relatively, we find that driving screw is the ideal solution that meets 1-3.

3.2 driving screw

General synonym: drive screw

Driving screw is basic frame for movement and is known for everyone of mechanical engineering work.For discussing and character, please refer to [Mechanical Engineering Design of Shigley, the 8.2nd joint].

From the formula of the driving screw of engineering textbook and handbook be with accurate result from calculus approach very much approximate, this has used enough accurately for us.

We have

$T_{\mathrm{raise}}=\frac{{\mathrm{Fd}}_m}{2}\left(\frac{l+\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m}{{\mathrm{\&pi;d}}_m-\mathrm{\&mu;}\sec \mathrm{\&alpha;l}}\right)=\frac{{\mathrm{Fd}}_m}{2}\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}+\mathrm{\&lambda;})---\left(1\right)$

$T_{\mathrm{lower}}=\frac{{\mathrm{Fd}}_m}{2}\left(\frac{\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m-l}{{\mathrm{\&pi;d}}_m+\mathrm{\&mu;}\sec \mathrm{\&alpha;l}}\right)=\frac{{\mathrm{Fd}}_m}{2}\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}-\mathrm{\&lambda;})---\left(2\right)$

Wherein:

T _raise=lifting moment

T _lower=reduction moment

Load on F=screw rod

D _mthe average diameter of the ring that=screw rod contacts with nut

α=thread angle

L=helical pitch (flight pitch)

λ=helical pitch angle

Coefficient of friction between μ=outside and internal whorl material

φ _eff=effective angle of friction, is defined as φ _eff=tan ^-1(μ sec α).For square thread, it has α=0, φ _eff=tan ^-1(μ) as the General Definition of angle of friction.

Shown in Fig. 3 .2-1 with the figure of the driving screw of above-mentioned symbol institute labelling.

Let us investigates why driving screw has three desirable properties now:

Character I: high-resolution, can be accurate thereby control.

Helical pitch l is generally very little.Consider to comprise one of modal embodiment of driving screw, linear stepping motor, l can be as small as 0.5mm, and the rotation of each step can be made into little of 7.5 °, thereby the stroke of each step (stroke) can be only

$\frac{0.5\mathrm{mm}}{360/7.5}=0.01042\mathrm{mm}$

And the thickness of IV pipe generally at 3mm between 4mm, so if we suppose that tube thickness can be compressed to " zero ", such distance can be divided into 300-400 part in theory.Certainly tube thickness can not be compressed into zero, and maintain 1mm or more while generally pressing completely, above-mentioned figure has still provided the accurate estimation of precision.

Character II: self-locking, thus do not need energy for maintaining control position.

$T_{\mathrm{lower}}=\frac{{\mathrm{Fd}}_m}{2}\left(\frac{\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m-l}{{\mathrm{\&pi;d}}_m+\mathrm{\&mu;}\sec \mathrm{\&alpha;l}}\right)=\frac{{\mathrm{Fd}}_m}{2}\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}-\mathrm{\&lambda;})---\left(2\right)$

For this character, please refer to [Mechanical Engineering Design of Shigley, ISBN0390764876, the 8.2nd joint].Intuitively, if the coefficient of friction between driving screw and nut surface is very big, if we want to make driving screw rotation will become difficult by pressing its axle; Or, if helical pitch l is zero, this means the movement that there is no vertical incremented/decremented, thereby the power on axle does not create moment, it also can not reverse rotation driving screw.On mathematics, if

$T_{\mathrm{lower}}=\frac{{\mathrm{Fd}}_m}{2}\left(\frac{\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m-l}{{\mathrm{\&pi;d}}_m+\mathrm{\&mu;}\sec \mathrm{\&alpha;l}}\right)=\frac{{\mathrm{Fd}}_m}{2}\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}-\mathrm{\&lambda;})>0$

?

$\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m-l>0\&DoubleRightArrow;\mathrm{\&mu;}>\frac{l}{\mathrm{\&pi;}{d}_m}\&CenterDot;\cos \mathrm{\&alpha;}=\tan \mathrm{\&lambda;}\cos \mathrm{\&alpha;}$

In fact this implying does not have independent axial force F, there is no outside moment, can make driving screw reverse rotation yet.

The important results of self-locking character is, can build the portable IV controller of power-efficient.By electromechanical power, create the moment in motor, and if still need moment to maintain position, by the constant consumption existing from the electric current of battery, thereby and in fact this electric current be that very large battery will be exhausted very soon.

Also can maintain position with the brake based on rubber or spring, this also can save energy, but this needs large volume and expensive additional mechanism.In addition,, for rubber brake, loss can make it lose gradually its ability.

Only need to meet an inequality μ >tan λ cos α, driving screw provides the simplest and the most reliable scheme of dealing with problems to us.

Character III: strong power output, thus can on IV pipe, apply enough pressure.

Due to

$T_{\mathrm{raise}}=\frac{{\mathrm{Fd}}_m}{2}\left(\frac{l+\mathrm{\&pi;\&mu;}\sec \mathrm{\&alpha;}{d}_m}{{\mathrm{\&pi;d}}_m-\mathrm{\&mu;}\sec \mathrm{\&alpha;l}}\right)=\frac{{\mathrm{Fd}}_m}{2}\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}+\mathrm{\&lambda;})---\left(1\right)$

This is made every effort to achieve and builds moment

$F_{\mathrm{raise}}=\frac{T_{\mathrm{raise}}}{d_m/2}=F\&CenterDot;\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}+\mathrm{\&lambda;})$

$\frac{F_{\mathrm{raise}}}{F}=\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}+\mathrm{\&lambda;})$

General λ is less, thus tan λ ≈ 0,

$\frac{F_{\mathrm{raise}}}{F}=\tan ({\mathrm{\&phi;}}_{\mathrm{eff}}+\mathrm{\&lambda;})=\frac{\tan {\mathrm{\&phi;}}_{\mathrm{eff}}+\tan \mathrm{\&lambda;}}{1-\tan {\mathrm{\&phi;}}_{\mathrm{eff}}\tan \mathrm{\&lambda;}}\&ap;\tan {\mathrm{\&phi;}}_{\mathrm{eff}}=\mathrm{\&mu;}\sec \mathrm{\&alpha;}$

For various materials, μ is generally between 0.1 to 0.3, and general α≤30 °, so

∴

μsecα≤0.3×1.1547=0.334641

So 1/3 F is enough to it to promote (promotion) at the most.And significantly, herein we as calculated ratio the upper limit, thereby for most metals combination of materials and the angle of thread, this ratio can be even less.

Therefore we reach a conclusion, and driving screw is the desirable frame for movement of controlling IV dropping speed.

Comment about linear electric machine:

The linear electric machine of nearly all type is all used driving screw as their linear actuators.Yet, not every linear electric machine, and even not every linear stepping motor has character I, II and III.Some linear electric machines are designed to have compared with low-friction coefficient μ and higher angle of thread λ, thereby do not meet II.In fact, rule request manufacturer will not consider this three requirements to the linear electric machine of general type simultaneously.

