CN107563371A - The method of News Search area-of-interest based on line laser striation - Google Patents

Abstract

The method of News Search area-of-interest of the invention based on line laser striation belongs to computer vision measurement technical field, is related to a kind of method of the News Search area-of-interest based on line laser striation.This method predicts the area-of-interest of striation using the quick method for calculating striation kinematic parameter of packet, one group of time series laser striation scan image is shot first, using the area-of-interest of polygon extraction testee, the area-of-interest of the first striation of rectangular extraction is used.It is then based on laser corner to be grouped all images, mixing difference image is detected using transverse edge, the inter-pixel speed of striation in every group of image is quickly calculated, according to striation area-of-interest in inter-pixel speed Dynamic Extraction image.This method can accurately define the uniform motion scope of line laser striation by being grouped to sequence image, adapt to the situation of striation variable motion, improve the computational efficiency of striation kinematic parameter.Method has high efficiency, high reliability.

Description

A Method of Dynamically Searching Regions of Interest Based on Line Laser Strips

technical field

The invention belongs to the technical field of computer vision measurement and relates to a method for dynamically searching interest regions based on line laser light bars.

Background technique

Usually, image feature extraction in a computer vision measurement system can be decomposed into three parts: target detection, searching for the region of interest (ROI) of the system; image segmentation, separating the target from the background; target feature extraction. Object detection is the premise of image feature extraction, and its search quality directly affects subsequent image processing. For continuous moving objects in a large number of time-sequential images, fast and accurate ROI is the key to improving the quality and efficiency of image feature extraction.

The existing object detection methods mainly include background segmentation method, difference method between adjacent frames, optical flow method and wavelet method. The background segmentation method establishes a background model, and compares the feature parameters of the image sequence with the background model to segment the background and the target, thereby obtaining the moving target. However, since there is no high-performance rule to define the target, this method is only suitable for fixed and comparative moving scenes. simple occasions. Adjacent frame difference method is to calculate the difference between two adjacent frames of images, and judge whether there is motion by the absolute value of the difference. This method is suitable for images with multiple moving objects, but it is susceptible to noise interference and is robust. Sex is poor. The optical flow method infers the moving speed and direction of the object by detecting the change of the gray value of the image pixel over time, which is not suitable for low frame rate cameras or high-speed moving objects. The wavelet method is to perform wavelet transformation on the image, and then use band-pass filtering and other methods to process the wavelet image to obtain the target area. Its advantage is that it can detect weak targets in complex scenes, but its efficiency and reliability are poor. Feng L, Po L M, X u X, et al. Dynamic ROI based on K-means for remote photoplethy smography [C]. ICASSP, 2015: 1310-1314 proposed a dynamic ROI method based on K-means clustering algorithm, First, fix the ROI and block the feature area of the image, calculate the two characteristic parameters such as the cross-correlation coefficient and signal-to-noise ratio of the block image, and then use the K-means algorithm to cluster based on these two characteristic parameters, and finally according to the clustering results A dynamic ROI is performed every two seconds. Measurement experiments show that the dynamic ROI method can effectively extract targets and improve target signal quality. The invention patent number of Zhao Zhiyuan et al. is CN201710156308.9 "A method and system for setting up vehicle flow information monitoring equipment". Based on the adjacent relationship between each initial interest area, a new unit of interest is obtained, and the vehicle information monitoring is calculated. The number of monitoring devices required, this method reduces the demand for monitoring devices by reducing the fineness of the region of interest, and maximizes the ability to cover vehicle information between regions of interest.

Contents of the invention

The technical problem to be solved by the present invention is to solve the problems of low extraction efficiency, low error tolerance rate and poor robustness in the traditional region of interest extraction method for a large number of line laser light strip images with high noise and complexity and small target area. A method for dynamically searching regions of interest based on line laser light stripes. This method is aimed at the time series images of a group of vertical laser light strips moving laterally. First, the area of the measured object is extracted based on the polygonal ROI method, and the light strips of the initial image are extracted based on the rectangular ROI method, and then the images are evenly grouped to ensure the same The frame-to-frame speed of the light strips in the group images is constant, and the horizontal edge convolution of the mixed difference image is quickly calculated to calculate the movement speed of each group of light strips. Finally, the position of the light strips is predicted based on the movement parameters of the light strips in the measured object area, and the dynamic region of interest is calculated. extract. By grouping sequence images, this method can adapt to the situation of linear laser variable-speed scanning, and improves the accuracy and robustness of light strip area extraction; by quickly analyzing the differential image after lateral edge convolution, it greatly improves the light strip motion. The calculation efficiency and reliability of the parameters, thus ensuring the efficiency of the method.

