CN115535525A - Conveyor belt longitudinal tear detection system and method based on image matching - Google Patents

Abstract

A conveyor belt longitudinal tear detection system and method based on image matching, involving the technical fields of machine vision and image processing, including a conveyor belt and a tear detection system, the tear detection system includes a red laser generator, a high-speed CCD camera, Siren, computer processing software, the red laser generator is installed obliquely below the conveyor belt, the laser line irradiated by the red laser generator is perpendicular to the running direction of the conveyor belt, and the high-speed CCD camera is set under the conveyor belt Used to capture the laser line, the signal of the high-speed CCD camera is connected to the computer processing software, and the signal of the computer processing software is connected to the alarm. With the advantages of good monochromaticity, high contrast, and strong directionality, the inline-shaped red laser line shows a bright monochromatic thin line on the surface of the conveyor belt, which highlights the surface characteristics of the conveyor belt and is more conducive to image recognition processing.

Description

Conveyor belt longitudinal tear detection system and method based on image matching

technical field

The invention relates to the technical fields of machine vision and image processing, in particular to a system and method for detecting longitudinal tearing of conveyor belts based on image matching.

Background technique

The conveyor is the key equipment for transporting coal, ore, etc., and the cost of the conveyor belt is relatively high, which is about half of the overall cost of the conveyor. During the long-term and high-load operation of the belt conveyor, the conveyor belt faces dangers such as idler rollers, roller friction, material compression, falling material impact, scratches and punctures by foreign objects mixed with materials, resulting in accidents such as tearing of the conveyor belt occur. The conveyor runs fast and has a long distance. Once a tearing accident occurs, it must be detected in time and shut down. If it is not solved effectively, it will cause major safety accidents such as longitudinal tearing, usually tens of meters or even hundreds of meters. The conveyor belt of rice is damaged, which leads to production stoppage, equipment damage, casualties and huge economic losses, seriously affecting safe production.

The existing mainstream methods for detection of abnormal conditions such as conveyor belt tearing include three categories: manual detection, contact detection and non-contact detection. The manual detection method has low efficiency, strong subjective factors, and high labor costs; contact detection methods such as leakage detection and pressure testing, although this method is simple in structure, has low accuracy, is easy to damage, and is prone to missed detection and false detection; non-contact detection Inspection methods such as embedded method, ultrasonic method, X-ray detection, radio frequency method, machine vision. Although the embedding method, ultrasonic method, and radio frequency method are reliable, they are expensive and have complex structures; X-ray detection accuracy is high, but they have radiation effects on the human body. During the operation of the conveyor belt, for the longitudinal tearing of the conveyor belt, the fault model is uncertain, the time is random, the causal relationship is complicated, and the fault diagnosis and detection theory is not perfect, which seriously affects the safe production of the transportation system and the transportation system. The online detection of conveyor belt longitudinal tear requires a new theoretical method. In the current related technologies, the relatively novel methods for tear fault detection use image processing for detection, but due to the complexity of the industrial site background and the possible interference of the conveyor belt itself, the accuracy of the existing detection methods is not good. high.

Contents of the invention

In view of the above defects, the object of the present invention is to provide a conveyor belt longitudinal tear detection system and method based on image matching, which can accurately detect the longitudinal tear problem of the conveyor belt , can give tearing alarm information in time in industrial production, and reduce economic and equipment losses caused by tearing accidents.

In order to achieve the above object, technical scheme of the present invention is:

The conveyor belt longitudinal tear detection system based on image matching includes a conveyor belt and a tear detection system, wherein the tear detection system includes a laser generator, a CCD camera, an alarm, and computer processing software, and the laser generator The device is arranged obliquely below the conveyor belt, the laser line irradiated by the laser generator is perpendicular to the running direction of the conveyor belt, and the CCD camera is arranged below the conveyor belt to capture the laser line, The CCD camera signal is connected to the computer processing software, and the computer processing software signal is connected to the alarm.

Among them, the following steps are included: S1: using a red laser generator to emit a red laser line as an auxiliary laser source, using a high-speed CCD camera combined with a red laser line, and extracting a continuous video of the normal operation of the conveyor belt through the CCD camera; S2 : The continuous video extracted by the above CCD camera is transmitted to the computer processing software for video processing. When there is no tearing, the red laser line is a continuous curve, but at the part where the tearing occurs, the red laser line will appear discontinuous. Point, after video processing, get the status quo video with the surface characteristics of the initial state of the conveyor belt, and obtain the status quo video of the next segment of the conveyor belt after the tear judgment for dynamic update; S3: combine the status quo video with the initial status of the conveyor belt after daily maintenance Comparing the template image formed in one circle, after obtaining the template image of the surface characteristics of the conveyor belt and the current situation comparison map, it is necessary to use the image matching algorithm to accurately locate the current situation video of the conveyor belt on the template image, and then analyze the current situation video The characteristic differences between the image and the template image, and finally analyze whether these differences are longitudinal tears.