Some may ask why we select term " driving screw " rather than " linear electric machine ", because most of linear electric machine is based on driving screw.This is mainly contained to two reasons:

1. linear electric machine or any motor, be assembling and the combination of different assemblies.When analyzing its engineering properties, first must analyze respectively each assembly, if and in fact we find that linear electric machine has character I, II and III really, in most of the cases (except when for character II, use braking, or while using other assemblies), this is because its driving screw has such character.So-called " linear electric machine " do not guarantee I, II and III.Therefore, we find more suitable relate to essential element rather than all together with miscellaneous part.

2. except in linear electric machine, also can use driving screws in a lot of other places.In fact, as shown in Fig. 3 .-1, for spinning movement is converted to linearity, can there is driving screw one or many at a lot of local places.For example, linear track/slide rail can have driving screw, but it may not use the electric parts of motor completely, so name it, is that " motor " is obviously also improper.Therefore be, to use separately title " driving screw " that our attention is guided to its peculiar property preferably.

3.2.1 difference driving screw

Can there is the modification of the driving screw depressor based on Fig. 3 .2-1.As by mentioned at 3.3 joints, speed-thickness relationship be very nonlinear and when thickness at zero-sum much smaller than between the value of the internal diameter of pipe time, can there is foreseeable variation.In order to go up among a small circle and provide meticulousr control on gamut at this, can use the difference combination of two or more driving screws.Axle is divided into two (or more) part and in each different piece, has the screw thread (spacing) of the difference guiding being polished out.In the example shown of Fig. 3 .2.1-1, the ratio between two spacing is 10:9.In order to realize this, let us can on depressor, make spacing and its corresponding shaft portion will be 1.0mm, and make spacing on fixture and corresponding shaft portion will be 0.9mm at pipe.When these two screw threads have same rigidity, so effective spacing of this configuration is 1.0-0.9=0.1mm.In order to make them become driving screw, certainly need to there is the combination of bearing, key/keyway or tooth bar/groove to prevent their rotations, this is in the drawings for visually knowing and omitted.In the example of this figure, for completely pressing tube is to its diameter 3mm, depressor need to be driven 30mm to the right, and the supporter of pipe is driven 27mm to the right, and relatively moving of they causes the pressing of complete 3mm of pipe.

To such be combined in micromechanics very common.Embodiment in figure, the fixing position of depressor make motor and pipe fixture has and relatively moves or use other similar realizations also.In principle, neither one modification and our example shown have essential difference.

3.3 Linear-moving guiding

In [3.1 pipe joint depressor and supporter], we have obtained the right a lot of selections of depressor/supporter, and we have investigated character I, II and the III of driving screw in [3.2 joint driving screw].Can both cause separately good tube thickness to be controlled to seem them, so why we still need Linear-moving guiding?

This is due to the existence from axial displacement.In Fig. 3 .3-1, the desirable direct of travel of depressor will be in the direction that is plotted as two parallel dotted lines.Yet due to the tolerance of manufacture view, the space between screw thread-nut coordinates and other skew/displacements (in rotary moving such as what come from rotor propagation), more or less always exist from axial displacement.

Angle δ in Fig. 3 .3-1 ₁and δ ₂exaggerate.Yet result is that it can make pipe not close completely, or liquid can not be controlled as enough drop lentamente, and this result is exist actually.Use at us is provided with in the experiment with several difform linear stepping motors of the mean quality of the depressors of manufacture very subtly, and the problem in Fig. 3 .3-1 exists and continues.During beginning, depressor can drop to speed specific size effectively, but after this, depressor is can not be further moving forward, and more double check finds that a side of depressor has touched piping support device and opposite side departs from it, and in this space, drop still can flow through, although slowly.

We want the experiment of emphasizing us to find that the relation between tube thickness and dropping speed is not nonlinear consumingly.For example, the common inner thickness of pipe is 3mm, and typically motor can have the little stroke to 0.0254mm, so in theory this 3mm is divided into 3/0.0L54=118.11 part.Yet, it is in no case linear that experiment illustrates tube thickness-drop Rate Relationship, and for most of initial step,, may for altogether in 118 more than 80, it is almost constant that drop speed maintains, and only in the end in several beating (dozen) step, just change, thereby each step can cause separately the observable variation of dropping speed.

If from axial displacement as so large in Fig. 3 .3-1, in fact, so this position that has stoped several final steps does not reach corresponding to the speed of this scope.This is unacceptable in actual applications, because

1. after drop finishes, pipe must be turned off in case Hemostatic Oral Liquid passes back into pipe.This is the basic demand of automatic IV monitoring and control system.

2. some application require low speed to control really.For example, for neonate, speed controlling can be low to moderate 1 drop/every 3 seconds.

So must solve the problem from axial displacement.Exist a lot of methods to solve this problem.

In Fig. 3 .3-2, use the combination of key/keyway; In Fig. 3 .3-3, use the combination of tooth bar/groove.Note herein for following not strict restriction:

1. key/keyway or tooth bar/groove combination is positioned at inner or outside.

Tooth bar/groove across whether on whole circle.

And although we are plotted as linear actuators by driving screw in two figure, we do not require on driving screw driver part, to use tooth bar/groove or the combination of key/keyway.They can be used to guiding from any assembly (a plurality ofs') Linear-moving.

Being used for guiding linearly moving another mode is by using bearing.Fig. 3 .3-4 illustrates bearing (a plurality of) and can be positioned at outside or agree with the groove/passage/track otch (cut) at moving-member.

All three methods used herein are verified effectiveness.They can be used alone, or are even used in combination to obtain optimum efficiency.

So far our treated actuator (although hypothesis is not finally driven by linear actuators by it yet).In the terminal stage of depressor pressing tube, we also can use the rotary components of a few types.

3.4 driving screws are unique type of Linear-moving parts

We want to be stressed that, above-mentioned Linear-moving guiding is not restricted to and is only applied to driving screw.They are general and are applied to all types of Linear-moving parts, anyway drive this Linear-moving parts.Fig. 3 .4-1 illustrates one of possible mode creating linearly moving One's name is legion, and wherein first driving screw makes lever one side conversion, and then the rotation of the lever other end causes the Linear-moving of slider/depressor.The nut of driving screw has less cylinder shape connector, and it agrees with in the groove in lever, and slider/depressor also has the identical adapter agreeing with in lever slot.For driving screw nut and slider nut, the two all uses bearing, but certainly also can use tooth bar/groove and key/keyway.

Remember, in the mechanical system schematic diagram of Fig. 3-1, the frame of " line movement " is only followed in line movement guiding, rather than the frame of " line movement after driving screw ".The example of Fig. 3 .4-1 shows this difference.

We are also noted that in the example of Fig. 3 .4-1, use lever further to strengthen the precision that depressor moves and increased power.Certainly, also can be different local uses repeatedly, so that it repeatedly conduces the precision of enhancing and the power of increase.

