Pipeline robot pipeline defect identification method and equipment for pipeline safety maintenance

Abstract

The invention relates to the technical field of image processing, and provides a pipeline robot pipeline defect identification method and device for pipeline safety overhaul, and the method comprises the steps: obtaining continuous frames of pipeline inner wall images through the driving of a pipeline robot in a pipeline; extracting a plurality of feature points from each frame of pipeline inner wall image to obtain a plurality of matched feature point pairs in two adjacent frames of pipeline inner wall images; based on the distribution of the edges of the two adjacent frames of pipeline inner wall images and other edges, obtaining the crack possibility of each edge in each pipeline inner wall image; the crack possibility difference of the same edge in the pipeline inner wall images of the adjacent scales is analyzed, and the interval regularity of all the edges is obtained; obtaining the gray uniformity of each edge; the crack authenticity of each edge is obtained; and acquiring the crack density of each pipeline inner wall image, and carrying out pipeline defect identification and alarm based on the crack density. The invention aims to solve the problem of recognition errors caused by the fact that internal defects of a pipeline are similar to texture features.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to a pipeline defect identification method and equipment for pipeline robots used in pipeline safety maintenance. Background Technology

[0002] Pipeline robots are intelligent equipment that can walk along the inside or outside of pipelines for inspection, monitoring, maintenance and repair operations. They can replace manual labor in entering high-risk, narrow, toxic and harmful pipeline environments and, combined with intelligent communication technology, achieve real-time monitoring of pipeline safety status.

[0003] Traditional pipeline defect detection methods are inefficient and risky. Pipeline robots are popular in pipeline defect detection, but they have not yet solved the problem of "automatic identification" of defects. In crack identification, various error information still interferes. For example, defects, cracks and textures inside the pipeline are indistinguishable due to their similar visual features, which can easily cause errors in the identification of defects inside the pipe wall. Summary of the Invention

[0004] This invention provides a pipeline defect identification method and equipment for pipeline safety maintenance using pipeline robots, to solve the problem of identification errors caused by the similarity between internal pipeline defects and texture features. The specific technical solution adopted is as follows: This invention proposes a pipeline defect identification method using pipeline robots for pipeline safety maintenance. The method includes the following steps: The pipeline robot travels through the pipeline and continuously collects images to obtain consecutive frames of images of the pipeline's inner wall. Several feature points are extracted from each frame of the pipe inner wall image by edge detection. Based on the distribution and grayscale representation of the feature points between two adjacent frames of the pipe inner wall image, several matching feature point pairs are obtained in the two adjacent frames of the pipe inner wall image. Based on the distribution of the matching feature point pairs between the corresponding edges of the two adjacent frames of the pipe inner wall image and other edges in the image, the crack probability of each edge in each pipe inner wall image is obtained. Multi-scale images of the pipe's inner wall are obtained by downsampling the pipe's inner wall image. The differences in the probability of cracks at the same edge in pipe inner wall images of adjacent scales are analyzed to obtain the spacing regularity of each edge. Based on the gray-level difference changes of adjacent pixels at the same edge in pipe inner wall images of each scale, the gray-level uniformity of each edge is obtained. Combining the spacing regularity, the authenticity of cracks at each edge is obtained. Based on the authenticity of cracks at each edge in the inner wall images of each pipe, the crack density of each inner wall image is obtained, and pipe defect identification and alarm are performed based on the crack density.

[0005] Optionally, the specific method for extracting several feature points from each frame of the pipe inner wall image through edge detection includes: For any frame of the pipe inner wall image, the Canny edge detection algorithm is used to obtain several edges of the pipe inner wall image in that frame; Feature points were detected on each edge using the FAST algorithm, and several feature points were extracted.

