A method for detecting damage area of a cushioning tape based on image segmentation

Abstract

The application belongs to the technical field of image processing, and particularly relates to a kind of buffer fabric tape damage area detection methods based on image segmentation, and the method comprises: the depth camera of two upper and lower synchronous acquisition fabric tape depth image is used, and target image and relative target image are obtained;According to the texture distribution law of target image, the regularity of target image is obtained, the regularity mean value of all depth image pairs obtained by two target image regularity mean values on the fabric tape is used to calculate the regularity of fabric tape texture;According to the difference value between the regularity of target image and the regularity of fabric tape texture, and the proportion of difference value in the regularity of fabric tape texture, the abnormal possibility of target image is calculated;Combined with the abnormal possibility of target image and the probability that all suspected areas and relative suspected areas in target image and relative target image belong to defect area, the defect degree of fabric tape is obtained, and is marked;The application improves the precision of buffer fabric tape damage area detection and positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a method for detecting damaged areas of a buffer webbing based on image segmentation. BACKGROUND

[0002] When an electrician works at a high altitude, the buffer webbing is the key line to maintain life safety, and its quality detection is crucial. In the scenario of high-altitude work, once a fall occurs, the buffer webbing needs to effectively absorb the impact force in an instant through scientific design of structure and high-quality materials to reduce the harm to the human body. In order to ensure the safety of electric power workers, it is necessary to regularly detect the safety of the buffer webbing on the protective equipment. If the buffer webbing is internally damaged, it may show bulges on the surface, that is, internal fibers are disordered and accumulated; local thinning, that is, internal fibers are broken and worn. The characteristics of these two defects on the surface of the webbing are not obvious, and if they are not detected, the electrician working at a high altitude may cause stress concentration in the instant of falling, causing the bulges or thinned places to break or collapse first.

[0003] In order to detect defects similar to the material of the buffer webbing, machine vision is currently used to scan the surface of the webbing for detection. In related technologies, for example, the Chinese patent document with the authorization announcement number CN112200790B discloses a fabric defect detection method, device and medium, which discloses using a convolutional neural network to obtain spatial and detail feature data respectively and fusing to judge fabric defects, which can improve detection accuracy.

[0004] However, the patent fabric defect detection method, device and medium have problems when facing complex damage conditions of the buffer webbing, such as local thinning and bulges caused by internal fiber breakage. Due to the lack of consideration of the physical structure and three-dimensional deformation of the webbing, the detection accuracy is not high, which brings hidden dangers to the high-altitude work of electric power workers. SUMMARY

[0005] To solve the problem that the prior art cannot detect local thinning and bulges of the webbing caused by internal fiber breakage, the present application provides a method for detecting damaged areas of a buffer webbing based on image segmentation, comprising:

[0006] The depth images of the woven belt are synchronously collected by using two depth cameras, a plurality of depth image pairs are obtained, a depth image set is formed, and any one of the depth image pairs is denoted as a target image and a relative target image; the target image is divided based on the depth distribution, a plurality of feature regions and a suspected region are obtained; the feature region distance is obtained according to the distribution rule of the feature regions of the target image; the depth similarity of all the feature regions on the target image is judged by using the feature region distance, and the regularity of the target image is calculated; the regularity of the woven belt texture is calculated by averaging the average values of the regularities of the target and the relative target images; the difference between the regularity of the target image and the regularity of the woven belt texture is compared, and the abnormality possibility of the target image is calculated; the depth extreme point is located in the suspected region, the vertical line of the extreme point is established, the vertical plane of the vertical line is moved to generate a scanning plane, the number of pixel points of each scanning plane is counted, the sequence of the number of pixel points is obtained, and the probability that the suspected region belongs to a defect is calculated in combination with the abnormality possibility of the target image; the probability that the relative suspected region belongs to a defect is located and calculated; the defect degree of the woven belt is obtained by combining the probabilities that all the suspected and relative suspected regions in all the depth image pairs belong to the defect region, and the damaged region is marked.

[0007] The prior art is difficult to find the local thinning, bulging and other defects caused by the internal fiber breakage of the woven belt during the detection of the buffer woven belt, and is easy to reduce the detection accuracy of the woven belt. The depth camera is used to collect the double-face depth images of the woven belt in the present application, and the defect degree of the woven belt is comprehensively judged by calculating a plurality of parameters such as the regularity of the target image, the regularity of the woven belt texture, the abnormality possibility of the target image and the probability that the suspected region belongs to a defect, so that the damage position is determined, and the accuracy and reliability of the damage detection of the buffer woven belt are greatly improved. The present application is more targeted, that is, it is specially designed for the buffer woven belt, fully considers the importance of the buffer woven belt in high-altitude operation and common damage forms such as woven belt wear and thinning, and uses the depth camera to collect the depth images, so that the internal damage can be detected, and the detection requirements of the buffer woven belt are better met.