Wherein, the video processing described in S2 includes the following steps: Step 1: Carry out single-frame conveyor belt section feature image extraction on the continuous video of the conveyor belt normal operation extracted by the CCD camera, and then perform RGV graying on each frame of conveyor belt section feature image. Step 2: Carry out noise reduction processing on the single-frame conveyor belt section feature image after RGV grayscale processing in step 1, and use weighted fast median filter algorithm for noise reduction processing; Step 3: Process the image in step 2 The video adopts the automatic threshold value optimization area segmentation algorithm to segment the area of the characteristic curve of the conveyor belt section; Step 4: Remove the background image of the video in Step 3; Step 5: After removing the background of the conveyor belt, the section characteristic curve of the conveyor belt Segmented slope curve correction; Step 6: Stitching and correction of the surface feature image of the conveyor belt, the stitching correction includes jitter correction and size correction, and then use the image stitching algorithm to stitch each frame of image to obtain the status quo with the surface features of the initial state of the conveyor belt Video; Step 7: Obtain the current status video of the next section of the conveyor belt for dynamic update after the tear judgment.

Wherein, the generating step of the template image includes after the daily maintenance of the conveyor belt system, using the code disc for positioning, collecting the video of the initial state of the conveyor belt when it is empty for one lap, and performing RGV grayscale for each frame of the image, Weighted fast median filter, area segmentation algorithm for automatic threshold optimization, background removal, shake correction and size correction, and finally stitch the video into a template image with the surface characteristics of the initial state of the conveyor belt.

Among them, the process of comparing the current video with the template image includes the following steps: matching the current video to the template image, finding the surface characteristics of the initial state of the conveyor belt corresponding to the current video, and judging which characteristic areas in the current video Is the newly added area, perform area filtering on the newly added area, exclude the smaller areas in the newly added area, and perform length filter detection on the remaining newly added area, if the length of the newly added feature area exceeds the pre-set tear If the crack alarm threshold is determined, it can be determined that the conveyor belt has undergone longitudinal tearing, and the code disc is used to locate the current video, and an alarm message is issued through the siren.

Wherein, the pattern matching adopts an improved normalized cross-correlation matching algorithm.

Wherein, the implementation method of the area filter: 1. use the morphological marking algorithm to carry out 8 neighborhood calibrations to the binary image f (i, j) with the surface features of the conveyor belt, and place each feature area in the order of positive integers Mark different gray levels, set the number of marked areas as q, then the maximum gray level of the marked area is q; 2. Calculate the number of each gray level, so as to calculate the area S _k of each feature area; 3 . When the area S _k of the feature region is smaller than the preset area critical threshold Tm, the corresponding gray level is cleared to zero.

Among them, the implementation method of length filter detection: 1. Use the morphological marking algorithm to perform 8-neighborhood calibration on the binary image f(i, j) with the surface features of the conveyor belt, and mark each feature area in the order of positive integers For different gray levels, set the number of marked areas as q, then the maximum gray level of the marked area is q; 2. Find the maximum and minimum points of each gray level in the vertical direction, and their difference b is The length of the characteristic region in the vertical direction; 3. When the length b of the characteristic region is smaller than the preset length critical threshold T1, the corresponding gray level is cleared to zero.

Wherein, the standard for judging that the longitudinal tear has taken place in the conveyor belt increases the inclination angle judgment, and the inclination angle judgment is added after the length filter detection, and the method for the inclination angle judgment is as follows: 1. use the morphological marking algorithm to 8-neighborhood calibration is performed on the newly added surface features after the area and length determination, and each feature area is marked as a different gray level; 2. Straight line fitting is performed on each gray level, and their point-slope formula is obtained Equation y=kx+b, wherein k is the slope of the straight line, and then obtain the inclination angle α by the formula α=arctan (k); 3. The inclination angle α between [80,100] meets the characteristics of longitudinal tearing and is retained. Eliminate the feature areas whose inclination angle a is not within this range.

Wherein, the size correction adopts image scaling processing, and the scaling method adopts bilinear interpolation.

After adopting above-mentioned technical scheme, the beneficial effect of the present invention is:

1. A conveyor belt longitudinal tear detection system and method based on image matching designed by the present invention, the system uses a red laser as an auxiliary laser source, and the line laser emitter is placed under the conveyor belt, so that the irradiated The laser line is perpendicular to the conveyor belt and forms a certain angle with the running direction of the conveyor belt, so that when the high-speed CCD camera is shooting, with the operation of the conveyor belt system, the laser line formed on the lower surface of the conveyor belt will scan to various parts of the conveyor belt , with the advantages of good monochromaticity, high contrast, and strong directionality of the inline red laser line, a bright monochromatic thin line appears on the surface of the conveyor belt, which highlights the surface characteristics of the conveyor belt and is more conducive to image recognition processing.

2. The region segmentation algorithm of automatic threshold optimization solves the problem that the global threshold is difficult to effectively segment the red laser line features. Aiming at the problem of the relative position change of the feature points caused by the deformation of the conveyor belt when it is loaded relative to the no-load condition and the left and right vibrations and up and down vibrations of the conveyor belt when it is running, the present invention proposes a method of correcting the red laser line according to the segmental slope. method, as well as conveyor belt shake correction and image size correction methods, these methods can effectively solve the problem of feature position deviation in the conveyor belt section template image and the current situation video, thereby increasing the accuracy of image matching, so that the system can accurately detect the conveyor belt The problem of longitudinal tearing can be solved in time, and the tearing alarm information can be given in time in industrial production, so as to reduce the economic and equipment losses caused by tearing accidents.