The technical scheme adopted in the present invention is a method for dynamically searching for an area of interest based on a line laser light strip, which is characterized in that the method uses a grouping method to quickly calculate the movement parameters of the light strip to predict the interest area of the light strip; Group time-series laser light strip scanning images, use polygons to extract the region of interest of the measured object, and use rectangles to extract the region of interest of the first light strip; then group all images based on the laser rotation angle, and use horizontal edge detection for mixed difference images, Quickly calculate the inter-frame pixel velocity of the light stripe in each group of images; finally, dynamically extract the region of interest of the light stripe in the image according to the inter-frame pixel velocity; the specific steps of the method are as follows:

The first step is the global polygon ROI and the initial ROI of the light bar

Build a visual measurement system based on the line laser light bar. The line laser 2 installed on the turntable scans the object 1 horizontally through the rotating motion, and projects the line laser bar onto the object 1 vertically. The laser follows the turntable at ω Angular speed rotation, the laser light bar on the measured object moves from the position of the first light bar 3 to the last position of the light bar 4, the moving speed of the light bar is v, the rotation angle of the laser in the corresponding process is θ, and a camera takes a picture group time series images;

For this time series image, the number of images is N, and the size of each image is U×V, and the i-th image in this group of images is recorded as G(i), i=1,2,...,N; for the first and last two The arithmetic average of the images is calculated according to the formula (1) to obtain the image G _r including the background and the first and last light bars;

Determine the edge of the measured object in G _r , use the polygon to roughly extract the edge, obtain the polygonal region of interest, and record it as ROI_poly; roughly extract the rectangular boundary of the first and last light strips on the measured object in G _r , and ignore the rectangle in the image The vertical length and abscissa of the two rectangles, the horizontal coordinates u ₁ and u ₂ of the upper left corner points of the two rectangles, the horizontal length width ₁ and width ₂ of the rectangle in the image are obtained; the initial ROI area of the light bar is calculated according to formula (2) RECV, that is, the left boundary of the first light strip is u ₁ , and the width is w=max(width ₁ ,width ₂ );

RECV＝{u ₁ ,max(width ₁ ,width ₂ )} (2)

The second step is to calculate the motion parameters of the line laser light bar based on the grouped images

Assuming that the light bar moves at a uniform speed within a small rotation angle, the N time series images are divided into θ groups on average, and the images included in the jth group are G(j,1),G(j,2),...,G(j,n _j ), j=1,2,...,θ. where n _j is the number of images in group j and θ° is the total rotation angle of the line laser and θ takes a positive integer;

For the jth group of images, according to the formula (3), calculate the difference between the 1st and _{njth images, and obtain the mixed difference image G j containing} only two light bars;

G _j =abs(G(j,n _j )-G(j,1)) (3)

Convolute the image G _j according to formula (4), and calculate the transverse gradient image G _jx of G _j ;

Obtain the pixel gray value of rows 0.4V, 0.45V, 0.5V, 0.55V, and 0.6V in G _jx , denoted as G _jx (k,l), where l=1,2,...,U, k∈ K, K={0.4V, 0.45V, 0.5V, 0.55V, 0.6V}; k takes five different values in K, draws five with l as the abscissa and G _jx (k,l) as the ordinate The histogram, according to the value of k, is recorded as the k-th row histogram; find the average abscissa corresponding to the two peak clusters in the k-th row histogram with According to formula (5), calculate the pixel velocity v _j between the adjacent frames of the jth group of image light bars;

The third step is the dynamic ROI of the light bar

According to the results of the first step, it is known that the left boundary of the ROI of the light strip in the first image in the first group is u ₁₁ =u ₁ , width w, and the right boundary u ₁₁ +w; The left boundary u _jp of the ROI of the pth light strip, then the right boundary is u _jp +w;

According to ROI_poly, extract the polygonal region of interest of the p-th image in the j-th group, and make the gray value of all pixels in other regions 0; calculate the dynamic ROI parameters of the light strip image, select the region of interest as a rectangle, and the four corners are (u _jp ,0), (u _jp +w,0), (u _jp +w,V) and (u _jp ,V); the rectangle formed by these four corner points is used as the p-th image of the j-th group area of interest. In this way, the regions of interest of the light bars in all N images can be obtained.