[0006] Optionally, the specific method for obtaining several matching feature point pairs in two adjacent frames of the pipe inner wall image includes: For any edge on any frame of the pipe inner wall image, all feature points on that edge are taken as target feature points. The pipe inner wall image of the next adjacent frame is obtained. For any target feature point, the Hamming distance between the target feature point and all feature points in the next adjacent frame of the pipe inner wall image is obtained. The feature point corresponding to the minimum value among all obtained Hamming distances is taken as the most similar point of the target feature point. The feature point corresponding to the second smallest value among all obtained Hamming distances is taken as the second most similar point of the target feature point. The ratio of the Hamming distance between the target feature point and its most similar point to the Hamming distance between the target feature point and its second most similar point is obtained and used as the matching ratio of the target feature point in the next adjacent frame of the pipe inner wall image. For all target feature points on the edge, obtain the matching ratio of the inner wall image of the pipe in the adjacent previous frame. Arrange all matching ratios from smallest to largest and obtain the difference between adjacent matching ratios. Take the minimum value of the two matching ratios corresponding to the maximum value among all the obtained differences as the matching threshold of the edge. Take several target feature points whose matching ratio is less than the matching threshold as several reference feature points on the edge. Take each reference feature point and its corresponding most similar point as a matching feature point pair for the edge.

[0007] Optionally, the probability of cracks at each edge in the inner wall images of each pipe is obtained using the following method: For any frame of the pipe inner wall image, obtain several reference feature points of each edge on the pipe inner wall image of that frame; take any one of these edges as the target edge, obtain the shortest distance between all reference feature points on any other edge besides the target edge and the target edge, obtain the absolute value of the difference between the shortest distances corresponding to any two reference feature points on that other edge, and use the mean of the absolute values ​​of the differences corresponding to any two reference feature points on that other edge as the distribution difference factor between that other edge and the target edge; Obtain the distribution difference factors between each other edge and the target edge, and use the mean of all distribution difference factors as the distribution difference coefficient of the target edge; perform linear normalization on the distribution difference coefficients of all edges in the pipe inner wall image of this frame, and use the result as the initial crack probability of each edge. For several matching feature point pairs of the target edge, obtain the edge in the adjacent previous frame of the pipe inner wall image where the most similar point of each reference feature point is located. The edge containing the most similar points is taken as the matching edge of the target edge in the adjacent previous frame of the pipe inner wall image. Obtain the initial crack probability of the matching edge corresponding to the target edge. Take the average of the initial crack probabilities of the target edge and its matching edge as the crack probability of the target edge.

[0008] Optionally, the specific method for downsampling the pipe inner wall image to obtain multi-scale pipe inner wall images includes: For any frame of the pipe inner wall image, downsample multiple times, and combine with the original scale of the pipe inner wall image of that frame to obtain pipe inner wall images at several scales.

[0009] Optionally, the regularity of the spacing between the edges is obtained by the following method: For any edge in any frame of the pipe inner wall image, obtain the probability of cracks in the pipe inner wall images at various scales for that edge; Calculate the absolute value of the difference between the crack probability of the edge in the pipe inner wall images of any two adjacent scales, and obtain the mean of the absolute values ​​of the difference between the crack probability of the edge at all two adjacent scales, as the crack variation factor of the edge at multiple scales. The ratio of the crack variation factor to the average crack probability of the edge in the pipe inner wall images at various scales is used as the spacing regularity of the edge.

[0010] Optionally, the grayscale uniformity of each edge is obtained by the following method: For any frame of the pipe inner wall image corresponding to a multi-scale pipe inner wall image, for any edge at any scale, obtain the absolute value of the difference between the gray values ​​of any two adjacent pixels on the edge, and use the reciprocal of the mean of the absolute values ​​of the differences between the gray values ​​of all two adjacent pixels on the edge as the gray uniformity factor of the edge at that scale. The gray-level uniformity of an edge is defined as the ratio of the mean to the variance of the gray-level uniformity factor across all scales.

[0011] Optionally, the specific method for obtaining the authenticity of cracks at each edge includes: For any edge in any frame of the pipe inner wall image, the reciprocal of the product of the spacing regularity and grayscale uniformity of the edge is used as the authenticity of the crack on that edge.

