CN102917171B - Based on the small target auto-orientation method of pixel - Google Patents

Abstract

The present invention relates to field of video monitoring, by the interlock of fixed camera and rotary camera, realize the small target auto-orientation linkage surveillance method based on pixel.Comprise target acquisition, target localization two steps, wherein target acquisition selects the panoramic video of fixed camera to trigger the startup of rotary camera by frame; Target localization is the complete HD video picture being obtained localizing objects by rotary camera, carry out image outline described point, visual profiles dot information is converted to sea chart coordinate, project on sea chart, projected position point on sea chart is by sea chart region drawing tool, be created as section object, by calculating the geometric center in this region, on sea chart, directly obtain the latitude and longitude information of this position, target location latitude and longitude information again by obtaining, the locking to this target arranging video camera is monitored, dynamic in real time to understand it.The present invention for rotary camera provides panoramic video picture, selects target accurately catch and locate by rotary camera to frame by fixed camera.

Description

Based on the small target auto-orientation method of pixel

Technical field

The present invention relates to field of video monitoring, by the interlock of fixed camera and rotary camera, realize the small target auto-orientation linkage surveillance method based on pixel.

Background technology

Along with developing rapidly of computer technology and image processing techniques, for improving safety monitoring ability to strengthen law enforcement dynamics, need to obtain clear and monitor video is as enforce foundation accurately, the video monitoring demand of user at different levels to intellectuality, accuracy, high Qinghua and removableization improves day by day.For a long time, video monitoring is mainly limited to the fixed area monitoring of fixed camera, and by the impact of pixel and intelligence degree, video monitoring has the problems such as the low and single picture display of region limitation, definition, brings difficulty to monitoring and law enforcement evidence obtaining.So need a kind of panorama, from various visual angles monitoring and the target localization that realize video monitor object with the monitoring mode of being association of activity and inertia.

Summary of the invention

The invention provides a kind of small target auto-orientation method based on pixel, realize the monitoring to different target in same picture and location.

For achieving the above object, the present invention takes following technical scheme:

Based on the small target auto-orientation method of pixel, comprise the following steps:

(1) target acquisition

1) videoscanning: fixed camera carries out scanning monitoring to the appointed area in field range;

2) view image calculates through background analysis, automatically triggers sound and light alarm when meeting default alarm rule;

3) according to alert locations information, alarming determining is carried out by the panoramic video of fixed camera;

4) select emphasis scene in video pictures and target by frame selection tool frame, frame selection operation triggers the startup of rotary camera;

5) obtain frame selected scenes scape geometric center location point coordinate by hind computation, and according to matching algorithm, calculate and obtain the position of this geometric center in rotary camera, thus obtain cloud platform rotation angle;

6) regulate rotation camera head orientation, zoom obtains the HD video that frame selects scene location;

(2) target localization

1) regulated by the video of rotary camera, obtain the complete HD video picture of localizing objects;

2) on HD video picture, carry out image outline described point, and by back-end data processing center, visual profiles dot information is converted to sea chart coordinate, projects on sea chart;

3) the projected position point on sea chart, by sea chart region drawing tool, being created as section object, by calculating the geometric center in this region, on sea chart, directly obtaining the latitude and longitude information of this position;

4) by sea chart referential function, event icons and explanation are added to this position, facilitate checking and sharing of user at different levels;

5) the target location latitude and longitude information by obtaining, the locking to this target arranging video camera is monitored, dynamic in real time to understand it;

6) incident report information is sent to relevant administrative department.

The present invention is that rotary camera is supplied to real-time, accurate and complete panoramic video picture by fixed camera, and selects the frame of highest priority in panoramic video picture, starts rotary camera realization to the accurate seizure of highest priority and location; Rotary camera, then by the positional information in the gain of parameter localized target such as camera pixel, focal length and visual field or region, is realized high definition to localized target in panoramic video picture by the rotation of rotary camera and Focussing, is amplified display.

Accompanying drawing explanation

Fig. 1 is FB(flow block) of the present invention;

The geometrical relationship figure that Fig. 2 is rotary camera, fixed camera and frame select target;

Fig. 3 is the schematic diagram of camera coordinate system;

Fig. 4 is the schematic diagram of image coordinate system;

Fig. 5 is the schematic diagram that pixel image catches.

Embodiment

Below in conjunction with drawings and Examples, content of the present invention is described further.