In summary, the image matching-based conveyor belt longitudinal tear detection system and method of the present invention solve the technical problem that the accuracy of the conveyer belt longitudinal tear detection method in the prior art is not high. The tearing detection system and method can accurately detect the longitudinal tearing problem of the conveyor belt, can give tearing alarm information in time in industrial production, and reduce economic and equipment losses caused by tearing accidents.

Description of drawings

Figure 1 RGV grayscale image and the renderings of three region segmentation algorithms;

Fig. 2 Simulation effect of tear detection system;

Fig. 3 detection system flow chart;

detailed description

Below in conjunction with accompanying drawing and embodiment, further elaborate the present invention.

The orientations involved in this specification are all subject to the orientations shown in the drawings, which only represent relative positional relationships, not absolute positional relationships.

Conveyor belt longitudinal tear detection system based on image matching, including conveyor belt and tear detection system, tear detection system includes red laser generator, high-speed CCD camera, alarm, computer processing software, red laser generator It is installed obliquely under the conveyor belt. The laser line irradiated by the red laser generator is perpendicular to the running direction of the conveyor belt. The high-speed CCD camera is set under the conveyor belt to capture the laser line. The signal of the high-speed CCD camera is connected to the computer processing software. The computer processing software signal connects the siren.

In the data acquisition stage, first simulate and analyze the tearing condition of the conveyor belt during operation. When the conveyor belt system is in normal operation, if there is a sudden longitudinal tear, the tear will have a gap perpendicular to the running direction of the conveyor belt. Or overlap, according to this characteristic, use digital image processing method, use the method of combining high-speed CCD camera and straight line red laser source, carry out transmission processing in the form of video. When no tearing occurs, the red laser line is a continuous curve, but at the site of tearing, the laser line will appear discontinuous breakpoints. In this way, after the processing of multiple frames of continuous images in the video, by detecting the breakpoint position of the longitudinal tear image of the conveyor belt, it can be compared with the template image formed when the conveyor belt runs a circle in the initial state after daily maintenance. Analyze and judge whether the conveyor belt has a longitudinal tear failure. This experimental method can effectively eliminate the interference of patches, soil, etc., and improve the accuracy of judging whether longitudinal tearing occurs. The details are as follows:

A detection method of a conveyor belt longitudinal tear detection system based on image matching, comprising the following steps:

S1: Use a red laser generator to emit a red laser line as an auxiliary laser source, and use a high-speed CCD camera combined with a red laser line to extract continuous video of the normal operation of the conveyor belt through the CCD camera;

First, a mining conveyor belt system was temporarily built in a company's production plant. The system includes a conveyor belt, a three-phase motor, a code disc, a metal bracket, a series of idlers, rollers, a metal detection system, and a tear detection system. The three-phase motor drives the conveyor belt to rotate, and the tear detection system includes a red laser generator, a high-speed CCD camera, an alarm, and computer processing software. The word red laser generator is placed under the conveyor belt, so that the irradiated laser line is perpendicular to the running direction of the conveyor belt and forms a certain angle with the running direction of the conveyor belt, so that the laser line can cover a larger area. The system uses a The word red laser generator is used as an auxiliary laser source, so that when the camera is shooting, as the conveyor belt system runs, the laser line formed on the lower surface of the conveyor belt will scan to various parts of the conveyor belt. With the advantages of good monochromaticity, high contrast, and strong directionality, the inline red laser line shows a bright monochromatic thin line on the surface of the conveyor belt, which highlights the surface characteristics of the conveyor belt and is more conducive to image recognition processing.

S2: Transmit the continuous video extracted by the above-mentioned CCD camera to the computer processing software for video processing. When no tear occurs, the red laser line is a continuous curve, but at the part where the tear occurs, the red laser line will appear discontinuous. Breakpoint, after video processing, the status quo video with the surface characteristics of the initial state of the conveyor belt is obtained, and the status quo video of the next segment of the conveyor belt is dynamically updated after the tear judgment;

Wherein, video processing includes the following steps:

Step 1: Extract a single-frame conveyor belt cross-section feature image from the continuous video of the normal operation of the conveyor belt extracted by the CCD camera, and then perform RGV grayscale processing on each frame of the conveyor belt cross-section feature image;

This algorithm uses R (red component), G (green component) in the RGB color space model, and V (brightness) in the HSV color space model, a total of three types of components, and assigns corresponding weights to perform comprehensive calculations. grayscale. The grayscale effect of the standard grayscale on the stains, edges, etc. in the conveyor belt image is not very ideal. The grayscale image calculated by RGV grayscale is clearer in detail than the standard grayscale, and the edge contrast at the position of the red laser line is more obvious.