[0008] Preferably, the calculation of the regularity of the target image comprises:

[0009] The feature region distance is calculated; and the regularity of the target image satisfies the following expression:

[0010] ;

[0011] In the formula, denotes the regularity of the target image; denotes the number of all the feature region sets in the target image; denotes the number of the feature regions of the i th feature region set; denotes the depth average value of the j th feature region in the i th feature region set, denotes the depth average value of the target image; represents the average depth of the feature regions spaced apart from the jth feature region by s feature region distances in the ith set of feature regions; represents a very small positive number, ensuring that the denominator is not 0; represents an absolute value function; represents a normalization function.

[0012] Preferably, the feature region distance is obtained by:

[0013] The feature regions smaller than the average depth are extracted from the target image to form a feature region, a 0°-180° straight line is generated with the center pixel point of the target image center feature region as the midpoint, the straight line containing the most feature regions is screened out as the target straight line, the feature regions on the straight line are the target set, and the target set period is calculated according to the number of target set regions and the length of the target straight line; the target straight line is translated multiple times until the image boundary is reached, multiple target set periods are obtained, and the average of all target set periods of the target image is calculated to represent the target image period, which is denoted as the feature region distance.

[0014] Preferably, the texture regularity of the braid is obtained by:

[0015] The regularity of the target image and the relative target image in the depth image pair in which the target image is located is calculated, the average of the regularity of the target image and the relative target image is calculated to obtain the texture regularity of the local area of the braid, and the texture regularity of the braid is calculated through the local regularity of all depth image pairs.

[0016] The present application can first determine the texture regularity of the braid, such as the interval of the texture and the periodicity of the depth variation, to provide a reference standard for subsequent judgment of where damage may exist, and facilitate the identification of abnormal areas deviating from the template.

[0017] Preferably, the target image abnormality possibility satisfies the following expression:

[0018] ;

[0019] In the formula, represents the target image abnormality possibility; represents the texture regularity of the braid; represents the regularity of the target image; represents a very small positive number, ensuring that the denominator is not 0; represents an absolute value function; represents a normalization function.

[0020] The application can determine whether an area may have an abnormality when the damage degree of the regularity of local texture is destroyed due to damage of the belt, which is equivalent to preliminary screening, marks a suspected problem area, reduces the range of subsequent key detection, and improves detection efficiency.

[0021] Preferably, the dividing the suspected area comprises:

[0022] Performing a region growing algorithm on the pixel points with a depth greater than or equal to the depth average in the target image to generate a first suspected area representing a suspected thinning area; performing a region growing algorithm on the pixel points with a depth less than the depth average to generate a second suspected area representing a suspected bulging area; and defining the first suspected area and the second suspected area as the suspected area.

[0023] Preferably, the calculating the probability that the suspected area belongs to a defect comprises:

[0024] Obtaining a pixel point number sequence; and the probability that the suspected area belongs to a defect satisfies the following expression:

[0025] ;

[0026] In the expression, z represents the number of times that the moving point of p moves on the baseline; 、 、 represents the number of pixel points scanned by the horizontal plane generated by the corresponding position of the moving point of p on the baseline after the zth, vth, and v+1th movement of p; represents the total number of times that the moving point of p moves on the baseline; represents the difference between the depth of the kth pixel point in the suspected area and the depth average of the target image; represents the possibility of the target image abnormality of the suspected area; represents a normalization function.

[0027] The application scans and analyzes the depth extreme points in the suspected area, records the changes in the number of pixel points scanned by the horizontal plane, which is equivalent to CT of the suspected area, can capture the shape, size, and other characteristics of the area in detail, and further analyzes the changes in the number of pixel points to provide a solid basis for judging whether it is a true defect.

[0028] Preferably, the obtaining the pixel point number sequence comprises:

[0029] Locate a depth extreme point p in the suspected area, establish a baseline perpendicular to the camera plane with p as the base point; generate a horizontal plane in the direction perpendicular to the baseline with a step size d, and record the number of pixel points on the suspected area swept by the horizontal plane of p after each movement; after three initial movements, if the number of pixel points is greater than 0 for three consecutive times, move in the original direction until it is equal to 0; if it is less than or equal to 0 for three consecutive times, move in the opposite direction; extract all movement times of the moving point of p, and the number of pixel points swept by the horizontal plane of the moving point of p in the suspected area after each movement of p, to obtain the pixel point sequence of the suspected area.