The lever length ratio herein illustrating is only for illustration purpose.According to the relative position of fulcrum, load and power, lever can be classified as three classes, and also can the type of working load between power and fulcrum.

And we also emphasize that we are also connected between lever and linearly moving parts (being nut and slider/depressor) with groove, also can use the connection of other modes herein.Yet neither one and lever principle shown here have essential difference.

3.5 rotary push devices

3.5.1 " utcracker " of pivotable

This mechanism is simple, low-cost and tested, very effective in operation.

Remember, the motivation of introducing " line movement guiding " is because from axial displacement.This displacement is very little and may not affect completely in a lot of application, but because pipe diameter is very little and tube thickness-dropping speed relation non-linear, causes problem for us.

The principle that key/keyway, tooth bar/groove and bearing are used is by maintaining, catch or tightly pushing away to prevent from axial displacement to it.The utcracker of pivotable is by dividing the axial displacement of dissociating to work.

Fig. 3 .5-1 illustrates this figure.The supporter of pipe and depressor, at pivotal axis place, be assembled together, on this pivotal axis depressor or depressor and supporter the two can rotate around it.Not agreeing with of junction should too tightly consequently cannot be rotated.

No matter whether line movement, guided by { key/keyway, tooth bar/groove, bearing etc. }, causes now that depressor upper-arm circumference rotates around pivotal axis.The rear support device of pipe will be fixed and not allow and move, or also can allow to rotate and be connected to driver part.Angle between closed depressor and piping support device has been compressed pipe, and it is contrary opening.

Due to the essence of this mechanism, to be depressor with supporter relative in rotary moving, thus both one of or both can driven rotation.For for purpose of brevity, in being described below, we only describe structure and the connection of the depressor with driver part, and mobile if supporter is also allowed to, this is equally applicable to supporter.

For line movement is converted to rotation, in depressor, cutting has the groove of even width.The width of groove is slightly larger than ball or the cylindrical diameter that is arranged on line movement parts head place, thereby when linear unit moves forward or backward along its axle, this ball or cylinder can have the movement of relative sliding in groove.

Does is a problem why we must use ball or cylindrical shape to agree with in groove? because when linear unit moves and depressor arm when rotation, if we use the head (with the contact component of pressing tank) of linear unit to study the relative action of depressor as zero, we find that in fact depressor arm rotates around contact component, follows and moves radially.This is for not adopting circle or being for the parts of rounded periphery (thereby allow connected something Smooth Rotation and do not repeat collide) to be impossible on its at least one cross section.The feature of such expectation is only had circle or rounded periphery shape in its at least one cross section has, and therefore this is necessary.

The length of depressor arm is generally much larger than the 3mm left and right diameter of pipe, and to be arranged on ball/cylinder on Linear-moving parts be also much larger than the distance of 3mm pipe diameter, to contact groove side with pivotal axis.Opening angle between depressor arm and piping support device, represents with θ, many times by being, is being less than the angle of 15 °.

Visually and be intuitively clear that, when ball/cylinder moves, the major part of its action will be converted into the rotation of depressor arm.In order to characterize quantitatively this point, see Fig. 3 .5.1-2, suppose ball current promoting forward depressor (due to its forward and backward, different from the contact point of groove) and contact point and pivotal axis between distance be R, and the movement of ball can be broken down into the component of two quadratures:

1.dy=forward moves

2.dx=moves from axle

And dy can be broken down into two orthogonal directions of another group subsequently:

(1) direction of rotation

(2) radial direction

Component in the radial direction will can not cause any effect.Let us is only considered direction of rotation.From Fig. 3 .5.1-2:

${\mathrm{dy}}_{\mathrm{rotation}}=\mathrm{dy}\cos \mathrm{\&theta;},\frac{{\mathrm{dy}}_{\mathrm{rotation}}}{\mathrm{dy}}=\cos \mathrm{\&theta;}$

${\mathrm{dx}}_{\mathrm{rotation}}=-\mathrm{dx}\sin \mathrm{\&theta;},\frac{{\mathrm{dx}}_{\mathrm{rotation}}}{\mathrm{dx}}=\sin \mathrm{\&theta;}$

a fact only very little part of telling that we move from axle will participate in rotary moving.Remember, non-linear relation means the speed controlling that actual key just occurs in the final stage of step, thereby can be even less corresponding to the θ value of these committed steps.If θ=10 °, |-sin θ |=| sin10 ° |=0.1736, that from axle, moves reaches 82% by radial direction " absorption ".

θ is less, and this " absorption " is better.When depressor and piping support device approach and when pipe is almost cut off completely, remaining, thus nearly all being absorbed from axial displacement, and we obtain the situation shown in Fig. 3 .3-1 at all never.

Except solving from the problem of axial displacement, the arm of pivotable has also strengthened precision and has reached a ratio, and this ratio is that ball/cylindrical contact point is to the distance D of pivotal axis and the ratio between the distance d between depressor arm and pipe and the contact point of pivotal axis.By lever principle, this also increase power reaches D/d.Ask for an interview Fig. 3 .5.1-3, wherein use similar triangle property that this relation is shown.

Please remember that we mention 0.0254mm(0.001 inch for linear stepping motor in [3.3 nodel line sexual acts guiding]) be common stroking distance from, and be roughly IV pipe 3mm internal diameter 1/118.By using " utcracker " depressor of this pivotable, these 118 parts are cut apart and can be further by the ratio of D/d, be segmented, and this obtains the even meticulousr precision aspect drop rate controlled.

We have gathered three key advantages using pivot structure:

1. absorbed linear actuators from axial displacement.

2. amplified pressing force.

3. strengthened precision.

And we mention possible modification: be not necessarily to use linear actuators push/pull depressor; Can push/pull depressor arm or piping support device (when it allows when mobile) create and relatively move; Also can there is more than one such linear actuators to drive both.

And we provide the formal characteristic of pivotable " utcracker " structure:

1. depressor is connected at one end with piping support device, and this allows relatively in rotary moving between two parts.

2. one or more linear actuatorss (a plurality of) push away or draw depressor and/or piping support device to open or close angle between the two.

Adapter between line movement parts and arm groove generally has ball or cylinder form.So-called cylindrical shape, also allows bearing.Its intrinsic propesties is, must at least one of numerous cross-sections surfaces, have rounded periphery, and this rounded periphery can allow it in groove, smoothly to move.The possible additional difficulty of take is cost, and other modification are also possible.

Also note that linearly moving parts can be at any position of any geometric configuration contact depressor, needn't from pivotal axis, start one half line/ray by the contact point with depressor or supporter near.In example shown in Fig. 3 .5.1-4.

" utcracker " of the pivotable of 3.5.2 rotating

" utcracker " of pivotable might not need linear actuators.Fig. 3 .5.2-1 illustrates a modification, by the axle of electric rotating machine being connected to depressor or supporter so that the rotation of motor will cause the variation of angle between depressor arm and supporter.In this configuration, do not need ball/cylindrical shape and groove.