Realize capital equipment of the present invention and comprise fixed camera and rotary camera, they are all high-definition cameras, and they all can carry out the adjustment of night vision mode and visible mode.Fixed camera is the principal phase machine of video monitoring system system default, keep fixing azimuth and fixing field range, can not carry out any change of camera position and visual field, the video pictures of fixed camera is the panoramic video that can arrive, and also has following operating function:

(1) visual field maps: the visual field image of video camera is become corresponding geographic orientation and distance with the form of pixel correspondence, and determines the geographic orientation of every bit in image with this;

(2) target frame choosing: by selecting scape frame arbitrary region in visual field chosen and locate;

(3) instruction generates: select the positional information of scape frame to generate corresponding orientation instruction, produce the driving instruction of rotating video camera and can identify;

(4) command: send video camera (The Cloud Terrace and motor-driven lens) instruction fixed point to corresponding rotary camera.

Rotary camera realizes region and the target monitoring of 360 ° by the rotation of The Cloud Terrace, and it can carry out the self-defined adjustment of position and visual field, can by zoom to monitoring objective carry out high definition amplify display, be to fixed camera monitoring supplement and perfect.Also there is following operating function:

(1) local operation: carry out common rotation and Focussing operation by the The Cloud Terrace of rotary camera and camera lens;

(2) combined operation:

1) command reception: receive the instruction of issuing this video camera;

2) target localization: according to the size in region, calculates the best monitoring visual field of target area in conjunction with the resolution, focal length etc. of rotating video camera;

3) camera driver: actuated camera is to specific orientation and focal length.

Realize operating procedure and the function of supervisory control system of the present invention:

(1) panoramic video is obtained

By video monitoring system, start the panoramic video scanning that fixed camera is fixed region, obtain the panoramic video picture of fixed area, panoramic video picture, by monitoring window layout pattern, carries out the display of single window, multiwindow form.

(2) selected monitoring objective

In the panoramic video picture of fixed camera, to the picture zoom function of highest priority by fixed camera, carry out target investigation and examination, user is concerned about, the not good destination object of display effect carries out monitoring objective by frame selection tool and selectes on panoramic video picture, amplifies its detailed picture information of display by the cloud platform rotation of rotary camera and zoom.

(3) rotary camera is started

After the monitoring objective of frame selection tool selected local, drag frame favored area to panoramic video picture to start rotary camera, realize the video interlink rotating video camera and fixed camera, then discharge mouse, obtain the high definition amplified video window of rotary camera selected target.After obtaining HD video picture, according to actual user's demand, picture extraction can be carried out, the operations such as video record.

(4) rotary camera target localization

1) position that frame selects target in fixed camera is determined;

2) the three dimensional coordinate space system of fixed camera and rotary camera relative position is set up;

3) by fixed camera focal distance f, the angle of fixed camera and horizontal plane, calculation block selects the position vector of target in fixed camera;

4) in three-dimensional system of coordinate, the anglec of rotation of rotary camera relative to fixed camera is calculated, and the vector position of rotary camera in XOY coordinate surface;

5) based on the rotational angle and the position moving parameter that calculate gained, rotary platform, captured target.

Wherein the target sizes decision method of rotary camera is as follows

Rotary camera obtains after frame selects target, obtains by zoom the HD video picture that frame selects target, now by the focal distance f of fixed camera and optical center to subject distance D, the size of calculating acquisition monitoring objective, its computing formula is as follows:

H=h*D/f calculates the horizontal range of photographic subjects;

V=v*D/f calculates the vertical range of photographic subjects;

Wherein: D: the optical center of rotary camera is to subject distance (geometric center distance)

F: rotary camera focal length

H: the photosensitive target surface horizontal range of CCD of rotary camera

The photosensitive target surface horizontal range of CCD of v: rotary camera A

H: the horizontal range of subject

V: the vertical range of subject

As shown in Figure 1, the present invention realizes the small target auto-orientation method based on pixel by fixed camera and rotary camera video interlink, and to realize monitoring to different target in same picture and location, concrete computer disposal flow process is as follows:

(1) target acquisition

1) videoscanning: fixed camera carries out scanning monitoring to the appointed area in field range;

2) view image calculates through background analysis, automatically triggers sound and light alarm when meeting default alarm rule;

3) according to alert locations information, alarming determining is carried out by the panoramic video of fixed camera;

4) select emphasis scene in video pictures and target by frame selection tool frame, frame selection operation triggers the startup of rotary camera;

5) obtain frame selected scenes scape geometric center location point coordinate by hind computation, and according to matching algorithm, calculate and obtain the position of this geometric center in rotary camera, thus obtain cloud platform rotation angle;

6) regulate rotation camera head orientation, zoom obtains the HD video that frame selects scene location;

(2) target localization

1) regulated by the video of rotary camera, obtain the complete HD video picture of localizing objects;

2) on HD video picture, carry out image outline described point, and by back-end data processing center, visual profiles dot information is converted to sea chart coordinate, projects on sea chart;

3) the projected position point on sea chart, by sea chart region drawing tool, being created as section object, by calculating the geometric center in this region, on sea chart, directly obtaining the latitude and longitude information of this position;

4) by sea chart referential function, event icons and explanation are added to this position, facilitate checking and sharing of user at different levels;

5) the target location latitude and longitude information by obtaining, the locking to this target arranging video camera is monitored, dynamic in real time to understand it;

6) incident report information is sent to relevant administrative department.