[0030] Preferably, the locating and calculating the probability of the relative suspected area belonging to defects comprises:

[0031] Locate the relative suspected area in the target image relative to the suspected area in the target image, and calculate the probability of the relative suspected area belonging to defects.

[0032] Preferably, the degree of the belt defect satisfies the following expression:

[0033] ;

[0034] In the formula, The degree of the belt defect is represented by Q, the number of depth image pairs is represented by Q, the number of defect regions on the target image is represented by Q, the probability of the wth suspected area on the target image belonging to defects in the wth depth image pair is represented by Q, the probability of the wth relative suspected area on the relative target image belonging to defects in the wth depth image pair is represented by Q, and the normalization function is represented by Q.

[0035] The present application can combine the conditions of the suspected areas in the images captured by the upper and lower cameras, verify each other, and finally comprehensively evaluate the degree of defects of the entire belt, so as to ensure that the detection result is more accurate, reduce misjudgment or omission, and finally clearly mark the damaged area for subsequent processing.

[0036] Compared with the prior art CN112200790B, the present application can accurately detect the damaged areas of the buffer belt, such as bulges and local thinning, and accurately locate the positions of these damages, thereby greatly improving the accuracy and reliability of the buffer belt damage detection. For high-altitude workers, the buffer belt is a key equipment to ensure safety, and through this method, potential hazards can be found in time to avoid safety accidents caused by belt damage. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 ​​​​​​is a flow chart illustrating a method for detecting damage area of a buffer belt based on image segmentation according to the present application;

[0038] Figure 2 is a schematic diagram of camera positions. DETAILED DESCRIPTION

[0039] The embodiment of the present application discloses a method for detecting damage area of a buffer belt based on image segmentation, referring to Figure 1 , comprising steps S1-S4:

[0040] S1: synchronously collecting depth images of the buffer belt by using two depth cameras above and below the buffer belt to obtain a target image and a relative target image.

[0041] It should be noted that the damage of the buffer belt may be hidden inside, such as fiber breakage leading to local thinning or appearing as surface abnormalities, such as bulges, and ordinary visual detection is difficult to capture the depth difference. Therefore, the depth camera is used to synchronously collect images from the top and bottom to obtain the regularity of the texture of the buffer belt, compare the difference between the regularity of the target image and the overall texture regularity of the buffer belt, calculate the abnormality possibility of the target image, divide the suspected area (bulge or thinning area) according to the relationship between the depth value and the average value, and scan and analyze the depth extreme points of the suspected area. This method can not only obtain the surface shape, but also reflect the internal structure change through the depth value, and ensure that the area where the damage may occur is covered in all directions.

[0042] It should be noted that the buffer belt of the power worker engaged in high-altitude work needs to be checked regularly, and the existing image processing technology can identify surface defects and wear conditions through image enhancement, edge detection, texture analysis and other methods when detecting the buffer belt of the safety belt used in high-altitude work, but there is certain limitation in detecting internal damage of the buffer belt. This step uses two depth cameras above and below the buffer belt to obtain a pair of depth images of the local buffer belt, and moves the buffer belt at equal steps to obtain a set of depth images.

[0043] Specifically, a used buffer belt to be detected is recovered; two depth cameras are placed above and below the buffer belt respectively, so that the camera optical axis is perpendicular to the plane of the buffer belt, and the field of view is completely overlapped, as shown in Figure 2The schematic diagram of the camera position is shown; a trigger synchronization mechanism ensures that the two cameras collect the same position of the tape image at the same time, eliminating the position deviation caused by the time difference; the upper and lower camera images are aligned and denoised; a depth image pair is obtained, which includes two depth images, namely the target image and the relative target image; the buffer tape is continuously moved along the extension direction of the tape at equal steps, and a plurality of depth image pairs are obtained to form a depth image set. It should be noted that each pixel value in the depth image represents the vertical distance from the point to the camera, usually in millimeters. The smaller the pixel depth value, the closer the vertical distance from the point to the camera; the larger the pixel depth value, the farther the vertical distance from the point to the camera.

[0044] At this point, the depth image set of the buffer tape to be detected is obtained.