[0012] Optionally, the crack density of the inner wall images of each pipe is obtained using the following method: For any frame of the pipe inner wall image, the average value of the crack authenticity of all edges in the frame of the pipe inner wall image is taken as the crack density of the frame of the pipe inner wall image.

[0013] The present invention also proposes a pipeline robot for pipeline defect identification for pipeline safety maintenance. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above method.

[0014] The beneficial effects of this invention are as follows: This invention acquires images during the axial movement of a pipeline robot inside a pipeline. First, it analyzes feature points in adjacent frames of pipeline inner wall images to filter and match feature point pairs. These pairs are then used to match edges in pipeline inner wall images from different perspectives and to remove feature points that do not clearly reflect edge features. Next, it quantifies edges with relatively regular distances to obtain the probability of cracks by considering the changes in distance between edges and other edges in the image. Furthermore, it performs multi-scale sampling on single-frame images. The smaller the change in the probability of an edge being a crack at multiple scales and the larger its overall appearance, the greater the probability that it is a crack, and the more regular its distribution with other edges. Smaller spacing leads to regularity; further analysis of grayscale differences in edge pixels at multiple scales reveals that a more uniform grayscale distribution indicates a greater likelihood of uniform texture, thus obtaining grayscale uniformity and ultimately quantifying crack authenticity; crack authenticity is used to identify crack defects, and the crack density of the entire pipeline inner wall image is quantified from different perspectives for defect identification, judgment, and alarm; a comprehensive assessment of regional crack status is achieved by integrating the analysis results of all edge lines from multiple viewpoints, further quantifying the authenticity and density of cracks within the captured area, providing a more comprehensive analysis for pipeline safety maintenance, and improving the robustness of pipeline crack defect detection. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a pipeline defect identification method using a pipeline robot for pipeline safety maintenance, provided in one embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 The diagram illustrates a pipeline defect identification method using a pipeline robot for pipeline safety maintenance, provided by an embodiment of the present invention. The method includes the following steps: Step S001: The pipe robot travels in the pipe and continuously collects images to obtain continuous frames of images of the inner wall of the pipe.

[0019] The purpose of this embodiment is to address the issue that when pipeline robots are used for pipeline safety inspection, the visual features of cracks and textures inside the pipeline are similar and difficult to distinguish, leading to errors in pipeline defect identification. Therefore, it is necessary to perform difference analysis by examining the differences between the texture distribution of the pipeline inner wall image and the appearance of cracks, thereby improving the accuracy of crack defect identification.

[0020] It should be noted that the pipeline robot moves forward along the pipeline axis according to a preset path. Equipped with a 20-megapixel zoom industrial camera, the lens is initially adjusted to face forward, and LED supplementary lighting is used to detect cracks and defects on the inner wall of the pipeline in real time. From the detection of a suspected crack to approaching the crack, the pipeline robot's lens slowly moves from facing forward to facing the crack directly in the pipe wall. The lens automatically focuses, the crack information gradually increases in size, and the brightness of the supplementary lighting is automatically adjusted, making the image clearer and clearer. When it is about to identify clear crack information, it will sound an alarm and send the crack location to the cloud.

[0021] Specifically, the brightness of the LED light panel is set according to the pipe diameter and material to ensure uniform lighting during shooting; the pipe robot detection path is set with a cruising speed of 0.2m / s; during the journey, the pipe wall is photographed for inspection, and the photos are uploaded to the cloud from the time a suspected crack is detected to the time of arrival. The photos of suspected cracks are taken from far away to near, and the cracks are taken from small to large, with the images gradually becoming clearer; during shooting, it is ensured that there is an 80% overlap area between two consecutive frames; the photos uploaded to the cloud are smoothed and denoised, and then grayscale processed to serve as consecutive frames of images of the inner wall of the pipe.