Standard grayscale formula:

I(i,j)=0.2989*RGB(i,j,1)+0.5870*RGB(i,j,2)+... 0.1140*RGB(i,j,3) (1)

RGV composite gray scale formula:

I(i,j)=0.358*RGB(i,j,1)+0.327*RGB(i,j,2)+... 0.314*HSV(i,j,3) (2)

Step 2: Perform noise reduction processing on the single-frame conveyor belt section feature image after RGV grayscale processing in step 1, and use weighted fast median filter algorithm for noise reduction processing;

Due to the complex environment of the mine where the conveyor belt system is located, the dust, light and other interference are relatively large, resulting in interference information in the image collected by the CCD camera, and noise will also be generated during the transmission and storage of the image, so the image needs to be processed. Noise reduction processing. The standard median filter requires a lot of sorting and comparison work, which is extremely time-consuming for the calculation of this study that requires noise reduction for each frame of the conveyor belt image. Therefore, on the basis of the standard median filter, the speed-up optimization is carried out, and the weighted fast median filter algorithm is adopted. The weighted fast median filter algorithm first considers the influence of the gray values of the two columns of pixels moved in and out on the median value, thus Reduces the amount of sorting calculations for areas with similar or even identical pixel gray levels. The weighted fast median filter algorithm not only achieves a better denoising effect than the standard median filter, but also better preserves the edge information of the original image. Experimental tests show that the weighted fast median filter is shorter than the standard median filter. More than 70%.

Step 3: For the video processed in step 2, use the automatic threshold value optimization region segmentation algorithm to segment the conveyor belt section characteristic curve region; after comparing the effect diagrams of the Otsu algorithm and the iterative threshold segmentation algorithm, it is found that this algorithm can effectively improve the conveyor belt. Accuracy with image segmentation.

The region segmentation algorithm of automatic threshold optimization first uses the image rotation algorithm to keep the running direction of the conveyor belt in the image as top-down or bottom-up, so as to keep the red laser line on the image belt in the general direction and keep it horizontal. Most of the traditional conveyor belts are dark in color, so the collected video data will also show a darker color, and the red laser line is used as an auxiliary laser source. A strip-shaped brightness area with higher brightness values than the normal conveyor belt area. Through observation, it can be found that the pixel gray value of this laser line irradiated area will be higher than the pixel gray value of other irradiated areas in the same column. Taking advantage of this feature, this study divides the conveyor belt into multiple regions of equal size in the longitudinal direction, and combines the advantages of iterative threshold segmentation and Otsu to perform automatic threshold value optimization on the conveyor belt image for region segmentation binarization.

The area binarization calculation method of automatic threshold optimization is as follows:

It is known that the pixel size of the original grayscale image is [m, a*n], and the image is divided into a parts according to the longitudinal average to calculate the optimal threshold value automatically, and the pixel size of each image is [m, n];

Pre-query the pixel width of the red laser line in the image as u;

Let I(i,j) be the pixel gray value of the image (i,j) position, where 1<i<m, 1<j<n;

Use the following calculation formula to calculate the optimal threshold Ta of each area of the image:

1. The gray level range of the original grayscale image is [0,1], and the pixel value 0 is selected as the initial threshold T0 of the entire image, and the threshold is set to increase & after each iteration, that is, the precision level of the threshold is &;

2. Use the initial threshold T0 to divide the image into two regions, the number of pixels with a gray level greater than T0 is N2, and the number of pixels with a gray level less than or equal to T0 is N1;

Determine whether the following inequalities hold:

N ₂ >=u*n (3)

3. If the inequality is established, use the following formula to calculate the new threshold Ti+1:

T _i+1 = T _i +& (4)

4. Repeat steps 2 and 3 until the inequality does not hold.

Step 4: remove the background image from the video in step 3;

Remove the background of the high-exposure conveyor belt: After the binarization of the region segmentation algorithm for automatic threshold optimization introduced above, a large area of white will appear in the background, and the comprehensive application of some morphological algorithms in digital image processing can be used to solve the problem. The problem of large white background elements caused by high exposure after binarization in the conveyor belt image.

Removal of low-exposure and normal-exposure conveyor belt background: In order to eliminate some interference points caused by low exposure outside the section characteristic curve, and to accurately segment and position the conveyor belt section characteristic curve in the image, the binary value of the conveyor belt is required Curve fitting is performed on the transformed image, and after comparative analysis, an iterative, low-order nonlinear least squares curve fitting algorithm is selected to fit the laser line in the image. The advantage of low-order compared with high-order is that it does not fit all points of known data into the curve, but while smoothing the curve, it is as close as possible to as many data points as possible, so that after a fitting After fitting, individual noise points in the image can be eliminated, and after multiple low-order fittings, the fitted curve will be as close as possible to the red laser line while eliminating interference points.

Step 5: After removing the background of the conveyor belt, the section characteristic curve of the conveyor belt is corrected by the segmental slope curve; the correction of the segmental slope curve can effectively solve the problem that the increase of the curvature of the laser line when the conveyor belt is loaded affects the vertical and horizontal surface of the conveyor belt. Ratio, thereby increasing the accuracy of image matching, so that the system can detect the tearing problem of the conveyor belt under the condition of no load and load.

The area irradiated by the red laser line in the image of the conveyor belt will show a straight line or a low curvature curve when the conveyor belt is unloaded, and the curvature of the area irradiated by the red laser line will be obvious due to the deformation of the conveyor belt when the conveyor belt is loaded. Increasing, also causes the pixel width of the conveyor belt in the image to decrease because the edge position of the conveyor belt shrinks inward. Therefore, the straight line correction of the laser line cannot directly use the conventional projection method. According to the changing law of the curvature of the red laser line in the conveyor belt, this research designs a segmented curve correction algorithm.