Target location algorithm process is as follows:

In video frequency following system, to the seizure of target by frame selection tool, destination object is extracted from background, image is carried out real-time dividing processing, thus obtain a width bianry image.In order to the target area information comprised in detected image, need mark gray value to be the pixel of 1, then check the connectedness between neighborhood pixels, will there is the pixel of connected relation as a target area, extract relevant characteristic quantity.

Frame selects the seizure of target to calculate coupling by grey scale pixel value and obtains, the calculating of geometric center position coordinates is by the Coordinate Conversion between the camera coordinate system set up and image coordinate system, utilize the relation of camera pixel, resolution and focal length, to obtain the coordinate position of targets of interest.

Pixel coordinate calculates: width dimensions x resolution=width pixel;

Height dimension x resolution=height pixel.

As shown in Figure 2,

A: the position of fixed camera; B: the position of rotary camera; C: frame selects target geometric center;

A1: the angle in fixed camera and y direction is fixing known parameters;

B1: the angle in rotary camera and y direction is unknown parameter;

C1: the relative position angle of fixed camera and rotary camera is unknown parameter.

Suppose,

1) fixed camera and rotary camera have the height of identical Y-direction, i.e. AA1=BB1=L;

2) fixed camera and rotary camera keep certain positional distance, and namely AB is fixed value;

3) fixed camera and rotary camera select the centre of location of target C to be the geometric center that frame selects grid to frame, and geometric center is obtained by fixed camera auto-focusing, namely ensure that the imaging point of rotary camera and fixed camera overlaps.

Geometric algorithm is as follows:

Known parameters is as follows:

1) the angle a1 in fixed camera and y direction;

2) rotary camera has the height L of identical Y-direction, AA1=BB1=L;

d:AB=A1B1=d

In right-angled triangle AA1C: AC=AA1/cosa1;

A1C=AC*sina1=AA1*tga1；

In right-angled triangle BB1C: according to universal formula, can obtain the rotational angle of rotary camera The Cloud Terrace in YOZ plane is:

$\mathrm{BC}=\sqrt{d^2-{\left(\mathrm{ltga}1\right)}^2}$

$b1=\mathrm{arctg}\sqrt{d^2-{\left(\mathrm{ltga}1\right)}^2}/l;$

In triangle ABC, according to the cosine law: AB ²=AC ²+ BC ²-2*AC*BC*cosc1,

Known rotary camera relative to the angle that horizontally rotates of fixed camera is:

$d^2=l^2+d^2-{\left(\mathrm{ltga}1\right)}^2-2*l/\cos a1*\sqrt{d^2-{\left(\mathrm{ltga}1\right)}^2}*\cos c1$

That is: $c1=\arccos \frac{1^2-{\left(\mathrm{ltga}1\right)}^2}{2-1/\cos a1\sqrt{d^2-{\left(\mathrm{ltga}1\right)}^2}}$

Target location algorithm is as follows:

(1) as shown in Figure 3, camera coordinate system is set up;

(2) as shown in Figure 4, image coordinate system is set up: image coordinate is divided into pixel coordinate system and physical coordinates system.Wherein image coordinate system is in units of pixel, with the image upper left corner for initial point, with u, v direction for Two coordinate axle.Image physical coordinates system is with the focus O of optical axis with picture plane ₁for the rectangular coordinate system of initial point.

(3) as shown in Figure 5, algorithm examples is caught based on pixel image as follows:

1) the 1st scan line, starts a new target label S (1) in the position of pixel (1,3); When processing pixel (1,6), learn mark S (1).Terminate at pixel (1,5) point, now by target label write S (1) in NS;

2) the 2nd scan line, in the position of pixel (2,2), from algorithm, it belongs to target label S (1).3rd scan line also can similarly process;

3) a new target label S (2) is started from (4,6) pixel of the 4th scan line.The pixel of the 5th scan line belongs to target label S (1) and S (2) respectively.

4) the 6th scan line, from pixel (6,1) the target label S (3) that beginning one is new, when scanning pixel (6,2) time, according to algorithm, target label S (1) will be merged in S (3), simultaneously the significant notation of S (1) is put invalid, and equivalent indicia points to S (3); When scanning pixel (6,6), merge the content of S (2) in S (3); When scanning pixel (6,8), due to S (2) significant notation zero setting, and equivalent indicia points to S (3) itself, so do not do any action; When scanning pixel (6,11), content in NS is written to S (3), the content now in S (3) is the merging of S (1), S (2) and NS.