Certainly, also may depressor and supporter by rotary components, drive, or or even in hybrid combining, the rotation of one of them parts is driven by Linear-moving parts, and the rotation of another parts is to be applied by rotary components.Essence result (and definition therefore) is by the relative rotation that is all the time two parts.

Rotation does not need directly by motor, to be driven.Preferably, between motor and depressor/supporter, may there are other mechanism, by this mechanism, carry out transfer line sexual act.For example, such mechanism can be used gear (a plurality of), and this has also strengthened precision and has increased power.

For the parts that apply rotation to depressor or supporter, whether there is fixedly rotating shaft and also do not require, this means and itself also can be allowed to mobile reaching to a certain degree.The characteristic that plays restriction effect is only to another (depressor/supporter), to apply spinning movement from parts.

And we provide the formal characteristic of pivotable " utcracker " structure:

1. depressor is connected at one end with piping support device, and this allows relatively in rotary moving between two parts.

2. directly or via other intermediate mechanisies, be connected to the revolving actuator (a plurality of) of depressor and/or piping support device, it rotates (a plurality of) and causes opening or closed angle between the two.

Can pipe slide?

A problem is when depressor is pressed or discharge, manage position and slide? as long as the surface of depressor and supporter is very not sliding, this can not occur.In reality realizes, the fixture of pipe should be there is in the position that approaches depressor/supporter, with the position of holding tube.

3.6 cam embodiment

Also can spinning movement be converted to linearity with cam.There is polytype cam and shown in us, be spiral in this illustrative embodiment.How the rotation that cam is shown in five positions shown in Fig. 3 .6-1 will drive the line movement of depressor.

In each subimage of Fig. 3 .6-1, the center circle in front view is motor shaft.Plate is connected to axle and cutting on this plate.The geometry of groove is the envelope (envelope) of circle or any shape at least one of its numerous cross sections with rounded periphery, along spiral curve, around its center, moves.If groove rotates clockwise, by agreeing with the cylinder shape connector in groove, depressor will be pressed to the right, and if groove counterclockwise rotate, will be pulled to a left side.

Why does is we must use the envelope of circle or have at least one cross section the shape of rounded periphery as the same problem having proposed in [" utcracker " of 3.5.1 joint pivotable] is? if we study depressor or the bar (parts that connect depressor and cam with the center of rotation of cam as zero.And need not to be bar, but convenient in order to cry herein) relative action, we find that bar reality has and moves radially in cam center rotation.This is for not taking circle or being for the parts of rounded periphery (thereby allow connected something Smooth Rotation and do not repeat collide) to be impossible on its at least one cross section.Such link must be circle or at least one cross section, there is rounded periphery, therefore and its envelope is the shape that groove need to have.

For for purpose of brevity, we use spiral of Archimedes as explanation, in practice, can use the cam of any shape, as long as can realize the control of expectation hereinafter.

If we suppose to use the geometry of spiral of Archimedes, because the polar equation of helix is:

r=c+αθ

C is constant always.If we want one completely rotation make depressor move the distance (this is the common internal diameter of IV pipe) of 3mm,

3mm=△r=α△θ=α·2π

∴α=3mm/2π

We can calculate the parameter of helix in this way.For ease of manufacturing the two ends of grooving, can not contact each other, we also can preferably select to be less than the rotation of whole circle.

Rotation and be linear from the relation between the line movement conversion of spiral cam.For the non-linear relation between containing pipe thickness and dropping speed better, we can and calculate the cam that designs other shapes certainly based on experiment.

Thereby the stable electric current that can need that cam is not self-locking maintains its position.In order to pin cam in the situation that there is no continuous current, in following steps, can:

1. use the incompatible amplification rotation of gear train so that little rotation can cause the larger rotation of plate.The gap that plate has or hole are dispersed in different directions equably or unevenly.

2. with electric magnet, come lifting to be connected to the wisp of other assemblies (a plurality of) of spring or plastics.When electric magnet leaves, spring pushes wisp in the hole or gap on plate, therefore locks its position; When electric magnet adsorbs, it plays wisp by lifting, and this plate, and then this motor shaft and cam, will be allowed to again move.

The same electrical Magnet that use has rubber-spring assembly is also possible.Except the hole on plate, the friction of rubber also can prevent cam and axle rotation.

Although also note that for good control, in the example of Fig. 3 .6-1, we have used cam in conjunction with bearing, to guide the line movement of depressor, and the cam structure that also can be used alone is so that edge directly contacts IV pipe.For the example of spiral of Archimedes, the variation of θ causes that therefore the variation of radical length can be used to direct pressing tube.This is also possible realization, although the item such as the slip of pipe need to be solved suitably.

Also have, on the contrary due to the essence of cam be for spinning movement is converted to linearity or, directly do not drive depressor, also can in other parts of system, use cam.Spinning movement is converted to linearity to any part that cam is used in system or line movement is converted to rotation, and this is to accomplish by its essence.

4. illumination

43.1 good and poor illuminations

In this section, how we should throw light on for IV monitoring system if describing.Owing to monitoring dropping speed by the cyclical signal extracting such as drop height or size, therefore crucial is that the image in observed region must be clear.

Fig. 4 .1 illustrates exemplary good illumination.In this example, LED light is injected by the light guide/stopper shown in Fig. 4 .10-1 from top of chamber.Overhead illumination does not create the false brighter hot spot from present drip chamber surface reflection in image.

Fig. 4 .1-2 illustrates three examples of poor illumination, and wherein LED light makes video/image process comparatively difficulty in the lip-deep reflection of present drip chamber.In these images, LED is placed on side, thereby some drip and drop are although be centered around in the brighter reflection in chamber, still invisible.If light from chamber above or inject below, due to trimmed book body (from below) or its reflection (light above), some drip and drop can be covered by brighter region (a plurality of) completely.

Fig. 4 .1-3 has explained the reason in bright pip/region.Distance table between ideal point light source and present drip chamber is shown to D, and the lengths table on present drip chamber surface is shown dx,

$\begin{array}{c}x=D\tan \mathrm{\&theta;}\&DoubleRightArrow;=\mathrm{dx}=D{\sec }^2\mathrm{\&theta;d\&theta;}\\ \frac{\mathrm{d\&theta;}}{\mathrm{dx}}=\frac{{\cos }^2\mathrm{\&theta;}}{D}\end{array}$

And because being distributed evenly at, the power of the light source on certain distance take this distance on the circular surfaces of radius, in the luminous power at dx place be inversely proportional to the square distance of point source.