The beneficial effect of the present invention is that the method can accurately define the uniform motion range of the line laser light stripe by grouping the sequence images, adapt to the situation of the light stripe moving at a variable speed, and improve the error tolerance and reliability of the light stripe extraction method; Based on the idea of the method, the motion prediction information is added, and the difference image after the lateral edge convolution is quickly analyzed, which improves the calculation efficiency and reliability of the motion parameters of the light strip, improves the efficiency of the region of interest extraction, and improves the area of the light strip. Extraction correctness and robustness.

Description of drawings

Figure 1 is a schematic diagram of a visual measurement system based on a vertical line laser light strip. In the figure, 1 is the measured object, 2 is the line laser, 3 is the first light bar, 4 is the last light bar, ω is the rotational angular velocity of the laser, θ is the laser rotational angle, and v is the moving speed of the light bar on the measured object .

Fig. 2 is a flowchart of a dynamic ROI method based on line laser light strips.

detailed description

The specific embodiments of the present invention will be described in detail below in conjunction with the technical solutions and accompanying drawings.

In this embodiment, the surface of the measured object 1 is a 2.5m×3.0m t800 composite material plate, and a blue-violet laser with a wavelength of 460nm is installed on the turntable, and the laser is vertically projected onto the measured object 1, and the laser rotates horizontally to scan. The laser 2 rotates with the turntable at an angular velocity of ω, and the laser light bar on the measured object 1 moves from the position of the first light bar 3 to the last position of the light bar 4. The moving speed of the light bar is v, and the laser light bar in the corresponding process The rotation angle is θ, and a set of time-series images are taken by the camera, as shown in Figure 1.

The invention adopts a camera equipped with a wide-angle lens to shoot light strip images. The camera model is view works VC-12MC-M/C 65 camera, resolution: 4096×3072, image sensor: CMOS, frame rate: full frame, maximum 64.3fps, weight: 420g. The wide-angle lens model is EF 16-35mm f/2.8L II USM, the parameters are as follows, lens focal length: f=16-35mm, APS focal length: 25.5-52.5, aperture: F2.8, lens size: 82×106. The shooting conditions are as follows: the picture pixel is 4096×3072, the focal length of the lens is 25mm, the object distance is 750mm, and the field of view is about 850mm×450mm.

Accompanying drawing 2 is the flowchart of the dynamic ROI method based on the line laser light strip. According to the operation process, the entire target detection process is divided into three steps: the global polygon ROI and the initial ROI of the light strip, the calculation of the motion parameters of the line laser light strip based on the grouped image, and the dynamic ROI of the light strip.

The first step is the global polygon ROI and the initial ROI of the light bar

First, build a visual measurement system based on line laser light strips, as shown in Figure 1, a set of time-series images are captured by a camera. Among them, the background of all images remains unchanged, the laser light bar in each image crosses the upper and lower boundaries of the measured object, and the shooting time difference between any two adjacent images is a constant value.

For a set of composite material laser light strip images captured above, the number of images is 200, and the size of each image is 4096×3072. Denote the i-th image in this group of images as G(i), where i=1, 2, . . . , 200. The image G _r including the background and the first and last light bars is calculated according to the formula (1). Use the polygonal ROI to manually extract the measured area of the composite plate in G _r and save it as ROI_poly. Use the rectangular ROI to manually extract the rectangular area of the first and last two light bars in G _r , get the upper left corner point u ₁₁ =u ₁ of the light bar, and get the horizontal length width ₁ and width ₂ of the two rectangular areas in the image, whichever is larger As the horizontal width w=max(width ₁ , width ₂ ) of the dynamic ROI.

The second step is to calculate the motion parameters of the line laser light bar based on the grouped images

It is known that when the above 200 time series images are taken, the line laser has a rotation angle of 20°. These images are divided into 20 groups, and the jth group includes images as G(j,1), G(j,2),...,G(j,10), j=1,2,...,20.