[0022] Step S002: Extract several feature points from each frame of the pipe inner wall image through edge detection. Based on the distribution and grayscale representation of the feature points between two adjacent frames of the pipe inner wall image, obtain several matching feature point pairs in the two adjacent frames of the pipe inner wall image. Based on the distribution of the matching feature point pairs between the corresponding edges of the two adjacent frames of the pipe inner wall image and other edges in the image, obtain the crack probability of each edge in each pipe inner wall image.

[0023] It should be noted that feature points that can represent cracks and textures are extracted from any two fields of view. The extracted crack and texture feature points differ due to their different characteristics. The same feature points in the two fields of view are located and matched based on the similarity of the feature points (Hamming distance).

[0024] Preferably, in one embodiment of the present invention, several feature points are extracted from each frame of the pipe inner wall image through edge detection. Based on the distribution and grayscale representation of feature points between adjacent frames of the pipe inner wall image, several matching feature point pairs in adjacent frames of the pipe inner wall image are obtained. The specific method includes: For any frame of the pipe inner wall image, the Canny edge detection algorithm is used to obtain several edges of the pipe inner wall image in that frame; for each edge, the FAST algorithm is used to detect feature points. On one edge, the pixels are traversed from left to right and from top to bottom to extract several feature points; Canny edge detection and FAST algorithm are existing technologies and will not be described in detail in this embodiment.

[0025] Furthermore, for any edge on the pipe inner wall image of this frame, all feature points on that edge are taken as target feature points; the pipe inner wall image of the adjacent previous frame is obtained; for any target feature point, the Hamming distance between the target feature point and all feature points in the adjacent previous frame pipe inner wall image is obtained; the feature point corresponding to the minimum value among all obtained Hamming distances is taken as the most similar point of the target feature point; the feature point corresponding to the second smallest value among all obtained Hamming distances is taken as the second most similar point of the target feature point; the ratio of the Hamming distance between the target feature point and its most similar point to the Hamming distance between the target feature point and its second most similar point is obtained, and this ratio is taken as the matching ratio of the target feature point in the adjacent previous frame pipe inner wall image; the calculation of Hamming distance is a prior art, and will not be described in detail in this embodiment.

[0026] Furthermore, for all target feature points on the edge, the matching ratio of the previous frame of the pipe inner wall image is obtained. All matching ratios are arranged from smallest to largest, and the difference between adjacent matching ratios is obtained (the difference is obtained by decreasing the larger value). The minimum value of the two matching ratios corresponding to the maximum value among all the obtained differences is used as the matching threshold of the edge. Several target feature points with matching ratios less than the matching threshold are used as several reference feature points on the edge. Each reference feature point and its corresponding most similar point are used as a matching feature point pair for the edge. According to the above method, several matching feature point pairs are obtained for each edge of the pipe inner wall image of the frame and the previous frame of the pipe inner wall image. It should be noted that the matching feature point pairs are only obtained based on any frame of the pipe inner wall image and the previous frame of the pipe inner wall image. The basic object is the pipe inner wall image of that frame. However, in the process of obtaining matching feature point pairs for the next frame of the pipe inner wall image, the basic object is the next frame of the pipe inner wall image.

[0027] It should be noted that by performing multi-view analysis of the same area using two adjacent frames of the pipe inner wall images, feature points on the same edge are matched and reference feature points are selected. In the subsequent analysis of the regularity of edge distribution, the influence of feature points that may not exist in all views is reduced, making the analysis of the regularity of edge distribution more accurate.

[0028] It should be further noted that the processing texture on the inner surface of the pipe wall exhibits a highly regular periodic distribution, and the distance between any two textures is basically the same at any two locations; while the location of cracks inside the pipe wall is random and unpredictable, and the direction of cracks is not unidirectional, often exhibiting multi-directional and irregular tortuous extension characteristics. Therefore, the shortest distance between the crack edge and any point between the regular textures continuously changes with different locations.