The specific content of the algorithm is as follows:

1. Make the conveyor belt image be divided into smaller areas according to the vertical direction, then the laser line in each area will be close to a straight line at this time;

2. Use the least squares method to fit the laser line in each area, and the fitting formula is (5);

3. Calculate the position of the laser line at both ends of each region according to the fitted linear coefficients a and b, and crop the region image according to the obtained position information to obtain graph G;

4. According to the fitted straight line slope a, use formula (6) to find the inclination angle α of the straight line;

α = arctan(a) (6)

5. Rotate the image G around the center point according to the inclination angle α, and use the nearest method for neighborhood interpolation, the formula is (7), to ensure that a complete rotated image H is generated;

Where x, y are both non-negative integers, and f(x, y) represents the pixel value at (x, y) of the source image.

6. Keep the length of the rotated image around the center point, and cut it with twice the pixel width of the laser line to keep the width of all areas of the conveyor belt image uniform;

7. The laser line images obtained through the above steps in all areas of the conveyor belt image are horizontally stitched together to obtain the corrected image of the conveyor belt section curve.

After the slope correction of the section feature image of the conveyor belt, the length of the section feature has been corrected, and at the same time it is straight, but the edge position of the section feature is uneven, which is not conducive to the subsequent splicing work, so it is necessary to smooth the section feature. Directly use the projection method for edge smoothing correction. The projection method is to project the image in the specified direction, and realize the purpose of image processing and analysis according to the characteristics of this projection.

Suppose the size of the image is m×n, that is, m rows and n columns, f(x, y) is the gray value at (x, y), after the region segmentation algorithm of automatic threshold optimization, the image becomes a binary image , where the black (gray value 0) area represents the background or fracture of the conveyor belt, and the white (gray value 1) area represents the cross-sectional features of the conveyor belt. The gray value of the image f(x, y) is accumulated vertically, and the average value is then projected in the horizontal direction to obtain the straight line feature T(y) of the conveyor belt section after the edge is smoothed. Its calculation formula is as follows:

Step 6: Stitching and correction of the surface feature image of the conveyor belt. The splicing correction includes shake correction and size correction, and then use the image stitching algorithm to stitch each frame of image to obtain the current situation video with the surface features of the initial state of the conveyor belt;

In the process of image acquisition, because the conveyor belt will vibrate left and right and up and down, in the conveyor belt video recorded by the CCD camera, there will be certain differences in the distance from the left and right edges of the conveyor belt to the left and right boundaries of the image in each frame of image. This will cause a left-right deviation of the tear position detected in each frame of image, so the edge of the conveyor belt needs to be corrected. In this paper, by detecting the initial position of the laser line on both sides of the conveyor belt in the image, the jitter correction is performed on the characteristic image of the conveyor belt section.

For the left edge of the laser line, we scan the characteristic image of the conveyor belt section from left to right, and record the boundary position where the white pixels rise sharply, so as to determine the left edge of the characteristic image of the conveyor belt section. For the right edge of the characteristic image of the conveyor belt section, we scan the image from right to left, and record the boundary position where the white pixels rise sharply, so as to determine the right edge of the characteristic image of the conveyor belt section.

Image size correction, in order to ensure that the number of pixels in each row of the conveyor belt cross-section characteristic image obtained through the above steps is the same for each frame of video data, and to ensure the success of the following splicing algorithm, it is necessary to perform size correction on the laser line breakpoint image, so that each The number of pixels in a row remains the same.

Image scaling processing is mainly used to change the size of the image. After the image is scaled, the size of the number of rows and columns will change accordingly. When the number of rows or columns of the image increases, the pixels of the image will become higher. The image becomes blurred; when the number of rows or columns of the image is reduced, smoothness and clarity are increased.

The zoomed image pixels cannot find the corresponding pixel points in the original image, so it is necessary to perform approximate processing according to the pixel values of the neighborhood. The general processing method is to perform interpolation processing, and common interpolation processing includes nearest neighbor point interpolation, bilinear interpolation and bicubic interpolation.

After the extraction of the surface feature area of the conveyor belt in this study, the spliced data volume is large, so the scaling method adopts bilinear interpolation, which can not only improve the calculation speed, but also have better processing results.

Each frame of image in the video, after the above steps to extract the section features of the conveyor belt and perform shake correction and size correction, is spliced together in the same order, so that the video data is converted into a binary image with the surface characteristics of the conveyor belt . The width of this feature image is the image width set during the correction of the conveyor belt section feature image size, and the number of pixels on the image height is the number of video frames multiplied by the pixel width of the laser line on the conveyor belt image.

Step 7: Obtain the status video of the next section of the conveyor belt for dynamic update after the tear judgment.

S3: Compare the video of the status quo with the template image formed when the conveyor belt runs a circle in the initial state after daily maintenance. The current situation video is precisely positioned on the template image, and then the characteristic differences between the current situation video and the initial unloaded state of the conveyor belt are analyzed, and finally whether these differences are longitudinal tears has occurred. It can be analyzed to determine whether the conveyor belt has a longitudinal tear failure.