$P_{\mathrm{light}}\&Proportional;\frac{1}{{\left(D\sec \mathrm{\&theta;}\right)}^2}=\frac{{\cos }^2\mathrm{\&theta;}}{D^2}$

Follow the average light power of the light on dx with proportional as follows

$\frac{\mathrm{d\&theta;}}{\mathrm{dx}}\&CenterDot;P_{\mathrm{light}}=\frac{{\cos }^2\mathrm{\&theta;}}{D}\&CenterDot;\frac{{\cos }^2\mathrm{\&theta;}}{D}\&Proportional;{\cos }^4\mathrm{\&theta;}---(4.1-1)$

This illustrates how to have formed brighter pip/region.

Based on two reasons, this reflection interference drop detection:

1. most of light is reflected back before can contacting drop and being reflected by drop.

These reflection regions (a plurality of) of causing general recently brighter from the reflection of drop or with drop crossover.

The principle that 4.2 reflection/luminance contrasts reduce

Its principle of eliminating/reducing is directly led in the reason analysis of brighter present drip chamber surface reflection hot spot:

1. increase the distance between light source and present drip chamber.This is shown in Fig. 4 .2-1.For given x, larger D → less θ → larger cos θ → larger cos ⁴θ, so the contrast of the brightness between diverse location will become less.

2. in the whole structure of a plurality of light/point sources, eliminate this effect.Referring to Fig. 4 .2-2, if a single point light source S ₁to a P ₁and P ₂penetrate zones of different average light power, if another point source S ₂be placed as around passing through mid point parallel lines and S ₁symmetry, because from S ₁and S ₂the difference of each generation be eliminated, due to S ₁and S ₂contribution, at P ₁and P ₂the zone leveling luminous power at place is by identical.Have more multi-point source, they will be more around present drip chamber scattering ground, and we will obtain better and eliminate.

Another mode is not attempted reducing reflection/luminance contrast; It moves to another position by them simply so that it does not enter in the visual field of image-capturing apparatus.

3. from a direction, penetrate light, so that less reflection/luminance contrast will be seen from the angle of the system of checking.

The method that great majority reduce reflection can be divided into above-mentioned three classes, and we will be referring to these two principles at that time when method uses.

First we define the light source of two types:

Main light source: a kind of light source, wherein other forms of physical quantity is converted into light.The light source that comprises LED lamp, electric filament lamp, infrared lamp, uviol lamp, laser instrument and any other type.

Secondary light source: a kind of light source, its light is guided from one or more main light sources by optics.Use the illumination of optical fiber, light pipe (tube)/light tunnel (pipe)/integrated, direct reflection assembly, reflecting surface all to belong to this type.

Be clear that, although secondary light source provides more motilities, these two types are all suitable for our application equally.And when we use word " light source " hereinafter, unless by limiting on the contrary, it refers to these two types.

More than 4.3 light source

As shown in Fig. 4 .3-1, by a plurality of positions of (might not be completely around) around present drip chamber, place a plurality of lamps, can alleviate or eliminate reflections affect.Under extreme and ideal situation, there is a plurality of little point source around present drip chamber, " point source effect " will not exist.

Significantly, this method relates to second principle " elimination " that reduces reflection/luminance contrast.

The light source of this type can be main light source type or secondary light source type herein.Light source labelling in Fig. 4 .3-1 appears to and has implied main light source type, and this is only in order to illustrate preferably when comparing with Fig. 4 .4-1.

About each of a plurality of light sources, how far should be placed, also not require.In fact, as we in second principle of " elimination " and in Fig. 4 .2-2 institute inferentially, even if secondary light source is the same near apart from present drip chamber with original light source, the major part that also can effectively eliminate in brightness is inhomogeneous.Therefore, can use completely close to original single light source position, centered by original single light source position or near the array of source of original single light source position, eliminate the inhomogeneous of each.Such array of source also can be manufactured to the other types that are integrated in the LED in encapsulation or comprise light-emitting device array.Also should be thought of as the situation of a plurality of light sources.

4.4 multiple light courcess from secondary light source

Fig. 4 .4-2 illustrates the principle of light pipe.Each light source (whatsoever type) in fact comprises a plurality of point sources.Each point source from lens (or is directly itself; Lens make illumination brighter, but must be included in product) image, will penetrate the ray of advancing by the reflection of varying number before leaving.Once leave light pipe, effect looks like numerous rays from numerous virtual images of point source, thereby similarly is no longer to work as point source, but in effect, is similar to scattering/diffusive light source.Please refer to [Smith, the Practical Optical System Layout and Use of Stock Lenses(practicalness optical arrangement of Warren J. and use stoke lens), the 105th page].

The principle of light pipe itself is unique.Although be similar to second " elimination " principle that reflection/luminance contrast reduces, be more suitable for it to be classified as an independent class rather than to be categorized as elimination principle.

Note, we use " light pipe " as the term of following blanket property

1. light pipe, light tunnel, integrated

2. optical fiber

3. above-mentioned boundling

Fig. 4 .4-1 illustrates and how to use single source (any type) to create a plurality of secondary light sources via light pipe.For the light pipe of the thinner type such as optical fiber, can use together its boundling.

When using light pipe in this way, second " elimination " principle that they have obviously used reflection/luminance contrast to reduce.

4.5 light sources from direct reflection

Fig. 4 .5-1 also illustrates can carry out direct light with mirror assembly, thereby can from single source, create a plurality of light sources.In the situation that not creating a plurality of light source, certainly can be only for guiding single source.

If by mirror, created a plurality of images from single source at the diverse location around present drip chamber, it has used second " elimination " principle that reflection/luminance contrast reduces.

If by mirror, be used to throw light on before present drip chamber in original light source, comparing original light source itself more away from the image of present drip chamber place formation light source, used first " increase distance " principle of reflections/luminance contrasts minimizing of 4.2 joints.

Can use the mirror of any shape, because it is only used to direct light, rather than image.Shape can be arbitrary surface or take geometry in particular.And also it doesn't matter: the mirror of shape can certain " unusual " effect of no generation when being used in common occasion like this.Can use the mirror of any type, as long as its bootable light.

The light source of 4.6 amplifications from lens

Another theory is that " amplification " original light source reduces or eliminates reflection/luminance contrast.In Fig. 4 .1-3 and Fig. 4 .2-2, if the size of the large I of light source and institute's illuminated objects (present drip chamber) is compared, effect can be similar to and exist a lot of small light sources to drip chamber from diverse location lighting point, thereby in polymerization effect, can eliminate the most non-uniform illumination of each.

This is illustrated in Fig. 4 .6-1.Suppose we use compared with thin lens and due to

$\frac{1}{S_{\mathrm{object}}}+\frac{1}{S_{\mathrm{image}}}=\frac{1}{f}$

Work as f<S _objectduring <2f, the real image of putting upside down and amplifying will be formed on lens opposite side.

Second " elimination " principle that this has obviously used reflection/luminance contrast to reduce.

Work as S _objectduring <f, form the virtual image of forward and amplification on will identical at lens (identical with light source) side.

Second " elimination " principle and first " increase distance " principle that this has used reflection/luminance contrast to reduce, because the image of light source is compared original graph image source now more away from present drip chamber.