[0029] Preferably, in one embodiment of the present invention, the probability of cracks at each edge in each pipe inner wall image is obtained based on the distribution of matching feature point pairs between corresponding edges in two adjacent frames of pipe inner wall images and other edges in the images. The specific method includes: For any frame of the pipe inner wall image, obtain several reference feature points for each edge on the frame of the pipe inner wall image; take any one of these edges as the target edge, obtain the shortest distance between all reference feature points on any other edge and the target edge, obtain the absolute value of the difference between the shortest distances corresponding to any two reference feature points on any other edge, and take the mean of the absolute values ​​of the differences between any two reference feature points on any other edge as the distribution difference factor between the other edge and the target edge; obtain the distribution difference factor between each other edge and the target edge according to the above method, and take the mean of all distribution difference factors as the distribution difference coefficient of the target edge; perform linear normalization on the distribution difference coefficients of all edges in the frame of the pipe inner wall image, and take the result as the initial crack probability of each edge.

[0030] It should be noted that by analyzing the distribution difference between edges through the distribution of reference feature points of each edge and the target edge, if the shortest distance between textures is relatively fixed, the average absolute value of the distribution difference factor, i.e., the difference of the shortest distance, will be small. However, if the average value is large, there may be crack edges between the two edges, resulting in uneven distribution.

[0031] Furthermore, for several matching feature point pairs of the target edge, the edges where the most similar points of each reference feature point are located in the adjacent previous frame of the pipe inner wall image are obtained. The edge containing the most similar points is taken as the matching edge of the target edge in the adjacent previous frame of the pipe inner wall image. The initial crack probability of the matching edge corresponding to the target edge is obtained, and the average of the initial crack probabilities of the target edge and its matching edge is taken as the crack probability of the target edge. It should be noted that if the pipe inner wall image in this frame is the first frame of the pipe inner wall image, then in the above processing, the matching feature point pairs of the edges are not obtained, and all target feature points are directly taken as reference feature points. At the same time, the target edge is not obtained, and the initial crack probability is directly taken as the crack probability.

[0032] It should be noted that by matching the edges of adjacent frames of pipe inner wall images, since adjacent frames of pipe inner wall images are actually the same area analyzed from two perspectives, the average crack probability is obtained by edge matching from multiple perspectives. This can reduce the occasional interference caused by factors such as lighting, thereby improving the accuracy of crack probability acquisition.

[0033] At this point, the probability of cracks at each edge in the inner wall images of each pipe is obtained.

[0034] Step S003: Downsample the pipe inner wall image to obtain multi-scale pipe inner wall images, analyze the difference in the probability of cracks at the same edge in pipe inner wall images of adjacent scales, and obtain the spacing regularity of each edge; based on the gray level difference change of adjacent pixels at the same edge in pipe inner wall images of each scale, obtain the gray level uniformity of each edge; combined with the spacing regularity, obtain the crack authenticity of each edge.

[0035] It should be noted that there is bound to be errors in assessing the probability of an edge line being a crack at a single scale. Regular textures may be photographed as cracks due to the influence of lighting and angle. If the edge where the feature point is located is more likely to be a crack at multiple scales, and the greater the difference in gray values ​​of all pixels on the edge line, the stronger the authenticity of the edge line as a crack.

[0036] Preferably, in one embodiment of the present invention, the method of downsampling the inner wall image of the pipe to obtain multi-scale inner wall images of the pipe, analyzing the difference in the probability of cracks at the same edge in the inner wall images of adjacent scales, and obtaining the regularity of the spacing of each edge includes: For any frame of the pipe inner wall image, downsampling is performed. In this embodiment, a total of 19 downsampling operations are performed. Combined with the original scale of the pipe inner wall image of that frame, a total of 20 scale pipe inner wall images are obtained. For any edge in the pipe inner wall image of that frame, the crack probability of that edge in the pipe inner wall images of each scale is obtained according to the above method. The absolute value of the difference between the crack probabilities of that edge in the pipe inner wall images of any two adjacent scales is calculated, and the mean of the absolute values ​​of the differences between the crack probabilities of that edge in all two adjacent scales is obtained as the crack variation factor of that edge in multiple scales. The ratio of the crack variation factor to the mean of the crack probability of that edge in the pipe inner wall images of each scale is used as the spacing regularity of that edge. It should be noted that in the ratio calculation process of this embodiment, in order to avoid the denominator being 0 and causing the fraction to be meaningless, a hyperparameter is added to both the denominator and the numerator for calculation. The hyperparameter is described as 0.001.