The dynamic generation of the current video of the conveyor belt, the dynamic collection of the current video during the normal operation of the conveyor belt, also converts each frame of the image according to the above-mentioned RGV grayscale, weighted fast median filter, and segmental automatic optimization. Region segmentation algorithm binarization, morphological background removal, curve fitting, jitter correction, size correction and other steps of calculation and processing, and finally splicing the video into a current video with the surface characteristics of the initial state of the conveyor belt, and obtaining the next segment after the tear judgment The status video of the conveyor belt is dynamically updated.

Feature area matching and tearing result judgment, matching and positioning the current video in the template image, finding the surface characteristics of the initial state of the conveyor belt corresponding to the current video, and judging which feature areas in the current video are newly added areas, and adding the new Perform area filtering on the added area, exclude the smaller areas in the newly added area, and perform length filter detection on the remaining newly added areas. If the length of the newly added feature area exceeds the tearing alarm threshold set in advance, it can be judged Longitudinal tearing of the conveyor belt occurs, use the code disc to locate the current video, and send an alarm message through the alarm. After the above algorithm is obtained to compare the template image of the surface characteristics of the conveyor belt with the current situation, it is necessary to use the image matching algorithm. The current status video of the conveyor belt is accurately positioned on the template image, and then the characteristic differences between the current status video and the initial unloaded state of the conveyor belt are analyzed, and finally whether the differences are analyzed for longitudinal tearing.

The improved normalized cross-correlation matching algorithm is used for image matching, assuming that S(x, y) is the current situation video with the surface characteristics of the conveyor belt of size m × n, and T(x, y) is M × N with The conveyor belt template image of the surface characteristics of the initial state of the conveyor belt, where 0<m≤M, 0<n≤N. Move the current video S on the conveyor belt template image T, and the area covered by the template image under the current video is called the template image subgraph T(x, y). (i, j) is the coordinate of the upper left corner pixel of the template image sub-image in the template image T, 0≤i≤M-m+1, 0≤j≤N-n+1, called the reference point. In the template image T(x, y), take (i, j) as the upper left corner, take the template image subimage of size m×n, calculate its gray value similarity with the current video, and traverse the current video through the entire template image, find the template subgraph most similar to the current situation video as the final matching result [56]. Usually, the following algorithm can be used as the similarity matching algorithm of T(x, y) and S(x, y):

The normalized cross-correlation algorithm (NCC), the formula is as follows:

Among them, E(Ti, j) and E(S) represent the average gray level of the current video sub-image and the current video at (i, j) respectively

Since the template image and the status quo video in this study are binary images after conveyor belt section feature extraction and splicing, there are fewer black problematic feature areas (pixel value 1), and more white normal conveyor belt areas (pixel value 0) , so the mean difference between the template image and the current video is very small, so the improved NCC algorithm is used, and the formula (11) is used to calculate the gray level correlation of the two images:

Since the template image in this study and the current video are corrected with the same size when they are stitched together by multiple images, the width of the two images is the same. When searching for matching points, only the rough positioning of the code disc is needed. The matching result can be obtained by performing top-down matching within a certain range of the template image.

After the image matching is successful, the program will feed back a position information, which is the corresponding position information of the first element in the upper left corner of the conveyor belt status video on the template image. Using this position information, according to the size of the current video of the conveyor belt, the corresponding position of the current video will be displayed on the conveyor belt template image with a red frame.

Because the initial surface state of the conveyor belt has a certain probability to be a characteristic area formed by objects that are easy to fall off such as mud and stains, this part of the area may fall off with the normal operation of the conveyor belt system, resulting in such characteristic areas on the conveyor belt. There are characteristic regions on the template image, but not on the live video.

Therefore, for the processing of the feature positions corresponding to the current video and the template image, only the feature areas added to the current video need to be preserved, and other feature areas are not reserved. When comparing the current status video of the conveyor belt with the template image, the required effect cannot be achieved with a simple AND or operation.

In this study, the current video is subtracted from the corresponding position of the template image, and the [0,1] part of the subtraction result is retained, and the other results are discarded. This method can effectively preserve the newly added surface feature area of the conveyor belt in the current video.

Area determination, when the conveyor belt system is in normal operation, there may be dirt, stains, conveyor belt joints, small patches and other objects attached to the surface of the conveyor belt, affecting the continuity of the red laser line, resulting in spots, In the case of small-area feature lines, etc., since the area of these feature points is too small, it is not considered to be longitudinal tearing. Therefore, before judging the tearing situation, it is necessary to filter these feature positions with small areas by using the area filter method. Set the threshold of the area filter to Tm, the area smaller than Tm will be filtered out, and the area larger than Tm will be retained.

The implementation method of area filtering:

1. Use the morphological marking algorithm to perform 8-neighborhood calibration on the binary image f(i, j) with surface features of the conveyor belt, and mark each feature area into different gray levels in the order of positive integers, and set the mark The number of regions is q, then the maximum gray level of the marked region is q;

2. Calculate the quantity of each gray level, thereby calculate the size of the area S _k of each feature region;

3. When the area S _k of the feature region is smaller than the preset area critical threshold Tm, the corresponding gray level is reset to zero.