5) coexist the 6th scan line, in pixel (6,13), opens up fresh target mark S (4); At pixel (6,14), check (5,15) the target label S (2) corresponding to, its significance bit is zero, S (3) is pointed in equivalence position, so by the content of S (3) write S (4), it is invalid that S (3) is set to, equivalent indicia points to S (4), check all target labels simultaneously, find that original S (1) and S (2) equivalent indicia point to S (3), existing should be sensing S (4) by the equivalent indicia of S (1) and S (2).In pixel (6,19), any target label is not taked to merge behavior according to algorithm.

6) the 7th scan line, pixel (7,6), this point belongs to its target label belonging to upper left pixel point, check this target label S (3), find that equivalent indicia points to S (4), so this point belongs to target label S (4); In like manner, scanning element S (7,13) also belongs to target label S (4).

7) check target label, S (1), S (2), S (3) significant notation is 0, only has S (4) significant notation to be 1, therefrom can learn, only have a target label in this pattern.

(4) frame selects destination object geometric center to locate:

Calculate target label position coordinates, the marked pixels coordinate of source location is selected by frame, be averaged by the coordinate figure of the different picture elements belonging to same connected component, using the position of mean value as the center-of-mass coordinate of this frame favored area or several adjacent, overlapping region, its coordinate computing formula is:

$x_c=\underset{i,j\∈\mathrm{\Ω}}{\mathrm{\Σ}}\frac{j}{N},y_c=\underset{i,j\∈\mathrm{\Ω}}{\mathrm{\Σ}}\frac{i}{N}$

Wherein: the longitudinal and transverse coordinate that (i, j) is image pixel;

N is total number of picture elements of connected component;

Ω is the set of the pixel belonging to same connected component;

Implementation is as follows:

The gray value of each picture element of step 1 search bianry image line by line, performs step 2 and calls recursive function when gray value is 0.

The gray scale of picture element is set to 1 by step 2, the abscissa of picture element and abscissa and be added, ordinate and ordinate and be added.Respectively to 8 communication direction search from present picture element point, judge whether the gray value of these 8 picture elements is 0, if 0 performs step 2, otherwise call end and return main program.

Step 3 according to the horizontal stroke returned, ordinate and the mean value that calculates transverse and longitudinal coordinate be the center-of-mass coordinate of the image of current region position.Continuing the next gray value of search is the picture element of 0, so obtains the center-of-mass coordinate of all micropores.By the centroid calculation of all pixels of object box favored area, obtain the position coordinates of this regional center point.

Claims (1)

1., based on the small target auto-orientation method of pixel, it is characterized in that: comprise the following steps:

(1) target acquisition

1) videoscanning: fixed camera carries out scanning monitoring to the appointed area in field range;

2) view image calculates through background analysis, automatically triggers sound and light alarm when meeting default alarm rule;

3) according to alert locations information, alarming determining is carried out by the panoramic video of fixed camera;

4) select emphasis scene in video pictures and target by frame selection tool frame, frame selection operation triggers the startup of rotary camera;

5) obtain frame selected scenes scape geometric center location point coordinate by hind computation, and according to matching algorithm, calculate and obtain the position of this geometric center in rotary camera, thus obtain cloud platform rotation angle;

6) regulate rotation camera head orientation, zoom obtains the HD video that frame selects scene location;

Wherein, the emphasis scene in video pictures and target is selected also to regulate rotation camera head orientation to comprise by frame selection tool frame:

Determine the position that frame selects target in fixed camera;

Set up the three dimensional coordinate space system of fixed camera and rotary camera relative position;

By fixed camera focal distance f, the angle of fixed camera and horizontal plane, calculation block selects the position vector of target in fixed camera;

In three-dimensional system of coordinate, calculate the anglec of rotation of rotary camera relative to fixed camera, and the vector position of rotary camera in XOY coordinate surface;

Based on the rotational angle and the position moving parameter that calculate gained, rotary platform, captured target;

(2) target localization

1) regulated by the video of rotary camera, obtain the complete HD video picture of localizing objects;

2) on HD video picture, carry out image outline described point, and by back-end data processing center, visual profiles dot information is converted to sea chart coordinate, projects on sea chart;

3) the projected position point on sea chart, by sea chart region drawing tool, being created as section object, by calculating the geometric center in this region, on sea chart, directly obtaining the latitude and longitude information of this position;

4) by sea chart referential function, event icons and explanation are added to this position, facilitate checking and sharing of user at different levels;

5) the target location latitude and longitude information by obtaining, the locking to this target arranging video camera is monitored, dynamic in real time to understand it;

6) incident report information is sent to relevant administrative department.