[0037] It should be noted that the smaller the difference in the probability of cracks at multiple scales, and the greater the overall probability of cracks, the greater the probability that the corresponding edge is a crack at multiple scales. Consequently, the probability that it is regularly distributed with other edges is smaller, the probability that it is a texture is smaller, and the spacing regularity is smaller.

[0038] It should be further explained that each edge contains multiple pixel values. Textures are regular, so the differences in pixel values ​​are small on edges belonging to textures. Cracks, on the other hand, are irregular, and their gray values ​​are uneven, varying in brightness in different edge regions depending on lighting, crack depth, surface dirt, noise, etc. Therefore, by calculating all gray values ​​on edges that are more likely to be cracks at a single scale, calculating the gray value difference between any two adjacent pixel values ​​on that edge line, averaging all gray value differences and taking the derivative, we can obtain the gray value uniformity of the edge at a certain scale. The larger the value, the more uniform the gray value of the edge at that scale.

[0039] Preferably, in one embodiment of the present invention, the gray-level uniformity of each edge is obtained based on the gray-level difference changes of adjacent pixels on the same edge in the pipe inner wall images at various scales. The specific method includes: For any frame of the pipe inner wall image corresponding to a multi-scale pipe inner wall image, for any edge at any scale, obtain the absolute value of the difference between the gray values ​​of any two adjacent pixels on the edge, and use the reciprocal of the mean of the absolute values ​​of the differences between the gray values ​​of all two adjacent pixels on the edge as the gray uniformity factor of the edge at that scale. It should be noted that, in order to avoid the denominator being 0 and the fraction being meaningless, this embodiment adds a hyperparameter to the numerator and denominator respectively for the reciprocal calculation. The hyperparameter is described as 0.1.

[0040] Furthermore, the ratio of the mean to the variance of the gray-level uniformity factor of the edge at all scales is taken as the gray-level uniformity of the edge. It should be noted that in this embodiment, in order to avoid the denominator being 0 and causing the fraction to be meaningless, a hyperparameter is added to both the denominator and the numerator for calculation. The hyperparameter is described as 0.001.

[0041] It should be noted that the smaller the difference in gray values ​​of pixels on the edge at a single scale, the larger the gray uniformity factor. At multiple scales, the smaller the variance between gray uniformity factors and the larger the overall mean, the more uniform the gray values ​​of pixels on the edge are and the less affected by scale transformation, thus the greater the gray uniformity.

[0042] Preferably, in one embodiment of the present invention, the method for obtaining the authenticity of cracks at each edge by combining the regularity of the spacing includes: For any edge in any frame of the pipe inner wall image, the reciprocal of the product of the spacing regularity and grayscale uniformity of the edge is used as the authenticity of the crack on that edge.

[0043] It should be noted that the greater the regularity of spacing and the greater the uniformity of grayscale, the more the edges conform to the texture characteristics. Therefore, by using the reciprocal to make an inverse proportion, the authenticity of the cracks at the edges can be directly reflected, which is the final probability of the cracks.

[0044] At this point, the authenticity of the cracks at each edge in the images of the inner walls of each pipe is obtained.

[0045] Step S004: Based on the authenticity of cracks at each edge in the inner wall images of each pipe, obtain the crack density of each inner wall image of the pipe, and perform pipe defect identification and alarm based on the crack density.

[0046] It should be noted that by comprehensively analyzing the authenticity of cracks at all edges of a single frame image of the inner wall of a pipe, the larger the mean value, the more likely there are cracks in the inner wall of the pipe from the current perspective, requiring timely warning.