In this way, the effect of area filtering can be achieved. It can be seen that area filtering can effectively filter out targets with smaller areas in the binary image, so the algorithm also has a good filtering effect on the status quo video representing the surface characteristics of the conveyor belt. Effect.

length determination. After area filtering, the newly added feature areas with small areas in the current video are excluded, and what is left is the newly added feature areas with a large enough area, but this feature area may be conveyor belt joints, horizontal patches, etc. In order to eliminate the misjudgment caused by this situation to the tear judgment, it is necessary to perform length filter detection on the feature area with a larger area newly added in the current video after area filtering.

The length filtering method is similar to the area filtering method, except that the area calculation in the middle part is changed to length calculation. Change the threshold Tm of the area filter to the threshold Tl of the length filter.

The implementation method of length filtering:

1 Use the morphological marking algorithm to perform 8-neighborhood calibration on the binary image f(i, j) with surface features of the conveyor belt, and mark each feature area into different gray levels in the order of positive integers. The number of regions is q, then the maximum gray level of the marked region is q;

2. Find the maximum and minimum points of each gray level in the vertical direction, and their difference b is the length of the feature area in the vertical direction;

3. When the length b of the feature region is smaller than the preset length critical threshold T1, the corresponding gray level is cleared to zero.

After the length filtering is completed, the rest of the newly added feature area map is the feature area that satisfies the tearing condition in both area and length. However, in order to make the tearing judgment more accurate, it is necessary to judge the tilt angle of the image.

Determination of tilt angle. According to the experience of several senior maintenance workers in the actual application site of the mine conveyor belt system, when the conveyor belt has a longitudinal tear failure, the tear angle will be between 80 degrees and 100 degrees. After the area and length are determined, the area and length of the newly added feature image that meet the characteristics of longitudinal tearing are retained. At this time, it is still necessary to detect and determine the inclination angle of the remaining feature area. If the feature area has been fitted by a straight line When the slope is between [5.67, +∞] and [-5.67, -∞], that is, when the inclination angle α is between 80 and 100, it can be considered that longitudinal tearing has occurred.

The method of judging the tilt angle is as follows:

1. Use the morphological marking algorithm to perform 8-neighborhood calibration on the image after the area and length of the newly added surface features are determined, and mark each feature area as a different gray level;

2. Carry out straight line fitting to each gray level respectively, obtain their point slope equation y=kx+b, wherein k is the slope of the straight line, and then obtain the inclination angle α by formula (12);

α=arctan(k) (12)

3. Those whose inclination angle α is between [80, 100] conform to the characteristics of longitudinal tearing and are retained, and the characteristic areas whose inclination angle α is not within this range are eliminated.

After the comprehensive application of the three determination methods of area, length and inclination angle, if these three characteristic conditions are met at the same time, it is recognized as a tear feature, and it can be determined that a longitudinal tear has occurred, and an alarm message will be issued through the alarm .

Measured data verification: In order to verify the feasibility of the scheme, we collected simulated experimental data on a temporary conveyor belt system in a company's production plant, and verified the algorithm on the corresponding mathematical software in the laboratory.

The experimental results show that by performing image matching tear detection analysis on a group of conveyor belt videos, the obtained template image and current video are shown in Figure 2, and the tear detection results are shown in Figure 2(b), and the red box is The area determined to be a tear location. Among the characteristic areas shown in Figure 2(a), the blue box is the feature area obtained by feature extraction after the interference caused by soil, patches, etc. to the laser line, the red box is the feature area caused by cracks, and the green box is The inside is the characteristic position caused by the conveyor belt patch. After matching and comparing the characteristic image with the template image, the original characteristic region of the initial state of the conveyor belt is excluded, and then the newly added area is excluded after judging methods such as area, length, and inclination angle. After interference caused by small pieces of soil and stains, etc., the remaining characteristic areas are the parts where longitudinal tears have occurred.

A kind of conveyor belt longitudinal tear detection system and method based on image matching described in the present invention illustrates the superiority of the improved algorithm and the effectiveness of the method through specific experimental data, and solves the problems caused by conveyor belt patches, joints, and soil The interference problem caused by single image tear fault identification can effectively improve the accuracy of conveyor belt tear detection results, which is of great significance in the fields of machine vision and image processing.

The present invention is not limited to the above-mentioned specific implementation manners, and various transformations made by those skilled in the art starting from the above-mentioned ideas without creative work all fall within the scope of protection of the present invention.

Claims (10)