[0045] S2: According to the texture distribution rule of the target image, the target image regularity is obtained, and the regularity of all depth image pairs obtained by the mean value of the regularity of the two target images is calculated to obtain the regularity of the tape texture.

[0046] It should be noted that the buffer tape is formed by interlacing warp and weft lines to form a regular texture, which is represented by equally spaced light and dark areas in the target image. The bright area represents a protrusion, and the dark area represents a depression. The depth value of the plain weave tape presents a periodic change of "protrusion-depression-protrusion". This periodicity is reflected in that the depth value of the corresponding feature area after the initial feature area in the target image moves along the alternating light and dark areas in any direction by a corresponding period is similar to the depth value of the initial feature area. Therefore, according to this feature, a method for obtaining the regularity of the texture of the buffer tape is obtained.

[0047] Specifically, one target image in a depth image pair is selected from the depth image set, the mean value of the depth values of all pixel points in the target image is calculated, and the mean value is denoted as the depth mean value; in the target image, the pixel points with a depth value less than the depth mean value are extracted to obtain a plurality of feature pixel points; region growing is performed on these feature pixel points to obtain a plurality of feature regions; the minimum circumscribed rectangle of all feature regions is generated, and half of the diagonal length of the minimum circumscribed rectangle of each feature region is calculated, which is denoted as the core distance. It should be noted that region growing is performed on these feature pixel points to aggregate adjacent feature points of the same type to form a feature region, which more clearly reflects the overall distribution form of the texture. The core distance is the maximum radius of the feature region, which is used to define the spatial range of the region and quantifies the size range of each feature region, providing a standard for judging whether the feature region belongs to the same rotating straight line, and ensuring that the selected target straight line can accurately reflect the real distribution direction and rule of the tape texture.

[0048] Preferably, the center pixel point of the target image center feature region is taken as the origin, a plurality of straight lines rotating around the origin are generated, the straight lines rotate every 1°, covering the range of 0°~180°; the perpendicular distance between each rotating straight line and the center pixel point of the feature region is calculated, if the distance is less than or equal to the core distance, it is determined that the feature region belongs to the current straight line, and the region is recorded as a target region; the number of target regions corresponding to each straight line is counted; the average value of the target regions of all straight lines is calculated, the straight line with the highest number of target regions and the corresponding direction are screened out, and the straight line is recorded as a target straight line; all target regions on the target straight line are recorded as a target set, the average distance between adjacent target regions on the target straight line is calculated by taking the ratio of the length of the target straight line to the number of target regions on the target straight line, and is recorded as the feature target set period. It should be noted that the texture of the buffer woven tape interlaced by warp and weft has significant periodicity, which not only embodies the interval law of the feature region in the same direction, but also shows the regular distribution in the vertical direction. By generating 0°~180° rotating straight lines and screening the target straight line with the most feature regions, the texture target straight line direction can be accurately captured, and the feature target set period can quantify the interval law in the direction; the target straight line is translated along the vertical direction and the average value of multiple set periods is taken, which can comprehensively reflect the texture characteristics in different directions, so as to more comprehensively extract the feature region distance.

[0049] Preferably, the distance between the center pixel points of any two adjacent feature regions perpendicular to the extension direction of the target straight line is calculated, which is set as the translation step of the target straight line on both sides along the direction perpendicular to the extension direction of the target straight line, and the target straight line is translated to both sides along the vertical direction until it stops at the image boundary, obtaining a plurality of translated straight lines and a plurality of sets corresponding thereto; all set periods are calculated and averaged to represent the average distance of all feature regions of the target image, which is recorded as the feature region distance.

[0050] It should be noted that the texture of the buffer woven tape is periodically distributed in the direction of the target straight line, and there is also a regular arrangement feature in the direction perpendicular to the target straight line. Taking the distance between the center pixel points of adjacent feature regions in the vertical direction as the translation step can not only ensure that the translated straight line can capture effective feature regions, but also conform to the distribution density of the woven tape texture in the vertical direction.

[0051] Preferably, the average depth of the feature regions in each set in the target image is obtained, and the similarity of the average depth of the feature regions spaced apart by a plurality of feature region distances is calculated to represent the regularity of the target image, and the regularity of the target image satisfies the following expression:

[0052] ;

[0053] In the formula, represents the regularity of the target image; N represents the number of all feature region sets in the target image; N represents the number of feature regions in the ith feature region set; N represents the depth mean value of the jth feature region in the ith feature region set, N represents the depth mean value of the target image; N represents the depth mean value of the feature region spaced s feature regions away from the jth feature region in the ith feature region set; N represents a very small positive number, which ensures that the denominator is not zero; N represents the absolute value function; N represents the normalization function.