For the jth group of images, calculate the transverse gradient image G _jx of the head-to-tail difference images in the group according to formulas (3) and (4). Take the gray value of 4096 pixels in the 1536th line of G _jx as the vertical coordinate, and the horizontal coordinate l value of each pixel in G _jx as the horizontal coordinate (l=1,2,...,4096), draw a histogram. Among them, as shown in Figure 1, the histogram includes four peaks, which represent the positions of the two light bars at the beginning and end of the j-th group on the left and right borders of line 1536. The two peaks with the closest distance are used as the peak clusters, and the peak clusters in each peak cluster are calculated The average value of the abscissa of the two peaks with Draw the histogram of row k according to the above method, and find the average abscissa corresponding to the two peak clusters in the histogram of row k with where k∈K, K={1229, 1382, 1536, 1690, 1843}. Calculate the pixel velocity v _j between adjacent frames of the jth group of image light bars according to the formula (5). Calculate v ₁ , v ₂ , . . . , v ₁₀ according to the above method.

The third step is the dynamic ROI of the light bar

It is known that the left border of the ROI of the light strip of the first image in the first group is u ₁₁ =u ₁ , and the width is w. According to the formula (6), the left boundary of the ROI of the p-th light strip in the j-th group is u _jp , and the right boundary is u _jp +w.

Extract the polygonal ROI of the p-th image in the j-th group according to ROI_poly, and make the gray value of all pixels in other areas be 0. Calculate the dynamic ROI parameters of the light strip image, the region of interest is selected as a rectangle, and the four corner points are (u _jp ,0), (u _jp +w,0), (u _jp +w,3072) and (u _jp ,3072 ); The rectangle formed by these four corner points is used as the region of interest of the pth image of the jth group. From this, the regions of interest of the light bars in all images can be obtained.

The present invention is aimed at a large number of line laser light strip images with high noise and complexity and small target areas, and the traditional region of interest extraction method cannot find the target area quickly and with high accuracy. Based on shooting a group of images of the vertical line laser light strips and scanning the measured object horizontally, the polygon ROI and rectangular ROI methods are used to extract the light strips of the measured object area and the initial image, and then group the images and quickly calculate each group of light strips Motion speed, and finally predict the position of the light strip based on the movement parameters of the light strip in the area of the measured object, and extract the dynamic region of interest. The invention can adapt to the situation of variable-speed scanning of line laser, improves the calculation efficiency of light strip motion parameters, and has the characteristics of high fault tolerance rate, high efficiency, high reliability and the like.

Claims (1)

1. a kind of method of the News Search area-of-interest based on line laser striation, it is characterized in that, this method is fast using packet Speed calculates the area-of-interest of the method prediction striation of striation kinematic parameter；One group of time series laser striation scanning is shot first Image, the area-of-interest of testee is extracted using polygon, uses the area-of-interest of the first striation of rectangular extraction；Then All images are grouped based on laser corner, mixing difference image is detected using transverse edge, quickly calculates every group of image The inter-pixel speed of middle striation；Finally according to striation area-of-interest in inter-pixel speed Dynamic Extraction image；Method has Body step is as follows：

First step overall situation polygon ROI and striation Initial R OI

The vision measurement system based on line laser striation is built, the laser line generator (2) on turntable is horizontal by rotary motion To scanning testee (1), line laser bar is vertically projected on testee (1), laser is as turntable is with ω angle speed Degree rotates, and the laser striation on testee is moved to the position of last striation (4), light from the position of first striation (3) Bar movement velocity is v, and the corner of laser is θ in respective process, and shoots one group of time-series image by video camera；

For the time-series image, the quantity of image is N, and every image size is U × V, remembers i-th image in this group of image For G (i), wherein i=1,2 ..., N；Two images of head and the tail are done sums average, as shown in formula (1), obtain comprising background with The image G of head and the tail striation_r；

<mrow> <msub> <mi>G</mi> <mi>r</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mi>G</mi> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>+</mo> <mi>G</mi> <mrow> <mo>(</mo> <mi>N</mi> <mo>)</mo> </mrow> </mrow> <mn>2</mn> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>

Determine G_rThe edge of middle testee, using the polygon coarse extraction edge, polygon area-of-interest is obtained, is designated as ROI_poly；Coarse extraction G_rThe square boundary of initial and end striation on middle testee, ignore the vertical extension of rectangle in the picture And abscissa, obtain the lateral coordinates u of two rectangle upper left angle points₁And u₂, the lateral length width of rectangle in the picture₁With width₂；The Initial R OI region RECV of striation are calculated according to formula (2), i.e., the left margin of first striation is u₁, width w= max(width₁,width₂)；