[0047] Specifically, for any frame of the pipe inner wall image, the average value of the crack authenticity of all edges in the frame of the pipe inner wall image is used as the crack density of the frame of the pipe inner wall image; a first threshold and a second threshold are preset, in this embodiment the first threshold is described as 0.2 and the second threshold is described as 0.5; if the crack density of the frame of the pipe inner wall image is less than the first threshold, no alarm is issued; if the crack density is greater than or equal to the first threshold but less than the second threshold, the edges in the frame of the pipe inner wall image with crack authenticity greater than the second threshold are obtained and sent to the cloud as crack defects, and the location of the pipe robot corresponding to the frame of the pipe inner wall image is uploaded at the same time; if the crack density is greater than or equal to the second threshold, an alarm is issued, the location of the pipe robot corresponding to the frame of the pipe inner wall image is uploaded, the edges in the frame of the pipe inner wall image with crack authenticity greater than the second threshold are obtained and sent to the cloud as crack defects, thereby realizing defect identification and alarm of the pipe inner wall through the pipe robot.

[0048] This concludes the embodiment.

[0049] Another embodiment of the present invention provides a pipeline robot for pipeline defect identification for pipeline safety maintenance. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the above-described method steps S001 to S004.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A pipeline robot pipeline defect identification method for pipeline safety inspection, characterized in that, The method includes the following steps: The pipeline robot travels through the pipeline and continuously collects images to obtain consecutive frames of images of the pipeline's inner wall. Several feature points are extracted from each frame of the pipe inner wall image by edge detection. Based on the distribution and grayscale representation of the feature points between two adjacent frames of the pipe inner wall image, several matching feature point pairs are obtained in the two adjacent frames of the pipe inner wall image. Based on the distribution of the matching feature point pairs between the corresponding edges of the two adjacent frames of the pipe inner wall image and other edges in the image, the crack probability of each edge in each pipe inner wall image is obtained. Multi-scale images of the pipe's inner wall are obtained by downsampling the pipe's inner wall image. The differences in the probability of cracks at the same edge in pipe inner wall images of adjacent scales are analyzed to obtain the spacing regularity of each edge. Based on the gray-level difference changes of adjacent pixels at the same edge in pipe inner wall images of each scale, the gray-level uniformity of each edge is obtained. Combining the spacing regularity, the authenticity of cracks at each edge is obtained. Based on the authenticity of cracks at each edge in the inner wall images of each pipe, the crack density of each inner wall image is obtained, and pipe defect identification and alarm are performed based on the crack density.

2. The method of claim 1, wherein, The specific method for extracting several feature points from each frame of the pipe inner wall image through edge detection includes: For any frame of the pipe inner wall image, the Canny edge detection algorithm is used to obtain several edges of the pipe inner wall image in that frame; Feature points were detected on each edge using the FAST algorithm, and several feature points were extracted.

3. The method of claim 1, wherein the method further comprises: The specific method for obtaining several matching feature point pairs in two adjacent frames of pipe inner wall images includes: For any edge on any frame of the pipe inner wall image, all feature points on that edge are taken as target feature points. The pipe inner wall image of the next adjacent frame is obtained. For any target feature point, the Hamming distance between the target feature point and all feature points in the next adjacent frame of the pipe inner wall image is obtained. The feature point corresponding to the minimum value among all obtained Hamming distances is taken as the most similar point of the target feature point. The feature point corresponding to the second smallest value among all obtained Hamming distances is taken as the second most similar point of the target feature point. The ratio of the Hamming distance between the target feature point and its most similar point to the Hamming distance between the target feature point and its second most similar point is obtained and used as the matching ratio of the target feature point in the next adjacent frame of the pipe inner wall image. For all target feature points on the edge, obtain the matching ratio of the inner wall image of the pipe in the adjacent previous frame. Arrange all matching ratios from smallest to largest and obtain the difference between adjacent matching ratios. Take the minimum value of the two matching ratios corresponding to the maximum value among all the obtained differences as the matching threshold of the edge. Several target feature points whose matching ratio is less than the matching threshold are used as several reference feature points on the edge. Each reference feature point and its corresponding most similar point are used as a matching feature point pair for the edge.