1. The conveyer belt longitudinal tear detection system based on image matching, comprises conveyer belt and tear detection system, it is characterized in that, described tear detection system comprises laser generator, CCD camera, siren, computer processing software, described The laser generator is arranged obliquely below the conveyor belt, the laser line irradiated by the laser generator is perpendicular to the running direction of the conveyor belt, and the CCD camera is arranged below the conveyor belt to capture the laser light. line, the CCD camera signal is connected to the computer processing software, and the computer processing software signal is connected to the alarm. 2. The detection method of the conveyor belt longitudinal tear detection system based on image matching is characterized in that, comprising the following steps: S1: Use a red laser generator to emit a red laser line as an auxiliary laser source, and use a high-speed CCD camera combined with a red laser line to extract continuous video of the normal operation of the conveyor belt through the CCD camera; S2: Transmit the continuous video extracted by the above-mentioned CCD camera to the computer processing software for video processing. When no tear occurs, the red laser line is a continuous curve, but at the part where the tear occurs, the red laser line will appear discontinuous. Breakpoint, after video processing, the status quo video with the surface characteristics of the initial state of the conveyor belt is obtained, and the status quo video of the next segment of the conveyor belt is dynamically updated after the tear judgment; S3: Compare the video of the status quo with the template image formed when the conveyor belt runs a circle in the initial state after daily maintenance. The current situation video is precisely positioned on the template image, and then the characteristic differences between the current situation video and the template image are analyzed, and finally whether these differences are longitudinal tearing occurs. 3. the detection method of the conveyer belt longitudinal tear detection system based on image matching according to claim 2, is characterized in that, video processing described in S2 comprises the following steps: Step 1: Extract a single-frame conveyor belt cross-section feature image from the continuous video of the normal operation of the conveyor belt extracted by the CCD camera, and then perform RGV grayscale processing on each frame of the conveyor belt cross-section feature image; Step 2: Perform noise reduction processing on the single-frame conveyor belt section feature image after RGV grayscale processing in step 1, and use weighted fast median filter algorithm for noise reduction processing; Step 3: For the video processed in step 2, the area segmentation algorithm of automatic threshold value optimization is used to segment the area of the characteristic curve of the conveyor belt section; Step 4: remove the background image from the video in step 3; Step 5: After removing the background of the conveyor belt, correct the section characteristic curve of the conveyor belt with a segmented slope curve; Step 6: Stitching and correction of the surface feature image of the conveyor belt. The splicing correction includes shake correction and size correction, and then use the image stitching algorithm to stitch each frame of image to obtain the current situation video with the surface features of the initial state of the conveyor belt; Step 7: Obtain the status video of the next section of the conveyor belt for dynamic update after the tear judgment. 4. the detection method of the conveyer belt longitudinal tear detection system based on image matching according to claim 2, is characterized in that, the generating step of described template image comprises after conveyer belt system daily maintenance, utilizes code disc to carry out positioning, Collect the video of the initial state of the conveyor belt when it is empty, and perform RGV grayscale, weighted fast median filter, automatic threshold optimization area segmentation algorithm, background removal, shake correction and size correction for each frame of the image. Finally, the video is stitched into a template image with the surface features of the initial state of the conveyor belt. 5. the detection method of the conveyor belt longitudinal tear detection system based on image matching according to claim 2, is characterized in that, the process of comparing the current situation video with the template image comprises the following steps: carrying out the current situation video on the template image Graphic matching, find the surface characteristics of the initial state of the conveyor belt corresponding to the current video, and determine which feature areas in the current video are newly added areas, and perform area filtering on the newly added areas to exclude areas with smaller areas in the newly added areas, Perform length filter detection on the remaining newly added areas. If the length of the newly added feature area exceeds the pre-set tear alarm threshold, it can be determined that the conveyor belt has undergone longitudinal tearing, and the current video is positioned using the code disc. And send out alarm information through the siren. 6. The detection method of the conveyor belt longitudinal tear detection system based on image matching according to claim 5, characterized in that, the image matching adopts an improved normalized cross-correlation matching algorithm. 7. the detection method of the conveyer belt longitudinal tear detection system based on image matching according to claim 5, is characterized in that, the realization method of described area filter: 1. with the marking algorithm of morphology, to belt conveyer belt Binary image of surface features

Carry out 8-neighborhood calibration, and mark each feature area into different gray levels in the order of positive integers. If the number of marked areas is q, then the maximum gray level of the marked area is q; 2. Calculate the gray level of each gray level Quantity, so as to calculate the area of each feature region

Size; 3. When the area of the feature area

If it is smaller than the preset area critical threshold Tm, the corresponding gray level is cleared to zero. 8. the detection method of the conveyer belt longitudinal tear detection system based on image matching according to claim 5, is characterized in that, the realization method of length filter detection: 1. with the marking algorithm of morphology, to belt conveyer belt surface binary image of features

Carry out 8-neighborhood calibration, and mark each feature area into different gray levels in the order of positive integers. If the number of marked areas is q, then the maximum gray level of the marked area is q; 2. Find each gray level in For the maximum and minimum points in the vertical direction, their difference b is the length of the feature area in the vertical direction; 3. When the length b of the feature area is smaller than the preset length critical threshold T1, the corresponding gray level is cleared to zero . 9. the detection method of the conveyer belt longitudinal tear detection system based on image matching according to claim 5, is characterized in that, the standard that judges that conveyer belt longitudinal tear has taken place increases inclination angle judgment, and described inclination angle judgment increases to After the length filter detection, the method for determining the angle of inclination is as follows: 1. use the morphological marking algorithm to carry out 8 neighborhood calibrations on the newly added surface features after area and length determination, and mark each feature region as Different gray levels; 2. Fit straight lines to each gray level to find their point-slope equations

, where k is the slope of the line, and then by the formula

Find the angle of inclination

; 3. Tilt angle

The characteristics of longitudinal tearing between [80,100] are preserved, and the inclination angle

Feature regions that are not within this range are eliminated. 10. The detection method of the conveyor belt longitudinal tear detection system based on image matching according to claim 3, wherein the size correction adopts image scaling processing, and the scaling method adopts two-line sub-interpolation.

Images