[0054] In the formula, N is used to measure the degree of deviation of the depth values of the jth feature region in the ith feature region set and the feature region spaced s feature regions away from the jth feature region in the ith feature region set; N represents the square of the deviation of the depth mean value of the jth feature region in the ith feature region set from the depth mean value of the target image; The closer to 1, the more consistent the depth deviation of the jth feature region in the ith feature region set and the feature region spaced s feature regions away from the jth feature region in the ith feature region set, i.e., the greater the similarity of the depth values of the two regions. When the depth deviation direction of the two regions is opposite or the amplitude difference is large, the value of the fraction tends to 0, i.e., the smaller the similarity of the depth values of the two regions. N represents the similarity mean value between the jth feature region in the ith feature region set and all feature regions in the ith feature region set, which is referred to as the similarity of the ith feature region; N represents the similarity mean value of all feature regions in the ith feature region set; N represents the average of the similarity mean value of all feature regions in all feature region sets in the target image, which is referred to as the regularity of the target image; It should be noted that this expression quantifies this regularity by calculating the similarity of the deviation of the depth mean value of the feature region spaced s feature regions away from the target image from the depth mean value of the target image. The closer to 1, the stronger the regularity. The regularity of the target image obtained by averaging and normalizing the similarity mean value of all feature regions in all feature region sets reflects the degree of regularity of the texture of the woven tape.

[0055] Preferably, the regularity of the relative target image in the depth image pair in which the target image is located is obtained; the average regularity of the target image and the relative target image is calculated, and the texture regularity of the local area of the braid corresponding to the target image and the relative target image is obtained through the average regularity of the target image and the relative target image; the average regularity of the target image and the relative target image in all depth image pairs in the depth image set is calculated, and is recorded as the average regularity of the depth image pair; the average value of the average regularity of all depth image pairs is calculated, and the average value is recorded as the texture regularity of the braid. It should be noted that the texture regularity of the continuous whole of the buffer braid should be consistent in the local area and as a whole. By obtaining the regularity of the relative target image and calculating the average of the regularity of the target image, the texture regularity of the local braid area can be comprehensively reflected, and errors that may exist in a single image can be avoided. Further calculating the average value of the average regularity of all depth image pairs as the texture regularity of the braid can comprehensively reflect the overall texture regularity of the whole braid.

[0056] At this point, the texture regularity of the braid is obtained.

[0057] S3: According to the difference between the target image regularity and the texture regularity of the braid and the proportion of the difference in the texture regularity of the braid, the target image abnormality possibility is calculated.

[0058] It should be noted that the manufacturing process of the buffer braid determines that the normal product has structured regular features, such as texture period, thickness uniformity, and fiber arrangement direction consistency. Defects such as bulging or local thinning will destroy the regularity of these features, i.e., local bulging or thinning will cause the profile curve in the two-dimensional image to be suddenly changed, disrupt the structured regular features of the buffer braid, and reduce the regularity of the target image. This step determines whether the surface of the braid local area is bulging or locally thinned according to whether the braid local area meets the texture regularity features of the buffer braid, and records the abnormality degree of the braid local area as the target image abnormality possibility.

[0059] Specifically, the texture regularity of the braid and the regularity of any target image in the depth image set are extracted, and the target image abnormality possibility is calculated. The target image abnormality possibility satisfies the following expression:

[0060] ;

[0061] In the formula, indicates the target image abnormality possibility; indicates the texture regularity of the braid; indicates the regularity of the target image; indicates a very small positive number, which ensures that the denominator is not zero; indicates the absolute value function; indicates the normalization function.

[0062] In the formula, The greater the value, the greater the deviation between the target image regularity and the tape texture regularity, the smaller the target image regularity, and the greater the possibility of target image abnormality. The greater the value, the greater the deviation between the target image regularity and the tape texture regularity, the smaller the target image regularity, and the greater the possibility of target image abnormality. It should be noted that the formula accurately captures the abnormal characteristics of the local area by quantifying the deviation between the target image regularity and the tape texture regularity.

[0063] S4: Combine the target image abnormality possibility and the probability that all suspected areas in the target image and the relative suspected area belong to the defect area, and obtain the tape defect degree for marking.