4. The method according to claim 3, wherein, The specific method for obtaining the probability of cracks at each edge in the inner wall images of each pipe is as follows: For any one frame of pipeline inner wall image, a plurality of reference feature points of each edge in the frame of pipeline inner wall image are obtained; any one edge is taken as a target edge, the shortest distance between all reference feature points on any one other edge except the target edge and the target edge is obtained, the absolute value of the difference between the shortest distances corresponding to any two reference feature points on the other edge is obtained, and the mean value of the absolute values of the differences corresponding to all any two reference feature points on the other edge is taken as the distribution difference factor of the other edge and the target edge; The distribution difference factors of all other edges and the target edge are obtained, the mean value of all distribution difference factors is taken as the distribution difference coefficient of the target edge, linear normalization is performed on the distribution difference coefficients of all edges in the frame of pipeline inner wall image, and the obtained result is taken as the initial crack possibility of each edge. For a plurality of matching feature point pairs of the target edge, the edges in which the most similar points of each reference feature point are located in the adjacent previous frame of pipeline inner wall image are obtained, the edge containing the most similar points is taken as the matching edge of the target edge in the adjacent previous frame of pipeline inner wall image, the initial crack possibility of the matching edge corresponding to the target edge is obtained, and the mean value of the initial crack possibilities of the target edge and the matching edge thereof is taken as the crack possibility of the target edge.

5. The method of claim 1, wherein the method further comprises: The method for obtaining the multi-scale pipeline inner wall image by downsampling the pipeline inner wall image comprises the following specific steps: Any one frame of pipeline inner wall image is downsampled, and the original scale of the frame of pipeline inner wall image is combined to obtain a plurality of scale pipeline inner wall images through multiple times of downsampling.

6. The method of claim 1, wherein the method further comprises: The spacing regularity of each edge is obtained through the following specific method: For any one edge in any one frame of pipeline inner wall image, the crack possibility of the edge in the pipeline inner wall image of each scale is obtained. The absolute value of the difference between the crack possibilities of the edge in any two adjacent scale pipeline inner wall images is calculated, the mean value of the absolute values of the differences between the crack possibilities of the edge corresponding to all adjacent two scales is obtained, and the mean value is taken as the crack change factor of the edge in the multi-scale. The ratio of the crack change factor to the mean value of the crack possibilities of the edge in the pipeline inner wall image of each scale is taken as the spacing regularity of the edge.

7. The method of claim 1, wherein the method further comprises: The gray uniformity of each edge is obtained through the following specific method: For the multi-scale pipeline inner wall image corresponding to any one frame of pipeline inner wall image, for any one edge in any one scale, the absolute value of the difference between the gray values of any two adjacent pixel points on the edge is obtained, the reciprocal of the mean value of the absolute values of the differences between the gray values of all adjacent pixel points on the edge is taken as the gray uniformity factor of the edge in the scale. The ratio of the mean value to the variance of the gray uniformity factors of the edge in all scales is taken as the gray uniformity of the edge.

8. The method of claim 1, wherein, The crack authenticity of each edge is obtained through the following specific method: For any one edge in any one frame of pipeline inner wall image, the reciprocal of the product of the spacing regularity and the gray uniformity of the edge is taken as the crack authenticity of the edge.

9. The method of claim 1, wherein, The crack density of each pipeline inner wall image is obtained through the following specific method: For any frame of the pipe inner wall image, the mean value of the crack authenticity of all edges in the frame of the pipe inner wall image is taken as the crack density of the frame of the pipe inner wall image.

10. A pipeline robot pipeline defect identification device for safe inspection of a pipeline, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor implements the steps of the pipeline robot pipeline defect identification method for pipeline safety overhaul as claimed in any one of claims 1-9 when executing the computer program.

Images