RECV={ u₁,max(width₁,width₂)} (2)

Second step is based on packet diagram picture and calculates line laser striation kinematic parameter

Assuming that striation uniform motion in the range of small angle tower, is equally divided into θ groups, jth group includes image by N width time-series images For G (j, 1), G (j, 2) ..., G (j, n_j), j=1,2 ..., θ；Wherein, n_jFor the amount of images in jth group and θ ° is the laser line generator total angle of rotation and θ takes positive integer；

For jth group image, according to formula (3) to the 1st width and n-th_jWidth image asks poor, obtains only including the mixing of two striations Difference image G_j；

G_j=abs (G (j, n_j)-G(j,1)) (3)

According to formula (4) to image G_jConvolution, calculate G_jTransverse gradients image G_jx；

<mrow> <msub> <mi>G</mi> <mrow> <mi>j</mi> <mi>x</mi> </mrow> </msub> <mo>=</mo> <mfenced open = "[" close = "]"> <mtable> <mtr> <mtd> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </mtd> <mtd> <mn>0</mn> </mtd> <mtd> <mn>1</mn> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>-</mo> <mn>2</mn> </mrow> </mtd> <mtd> <mn>0</mn> </mtd> <mtd> <mn>2</mn> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>-</mo> <mn>1</mn> </mrow> </mtd> <mtd> <mn>0</mn> </mtd> <mtd> <mn>1</mn> </mtd> </mtr> </mtable> </mfenced> <mo>*</mo> <msub> <mi>G</mi> <mi>j</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>

Obtain G_jxIn 0.4V, 0.45V, 0.5V, 0.55V, 0.6V row grey scale pixel value, be designated as G_jx(k, l), wherein, l=1, 2 ..., U, k ∈ K, K={ 0.4V, 0.45V, 0.5V, 0.55V, 0.6V }；K takes five different values in K, using l as abscissa, G_jx (k, l) is that ordinate draws five histograms, according to k value, is designated as row k histogram；Two are searched in row k histogram Average abscissa corresponding to crest clusterWithPixel speed between jth group image striation consecutive frame is calculated according to formula (5) v_j；

<mrow> <msub> <mi>v</mi> <mi>j</mi> </msub> <mo>=</mo> <mfrac> <mrow> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>&amp;Element;</mo> <mi>K</mi> </mrow> </munder> <mrow> <mo>(</mo> <msub> <mover> <mi>x</mi> <mo>&amp;OverBar;</mo> </mover> <mrow> <mi>k</mi> <mo>,</mo> <mn>2</mn> </mrow> </msub> <mo>-</mo> <msub> <mover> <mi>x</mi> <mo>&amp;OverBar;</mo> </mover> <mrow> <mi>k</mi> <mo>,</mo> <mn>1</mn> </mrow> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mn>5</mn> <mrow> <mo>(</mo> <msub> <mi>n</mi> <mi>j</mi> </msub> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>

3rd step striation dynamic ROI

According to the result of the first step, it is known that the ROI left margins of the first width image striation are u in first group₁₁=u₁, width w, the right Boundary u₁₁+w；The ROI left margins u of pth width striation in jth group is calculated according to formula (6)_jp, then right margin is u_jp+w；

<mrow> <msub> <mi>u</mi> <mrow> <mi>j</mi> <mi>p</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>u</mi> <mn>1</mn> </msub> <mo>+</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>s</mi> <mo>=</mo> <mn>1</mn> </mrow> <mrow> <mi>j</mi> <mo>-</mo> <mn>1</mn> </mrow> </munderover> <msub> <mi>n</mi> <mi>s</mi> </msub> <msub> <mi>v</mi> <mi>s</mi> </msub> <mo>+</mo> <mrow> <mo>(</mo> <mi>p</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> <msub> <mi>v</mi> <mi>j</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow>

The polygon area-of-interest of pth width image in jth group is extracted according to ROI_poly, makes all pixels in other regions Gray value is 0；Optical strip image dynamic ROI parameters are calculated, area-of-interest elects rectangle as, and four angle points are (u_jp,0)、(u_jp+w, 0)、(u_jp+ w, V) and (u_jp,V)；Area-of-interest using the rectangle that this four angle points are formed as jth group pth width image；By This obtains the area-of-interest of striation in all N width images.