[0064] It should be noted that this step can achieve comprehensive evaluation from local to global by combining the target image abnormality possibility and the defect probability of the suspected area and the relative suspected area. The relative suspected areas of the upper and lower images confirm each other, which can reduce the misjudgment caused by single-angle shooting. The tape defect degree obtained through the normalized calculation can intuitively reflect the damage severity of the entire tape, and accurately mark the damage area accordingly, meeting the detection needs of the buffer tape as a continuous whole.

[0065] Specifically, the depth mean value in the target image is extracted, and all pixel points with a depth value greater than or equal to the depth mean value are regionally grown to generate a first suspected area, which represents a suspected thinning area. All pixel points with a depth value less than the depth mean value are regionally grown to generate a second suspected area, which represents a suspected bulging area. The first suspected area and the second suspected area are both referred to as suspected areas. It should be noted that the thickness distribution of the buffer tape is relatively uniform in the normal state, and the depth value of the pixel points in the depth image should present regular fluctuations around the depth mean value. When the tape has local thinning, the depth value of the area will be greater than or equal to the depth mean value. When there is a bulging defect, the depth value of the area will be less than the depth mean value.

[0066] Preferably, an arbitrary suspected area is selected, and a pixel point p with an extreme value is located, and the depth of point p is denoted as . The three-dimensional position of the point p is taken as a base point, the vertical direction of the camera upward is a height reduction direction, and a straight line of the vertical camera from the point p is recorded as a base line of the point p. It is assumed that a moving point on the base line of the point p moves along the base line multiple times from the point p at a uniform speed, the moving step is d, and a plane perpendicular to the base line where the moving point is located is recorded as a horizontal plane of the moving point. When the moving point moves once, the number of pixel points scanned by the horizontal plane of the moving point in the suspected area is recorded. After three consecutive movements, if the number of pixel points is greater than 0 continuously, the moving point continues to move in the direction; if the number of pixel points is less than or equal to 0 continuously, the moving point continues to move in the opposite direction, and the movement stopping condition is that the number of pixel points is 0. The number of all movements of the moving point of the point p and the number of pixel points scanned by the horizontal plane corresponding to each movement of the moving point in the suspected area are extracted, and a pixel point number sequence of the suspected area is obtained. The change condition of the number of pixel points in the pixel point number sequence of the suspected area is analyzed, and the probability that the suspected area belongs to a defect is obtained. The probability that the suspected area belongs to a defect satisfies the following expression:

[0067] ;

[0068] In the formula, g represents the probability that the suspected area belongs to a defect; represents the number of pixel points scanned by the horizontal plane generated at the corresponding position on the base line after the vth movement of the moving point of the point p in the suspected area, and e represents the number of pixel points scanned by the horizontal plane in the suspected area; represents the number of pixel points scanned by the horizontal plane generated at the corresponding position on the base line after the (v+1)th movement of the moving point of the point p in the suspected area; represents the total number of movements of the moving point of the point p on the base line; represents the corresponding number of pixel points when the zth movement of the moving point of the point p occurs; represents the difference between the depth of the kth pixel point in the suspected area and the average depth of the relative target image; represents the abnormality possibility of the target image; represents a normalization function;

[0069] In the formula, represents the total number of pixel points corresponding to the zth movement of the moving point of the point p, and the greater the value, the greater the area of the suspected area; represents the change amount between the number of pixel points scanned by the horizontal plane generated at the corresponding position on the base line after the (v+1)th movement of the moving point of the point p in the suspected area and the number of pixel points scanned by the horizontal plane generated at the corresponding position on the base line after the vth movement of the moving point of the point p in the suspected area; represents the total change amount of the pixel points scanned by the horizontal plane generated by adjacent movements of the moving point of the point p on the base line for n times in the suspected area. This represents the average change in the number of pixels swept across the suspected area by the horizontal plane generated when the moving point p moves n adjacent times on the baseline. The larger this value is, the greater the degree to which the suspected area belongs to the defect area. This represents the sum of the differences between the depth of each pixel in the suspected region and the mean depth of the target image. A value greater than 0 indicates that the suspected region is a bulging region, and a value less than 0 indicates that the suspected region is a thinning region. The larger the value, the greater the probability that the suspected area is a defect. It should be noted that the damaged area of ​​the buffer webbing exhibits significant three-dimensional morphological differences in depth features, making it difficult to represent the degree of defect using only a two-dimensional image. This step involves locating the depth extrema of the suspected area and scanning along the baseline on a horizontal plane. The recorded sequence of pixel counts visually reflects the extent and morphological changes of the defect in the depth direction. After three consecutive movements, the direction is adjusted based on the number of pixels to ensure complete coverage of the defect area and avoid missing any defective areas. The expression for the probability that a suspected area is a defect combines region size, morphological changes, depth deviation, and the probability of anomalies in the target image. This method integrates the physical characteristics of the defect and the abnormal trends of the webbing texture, allowing the calculation results to accurately reflect the likelihood that the suspected area is a real defect.