In both cases, we have obtained the light source amplifying, and by this measure, can reduce present drip chamber surface reflection.

Work as S _objectduring=f, will cause the situation of the light of parallel outgoing not to be listed, because true light source has specific size.Work as S _objectduring close to f, the point on it can be in these three kinds of situations arbitrarily.Yet whole structure will be to have reduced present drip chamber surface reflection all the time.

There are a lot of modes that create lens.For simple lens, each surface can be projection, recessed or smooth; For thick lens, the ability of convergence light also depends on its thickness; For lens combination, probability can not be enumerated.

But we have the well-formedness of the lens of the application that universal performance represents us really: focal length, or the effective focal length in lens combination or thick lens (EFL).

Provable, the optical system (concept of upper summary single lens) that only needs to have positive focal length length or EFL can create enlarged image or image is appeared at compares the original observer farther place being positioned at opposite side.Therefore we reach a conclusion: for our application, can create by any optical system with positive EFL the enlarged image of light source, described image is compared original light source or further from present drip chamber or more approaching.

4.7 use reflecting surface

We also can eliminate chamber surface reflection effect with reflecting surface.The method can be regarded as the upper summary of [4.5 joints are from the light source of direct reflection], the non-flat forms because surface can be bending.

Given shape has our available useful geometric properties, sees Fig. 4 .7-1:

1. ellipse or ellipsoid: the optical fiber sending from a focus will all be reflected onto another focus.

Effectiveness: ellipsoid have smear (smear out) light source image a large amount of " comet " thus (coma) and the mirror of a lot of small sizes final image of seeing from different perspectives light source light source by fuzzy [the practicalness optical arrangement of Warren J.Smith and use stoke lens, the 105th page].

Principle: although there are some similarities with second principle " elimination " that reduces reflection/luminance contrast, be preferably it is divided into its oneself class.

2. parabola or parabola: the light sending from focus will be reflected so that all light becomes parallel with axis (axis of symmetry of this shape itself).

Effectiveness: the explanation of Fig. 4 .1-3 is directly invalid, because light appears to the parallel lines from infinity now.

Principle: be preferably it is divided into its oneself class.

3. hyperboloid or hyperbola: the reverse extending line of the light sending from a focus is converged in another focus, thus appear to light from the virtual image that is positioned at another place, source.

Effectiveness: the virtual image of light source will this means that the D in Fig. 4 .1-3 becomes larger than light source itself further from present drip chamber, therefore for being positioned at the lip-deep ad-hoc location x of present drip chamber, it is less that θ will become.Remember, in formula (4.1-1), for average emitted luminous power and the cos of specific dx ⁴θ is proportional.The increase of D has reduced θ, so cos θ and cos ⁴θ increases, and therefore will reduce luminance contrast.If the distance D between the virtual image of light source and present drip chamber is enough large so that in present drip chamber, have a cos θ enough large (such as, similarly be 0.9), 4 of cos θ powers will not be very little numerals, thus luminance contrast will be not strong.

Principle: first principle that reduces reflection/luminance contrast " increases distance ".

Note, by using " ellipsoid ", " parabola " and " hyperboloid ", 3D shape that I refer to, rather than two-dimensional curve.For 3D shape, it is by around axisymmetry; For 2D curve, must make curve move to create along space curve the surface that it skims over.Ask for an interview Fig. 4 .7-2.

Also please note in Fig. 4 .7-1 and Fig. 4 .7-2, with photoresistance lug-latch, stop the directapath between light source and present drip chamber.This is necessary element, not so will have reflection/luminance contrast.

The defined property that the reflection of this classification/luminance contrast reduces is:

1. photoresistance lug-latch (part that can be made into/be integrated into light source prevent light source by light scattering to all directions or separated from light source), the directapath between impedance light source and present drip chamber.

2. reflecting surface, its reflection carrys out non-directly illuminated objects for light source.

As long as meet above-mentioned two characteristics, whether surface is very level and smooth, level and smooth or coarse, not essential distinction, because inherently, we for reflecting surface itself without any particular requirement and for obtaining its image (fuzzy or clearly) and loseing interest in, thereby and just with this surface, rebooting light can reduce reflection/luminance contrast.

In [4.9 joints are in conjunction with reflection and rough surface definition], we combine definition, thereby have defined reflecting surface for all other roughness of level.

4.8 use rough surface

We also can reduce or eliminate reflection or luminance contrast with rough surface.So-called " coarse ", I am intended to refer to have from the teeth outwards a lot of inhomogeneous places, thus light or thin-beam will be scattered to all directions of a single point vicinity.Yet, due to these adjectival subjective essence, what is difficult to define quantitatively for " smoothly " or " coarse ".Herein, I define:

Definition:

1. in our term, " smoothly " is equal to " reflection "

2. by the smoothness that only has two types: reflection or coarse.We are any surface or the part that shows, are categorized as a class of two apoplexy due to endogenous wind and a class only.

Rough surface is shown in Fig. 4 .-81, and it has also required photoresistance lug-latch to prevent light direct lighting present drip chamber.

The defined property that the reflection of this classification/luminance contrast reduces is:

1. photoresistance lug-latch (part that can be made into/be integrated into light source prevent light source by light scattering to all directions or separated from light source), the directapath between impedance light source and present drip chamber.

2. rough surface, its reflection carrys out non-directly illuminated objects for light source.

By referring to Fig. 1 .1-1A, can confirm that rough surface is used for reducing the effectiveness of reflection/luminance contrast, this takes in indoor environment, and wall is rough surface, and the fluorescent lamp that is blocked illumination IV is directly set.On present drip chamber both sides, only have weak reflection, this does not disturb the top intermediate image of drop (a plurality of).

4.9 in conjunction with reflection and rough surface definition

By as the cohesive process of giving a definition, can solve the difficulty of two sub reflectors and rough surface:

1. due to for rough surface, shape is indifferent, so desirable any shape (comprising the shape that reflecting surface is used), for example, and conic section shape.

2. by combination two subjective words " reflection " and " coarse ", we can define " surface of any rank smoothness/roughness " simply.

3. therefore, we reach a conclusion " surface of any rank smoothness/roughness ", suppose suitable or geometry arbitrarily, thereby, when original light source is blocked direct lighting present drip chamber, it is for the reflection of original light source, the present drip chamber of throwing light on, so that there is no or only have faint reflection/luminance contrast, can be used to our application.

4.10 avoid photographing reflection/luminance contrast

Our disclosed last method does not also require above-mentioned any optical component.This method has been used the 3rd principle of 4.2 joints " from a direction, to penetrate light, so that will see less reflection/luminance contrast from the angle of the system of checking ".

Fig. 4 .10-1 illustrates us and can use the light guide/stopper extending from the covering part of present drip chamber to cover the part of light source (being positioned at present drip chamber top or bottom).Thereby light is by directed when video camera (or any lens of optical system, referring to figure 0.1-1) is positioned at the horizontal side of present drip chamber, and light enters present drip chamber from top/bottom.