[0070] Preferably, the suspected region in the target image relative to the target image is located and denoted as the relative suspected region; the probability that the relative suspected region belongs to the defect is calculated using the same method as calculating the probability that the suspected region belongs to the defect. It should be noted that the buffer webbing damage corresponds to the region features in the forward and reverse depth images, and forward and reverse verification can be performed by locating the relative suspected region.

[0071] Preferably, the confidence level of a suspected region being a defect is analyzed by the relationship between the probability that a suspected region in the target image is a defect and the relative probability that a suspected region is a defect; the confidence level of all suspected regions in the target image being defects is calculated as the defect confidence level of the depth image pair containing the target image; the defect confidence level of all depth image pairs in the depth image set is calculated as the webbing defect degree. The webbing defect degree satisfies the following expression:

[0072] ;

[0073] In the formula, Indicates the degree of webbing defect; Q represents the number of depth image pairs; Indicates the number of defective regions on the target image; Represents the target image in the w-th depth image pair. The probability that a suspected area is a defect; Represents the relative target image in the w-th depth image pair. The probability that a relatively suspected area is a defect; This represents the normalization function.

[0074] In the formula, Represents the w-th depth image pair. The suspected area and the first The sum of the probabilities that each relatively suspected region is a defect is denoted as the i-th region. The credibility of defects in suspected areas; Represents all of the w-th depth image pairs The sum of the defect confidence scores of each suspected region is denoted as the defect confidence score of the w-th depth image pair. The sum of the defect confidence scores of the Q pairs of images at the w-th depth is denoted as the webbing defect confidence score; finally, it is expressed as... This indicates the degree of webbing defect. It should be noted that as a continuous whole, the defects in the buffer webbing may be distributed across different areas, and a single depth image pair cannot fully reflect the overall damage condition. By combining the defect probabilities of suspected areas in the target image and the relative target image, the defect confidence level of a single depth image pair can be calculated, reflecting the damage degree of that local area. Further calculation and normalization of the defect confidence levels of all depth image pairs yields the webbing defect degree, comprehensively covering the damage situation of the entire webbing.

[0075] Preferably, a first threshold is preset, if If the defect level F exceeds the first threshold, it indicates that there are damaged areas on the webbing, and all damaged areas of this cushioning webbing are marked. It should be noted that the first threshold is set by the implementer based on the actual implementation situation; for example, the first threshold can be set to 0.8. When the calculated webbing defect level F is greater than 0.8, it indicates that the overall damage to the webbing has exceeded the safe range, posing a risk to safe use. Marking all damaged areas at this time not only clearly indicates the specific location of the problem, facilitating subsequent repair or scrap assessment, but also ensures that the test results match the safety requirements of actual applications, avoiding safety accidents caused by ambiguous defect level judgments.

[0076] Thus, a method for detecting damaged areas in buffer webbing based on image segmentation has been completed.