Under this configuration, be incident on that light on present drip chamber top surface will partly be reflected and residual ray will enter present drip chamber.A part for the light being come also can touch the surface that present drip chamber is not covered by photoresistance lug-latch, but the less angle of incidence that can form with present drip chamber surface due to these limited light amounts and they does not create brighter hot spot.On the other hand, drop (a plurality of) is generally assumed to be spherical or elongated circle, thereby can effectively incident illumination be reflexed to other directions, for this reason, can in the visual field of video camera, become brighter hot spot (a plurality of).

Use the guiding of guide/stopper from the light source at top the effect of obtaining from the side video, to be illustrated in 4.1-1 that (these images are used and obtain via the emitted LED light of the light guide/stopper that is positioned at top.Second segment referring to the 4th joint illumination).The good quality of image is the prerequisite of accurately monitoring.

Fig. 4 .10-2 illustrates modification, wherein the guiding of light guide/stopper is from the light of inclination or horizontal direction, and therefore photographing unit can be placed in a position and have orientation so that it checks present drip chamber from a visual angle, from this visual angle, can see the present drip chamber reflection of small amount, and/or the image of drop (a plurality of) is clearer.For photographing unit and light guide/stopper, whether should be positioned at the also not restriction of present drip chamber homonymy.The characteristic that plays restriction effect is " from a direction, penetrating light, so that will see less reflection/luminance contrast from the angle of the system of checking " all the time.

In Fig. 4 .10-3, we illustrate light guide/stopper also needn't extend to the Lights section (or near locating) from all directions of the part of present drip chamber (or near).Light guide/stopper can be placed in to both any sides, or in the situation that not linking together, more than one light guide/stopper can be placed near light source/around light source and be placed near present drip chamber/around present drip chamber.Light guide (a plurality of)/stopper (a plurality of) also can be integrated:

1. as fixture, the chamber of present drip chamber or maintain the part of part.

2. use this light, thereby effectively work as torch, wherein emergent ray is directed.

As us shown in Fig. 4 .10-1 and Fig. 4 .10-3, for the light guide/stopper not being connected in the whole path that does not have covering from light source to present drip chamber and for { the fixture with present drip chamber, chamber, maintain part } or the integrated light guide/stopper of light source, for only no, from top, penetrate and also do not limit requirement.As long as the image that photographing unit can be avoided seeing reflection, be clear that monitored area simultaneously, this configuration is just effective all the time.

General introduction

In image-capturing apparatus, how can reduce or avoid the method for reflection/luminance contrast to be summarised in above table for reference.

Claims (19)

1. Attention: this PCT application requirement my priority date of two U. S. applications:

   The electro-mechanical system that U. S. application 13019698:IV controls

   U. S. application 13356632: IV monitoring is controlled and illumination with image processing, frequence estimation, the machinery of control system automatically

   The major part of the application's content is from U. S. application 13356632, and the some parts that the 3rd joint machinery is controlled is from U. S. application 13019698.The Histogram Matching of figure image intensifying is new introducing in this PCT application.In following claim, if a part for claim or this claim does not specialize its desired priority date, what it required is the priority date of U. S. application 13356632; For requiring any claim of priority date of U. S. application 13019698 or a part for claim, will in bracket, specialize; For its respective description, be any claim of new introducing in this PCT application or a part for claim, also will clear and definite labelling.

   My requirement

   1. an equipment, is used any video/image technology (a plurality of) to come from IV observation process extracting cycle signal, and measures dropping speed by any frequence estimation technology.
2. 2. equipment as claimed in claim 1, is characterized in that, described equipment strengthens described image by any image enhancement technique, described technology include but not limited to any order, any combination, use in the following method of any number of times arbitrarily:

   (1) gradation conversion, includes but not limited to:

   A. power law conversion

   B. exponential transform

   C. stagewise linear transformation and question blank

   D. use the technology of histogram information, comprise Histogram Matching/appointment or equilibrium

   [in this PCT application, newly introducing]

   And the additive method of greyscale transformation.

   (2) frequency domain technique, includes but not limited to:

   A. with the frequency domain filter of spatial domain wave filter equivalence;

   B. the wave filter directly designing in frequency;

   (3) wavelet method of figure image intensifying.
3. 3. equipment as claimed in claim 1, is characterized in that, described equipment is binary picture by threshold process method by greyscale image transitions, include but not limited to any order, any combination, use in the following method of any number of times arbitrarily:

   (1) alternative manner

   (2) value of choosing arbitrarily of threshold process

   (3) manual definite value of threshold process

   (4) the area pixel value of threshold process is average

   (5) intermediate value of the area pixel value of threshold process

   (6) additive method of threshold process.
4. 4. equipment as claimed in claim 1, it is characterized in that, described equipment frequency of utilization estimating techniques determine dropping speed according to processing picked up signal via image, described technology include but not limited to any order, any combination, use in the following method of any number of times arbitrarily:
5. 5. equipment as claimed in claim 4, is characterized in that, described equipment is used nonparametric technique for frequence estimation, described technology include but not limited to any order, any combination, use in the following method of any number of times arbitrarily:

   (1) simple time domain approach, includes but not limited to:

   A. find out and count value is crossed the border, value threshold process, zero passage or zero value detection

   B. find out and count local maximum/minima.

   (2) Time-domain Statistics method, includes but not limited to:

   A. frequence estimation have partially or without partial autocorrelation

   B. frequence estimation has partially or the inclined to one side auto-covariance of nothing

   C. have partially or without inclined to one side average amplitude difference function (AMDF)

   (3) method of Fourier or Fourier correlation, includes but not limited to:

   A. periodogram

   The periodogram of b.Bartlett is average

   C. discrete time Fourier transform (DTFT)

   D. autocorrelative correlogram or periodogram

   E. the DTFT of auto-covariance or periodogram

   F. discrete cosine transform (DCT)

   G. discrete sine transform (DST)

   (4) wavelet method.
6. 6. equipment as claimed in claim 4, is characterized in that, described equipment operation parameter method is for frequence estimation, described frequence estimation include but not limited to any order, any combination, use in the following method of any number of times arbitrarily:

   (1) autoregression or autoregression average-average frequency spectrum estimation, this includes but not limited to

   A.Yule-Walker method

   (2) characteristic vector/subspace method or can, from any method of the pseudo-frequency spectrum estimation frequency of signal, include but not limited to:

   A.Pisarenko Harmonic Decomposition method

   B. multiple emitter location and signal parameter estimation (MUSIC).
7. 7. a machinery, changes the thickness of IV pipe or the speed that diameter is controlled IV drop by the IV dropping speed recording according to the monitoring equipment of processing based on video/image.
8. 8. device as claimed in claim 7, is characterized in that, described device combines to press or discharges described pipe with pipe depressor and the supporter of any shape and material, so that the speed of described pipe is controlled.
9. 9. device as claimed in claim 7, it is characterized in that, described device in any part, with arbitrary number of times, use the difference driving screw combination of driving screw or any type, with other assembly combination in any or by the difference driving screw combination itself of described driving screw or any type, for by the linearity that is converted in rotary moving.