[0077] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for detecting a damaged area of a cushioning webbing based on image segmentation, characterized by, The application relates to a method for detecting defects in a textile belt, comprising the following steps: Synchronous acquisition of depth images of the textile belt by using two depth cameras, to obtain a plurality of depth image pairs, to form a depth image set, and to record any one of the depth image pairs as a target image and a relative target image; Based on the depth distribution, the target image is divided to obtain a plurality of feature regions and a suspected region; according to the distribution rule of the feature regions of the target image, a feature region distance is obtained, and the feature region distance is the average distance of all the feature regions of the target image; The feature region distance is used to judge the depth similarity of all the feature regions on the target image, and the regularity of the target image is calculated; The regularity of the target image satisfies the following expression: ; wherein, denotes regularity of the target image; denotes the number of all feature region sets in the target image; denotes the number of feature regions of the i-th feature region set; denotes the depth mean of the j-th feature region in the i-th feature region set, denotes the depth mean of the target image; denotes the depth mean of the feature region which is spaced by s feature region distance from the j-th feature region in the i-th feature region set; denotes a very small positive number, which guarantees the denominator not to be zero; denotes the absolute value function; denotes the normalization function; The regularity of the textile belt texture is calculated by averaging the average values of the regularities of all the target images and the relative target images; the difference between the regularity of the target image and the regularity of the textile belt texture is compared, and the abnormality possibility of the target image is calculated; In the suspected region, a depth extreme point is located, a vertical line of the extreme point is established, and a scanning plane is generated by moving the vertical plane of the vertical line; the number of pixel points of each scanning plane is counted, a pixel point number sequence is obtained, and the probability that the suspected region belongs to a defect is calculated in combination with the abnormality possibility of the target image; The target image abnormality likelihood satisfies the following expression: ; in which, represents the target image abnormality likelihood; represents the regularity of the braid texture; The probability that the suspected region belongs to a defect satisfies the following expression: ; wherein, , , represents the number of pixel points scanned by the horizontal plane generated by the moving point of p after the zth, vth, and v+1th movement on the baseline over the suspected area; represents the total number of movements of the moving point of p on the baseline; represents the difference between the depth of the kth pixel point in the suspected area and the relative target image depth average. The probability that the relative suspected region belongs to a defect is located and calculated; the suspected region in the relative target image of the target image is located, and is recorded as a relative suspected region; The defect degree of the textile belt is obtained by combining the probabilities that all the suspected regions and the relative suspected regions in all the depth image pairs belong to defect regions, and a damaged region is marked.

2. The method for detecting the damaged area of the cushioning fabric tape based on image segmentation according to claim 1, characterized in that, The feature region distance is obtained, comprising the following steps: A feature region is formed by growing a point smaller than a depth average in the target image; a 0-180-degree rotating straight line with a center pixel point of a central feature region of the target image as a midpoint is generated; a straight line containing the most feature regions is screened out as a target straight line; a feature region on the straight line is a target region; all the target regions on the target straight line are recorded as a target set; a ratio of the length of the target straight line to the number of the target regions on the target straight line is calculated to obtain a target set period; the target straight line is translated multiple times until the image boundary is reached; multiple target set periods are obtained, and an average value of all the target set periods of the target image is calculated to represent a target image period, which is recorded as the feature region distance.

3. The method of claim 1, wherein the method comprises: The regularity of the textile belt texture is calculated, comprising the following steps: The regularity of a relative target image in a depth image pair in which the target image is located is obtained; the average value of the regularities of the target image and the relative target image is calculated to obtain the texture regularity of a local region of the textile belt; and the texture regularity of the textile belt is calculated by using the average value of the average values of the local regularities of all the depth image pairs.

4. The method for detecting the damaged area of the cushioning fabric tape based on image segmentation according to claim 1, characterized in that, The suspected region is obtained, comprising the following steps: A region growing algorithm is performed on pixel points with a depth greater than or equal to a depth average in the target image to generate a first suspected region representing a suspected thinning region; a region growing algorithm is performed on pixel points with a depth smaller than the depth average to generate a second suspected region representing a suspected bulging region; and the first suspected region and the second suspected region are jointly defined as the suspected region.

5. The method for detecting the damaged area of the cushioning fabric tape based on image segmentation according to claim 1, characterized in that, The pixel point number sequence is obtained, comprising the following steps: In the suspected area, a depth extreme point p is located, a baseline perpendicular to the camera plane is established with p as the base point; a horizontal plane is generated by moving in the direction perpendicular to the baseline with a step size d, and the number of pixel points scanned by the horizontal plane of p after each movement in the suspected area is recorded; after initial movement for three times, if the number of pixel points is greater than 0 for three times in succession, movement is carried out in the original direction until the number of pixel points is equal to 0; if the number of pixel points is less than or equal to 0 for three times in succession, movement is carried out in the reverse direction; all movement times of the moving point of p and the number of pixel points scanned by the horizontal plane of the moving point of p in the suspected area after each movement of p are extracted, and a pixel point number sequence of the suspected area is obtained.

6. The method of claim 1, wherein the method is characterized by: The degree of the fabric belt defect satisfies the following expression: ； wherein represents the degree of tape defect; Q represents the number of depth image pairs; represents the number of defect regions on the target image; represents the probability that the w-th relative suspect region on the relative target image in the w-th depth image pair belongs to a defect; represents the probability that the w-th suspect region on the target image in the w-th depth image pair belongs to a defect; represents the probability that the w-th relative suspect region on the relative target image in the w-th depth image pair belongs to a defect; represents the probability that the w-th suspect region on the target image in the w-th depth image pair belongs to a defect; represents a normalization function.

Images