   [this claim has required the priority date of U. S. application 13019698]
10. 10. device as claimed in claim 7, it is characterized in that, described device is used the assembly (a plurality of) of single type or the combination of dissimilar assembly (a plurality of) to guide the action of one or more Linear-moving parts, its object can comprise prevent, reduce or control described Linear-moving parts from axle action, and can include but not limited to as lower member:

    (1) key/keyway combination,

    (2) tooth bar/groove combination,

    (3) be positioned at described linear unit inside or outside or other local bearings.
11. 11. devices as claimed in claim 7, is characterized in that, described device is used lever in any part, use arbitrary number of times, with other assembly combination in any or by himself, object can include but not limited to:

    (1) strengthen mobile precision

    (2) amplification power

    (3) action of parts (a plurality of) is converted to another parts (a plurality ofs') action.

    [this claim has required the priority date of U. S. application 13019698]
12. 12. devices as claimed in claim 7, is characterized in that, described device use relative to any reference point, have absolute or relative to the rotatable parts that move or a plurality of parts, with other actions side by side or side by side non-, press or discharge described IV pipe.
13. 13. devices as claimed in claim 12, and be characterised in that, described device has

    (1) hub switch side, around one or parts rotatable;

    (2) open area that the part of described IV pipe can be passed, and due to one or more moving-members (a plurality of) in sub-claim and stationary parts (a plurality of) relatively move or moving-member between the variation of scanning action this region of bringing that relatively moves and cause, cause compression or the release of described IV pipe.
14. 14. devices as claimed in claim 13, it is characterized in that, described device is on its movable part one or more, there is groove, otch, or opening, it can have or not have even width, and described groove (a plurality of), otch (a plurality of), or opening (a plurality of) can be with linearly moving parts or a plurality of parts being connected Anywhere or not connecting in geometric configuration, thereby the rotation of described linear unit or a plurality of parts can be exchanged into the rotation of rotary part or a plurality of parts (as defined in claim 13), and described device can have at the join domain place of described rotation and linear unit as lower one or more:

    (1) be connected with described linearly moving parts and agree with ball in described groove, otch or opening or any assembly of ball shape;

    (2) be connected with described linearly moving parts and agree with cylinder in described groove, otch or opening or any assembly of cylindrical shape;

    (3) one or more bearings;

    (4) in its a plurality of cross sections, there is any assembly that at least one is taked circle or has circular periphery.
15. 15. devices as claimed in claim 13, it is characterized in that, the rotation of one or more parts is to be applied by another assembly also rotating, described another assembly can have or not have fixed axis and can have simultaneously or not have another and moves, and described device can or select one or both get both:

    (1) use gear (a plurality of) to apply rotation to described pivotable member (a plurality of);

    (2) with electric rotating machine, to described pivotable member (a plurality of), apply rotation.
16. 16. devices as claimed in claim 7, is characterized in that, on the contrary in one or more parts of described system with cam or a plurality of cam by the line movement of other assemblies be converted to rotation or or with the edge of described cam, directly press described IV and manage, and

    (1) the mobile bearing of described cam (being connected with Linear-moving parts) can be taked spiral-shapedly in some or a plurality of parts, includes but not limited to spiral of Archimedes;

    (2) described cam can have groove or otch or opening, described Linear-moving parts are connected by it with assembly, and described coupling assembling can take to have in a plurality of cross sections at least one for the shape of circle or circular periphery, and in this situation, described groove can, but nonessential, take the shape of the envelope that described coupling assembling moves along specific curves.

    [this claim has required the priority date of U. S. application 13019698]
17. 17. 1 kinds of illuminators, illumination present drip chamber, so that can obtain image clearly to the IV monitoring system of processing based on video/image; Described system can be used one or two in following principle:

    (1) with the combination of an optics or optics, tell on, so that light more sends away from the distance of described present drip chamber from comparing original light source;

    (2) with the combination of an optics or optics, tell on, to eliminate reflection (a plurality of) or the inhomogeneous brightness in described present drip chamber, because the light sending from the ideal point light source in original light source seems to send from comparing the more separated point source of their actual origin;

    Reduce or eliminate reflection (a plurality of)/luminance contrast in image.
18. 18. illuminators as claimed in claim 17, described illuminator using method, described method includes but not limited to as lower single, a plurality of or combination:

    (1) a plurality of light sources, or relative separation or close or separated, around or part around described present drip chamber not around or the packaged light source that comprises arbitrary source or comprise a plurality of light-emitting components;

    (2) a plurality of light sources, described a plurality of light source via single source one or more single-piece is integrated or encapsulation containing a plurality of light-emitting component luminaires, through light pipe (a plurality of), light tunnel (a plurality of), integrated (a plurality of) or optical fiber (a plurality of) or their boundling, guide;

    (3) mirror or combinations of mirrors, smooth, arbitrary curve or take geometry in particular, the light of original light source (a plurality of) is conducted through it or they;

    (4) lens or a plurality of lens, its surface can have any shape, thin or thick, or the lens combination of formation optical system (a plurality of), there is positive focal length length (for thin lens) or positive effective focal length (for thick lens or optical system), produce enlarged image or a plurality of image of described original light source, described image compare described original light source further from or more approach described present drip chamber;

    (5) photoresistance lug-latch, itself or a part that is made into/is integrated into described light source prevent that described light source from scattering light to all directions or separated with described light source, and the tool surface of any smoothness of described present drip chamber of throwing light on, thereby not there is not or only exist faint reflection/luminance contrast; The shape on described surface can include but not limited to:

    A. ellipse or ellipsoid;

    B. parabola or parabola;

    C. hyperboloid or hyperbola;

    D. or by above-mentioned sub-claim (a), (b) and (c), scan arbitrarily the shape that action forms.
19. 19. illuminators as claimed in claim 17, described illuminator and image-capturing apparatus are set together according to a configuration, and described configuration makes can see from the visual angle of described image-capturing apparatus less reflection/luminance contrast; Can comprise, single or multiple, following, but be not limited to this:

    (1) light guide/stopper extending between described light source and described present drip chamber, cover in described light source or present drip chamber parts or do not cover, thereby light can be from present drip chamber described in a directional lighting, and described direction can cause the less reflection/luminance contrast in image-capturing apparatus;

    (2) light guide/stopper does not extend between light source and described present drip chamber, and be only to extend beyond and/or cover light source or present drip chamber parts (a plurality of) one or both, thereby light can be from present drip chamber described in a directional lighting, and described direction causes the less reflection/luminance contrast in image-capturing apparatus;

    (3) light guide/stopper as described in sub-claim (1) and (2), it is integrated

    A. as fixture, the chamber of described present drip chamber or maintain the part of part;

    B. use this light, thereby it works as torch effectively, wherein emergent ray is directed.