A visual inspection method for track quality of a fully automatic track inspection system

Through pre-trained neural network model and chain code analysis, the shadow area characteristics in orbital quality visual detection were calculated, which solved the problem of cement skin misjudged as foreign body on the track bed, and improved the accuracy of orbital quality detection.

CN118692035BActive Publication Date: 2024-11-05SHANDONG BAOKE INFORMATION TECH CO LTD
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
CN202411169942.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2024-11-05
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

When detecting track quality, the existing fully automatic track patrol system is prone to misjudgment of cement skin as a foreign body on the track bed, resulting in low accuracy of visual inspection of track mass.

Method used

The pre-trained neural network model is used to obtain suspected foreign object regions in the track bed image. Through clustering and boundary chain code sequence analysis, the outline matching degree and shadow region probability of the connecting domain are calculated, and the shadow region stability and proportion of different track bed images are combined. , calculate the probability of foreign objects on the track bed at the position to be detected, and then realize visual detection of orbital mass.

Benefits of technology

The edge features of the connecting domain are represented by the chain code, and the shadow area is accurately screened to distinguish suspected foreign body areas and cement epidermis are missing, improving the accuracy of visual detection of track quality and avoiding misjudgment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118692035B_ABST
    Figure CN118692035B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of image processing technology, and specifically to a visual detection method for track quality of a fully automatic track inspection system, comprising: obtaining a suspected foreign body area in a track bed image of a position to be detected; obtaining the contour matching degree of a connected domain according to the shapes of different connected domains in the suspected foreign body area of ​​the track bed image; obtaining all shadow areas of the suspected foreign body area according to the contour matching degree and brightness of the connected domain; obtaining the shadow area stability and shadow area proportion of every two track bed images according to the contour similarity and number of pixels of the shadow areas in different track bed images; obtaining the probability of a track bed foreign body at the position to be detected according to the shadow area stability and shadow area proportion of a track bed image and other track bed images, and realizing the visual detection of track quality of a fully automatic track inspection system by using the track bed foreign body probability. The present invention aims to solve the problem of shadow interference in foreign body detection of a single track bed image, so as to improve the detection efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a track quality visual detection method for a fully automatic track inspection system. Background Art

[0002] The development of rail transit is an important part of the modern urbanization process. It not only provides efficient urban internal transportation services, but also has a profound impact on urban development, economic growth, environmental protection and the quality of life of residents. For the detection of track quality, an intelligent and fully automatic track inspection system is often used to achieve unattended operation inspection, rapid image judgment and analysis, and intelligent fault identification, so as to improve the efficiency of track inspection operations.

[0003] When the fully automatic track inspection system conducts track quality inspection, it is easy to misjudge the missing cement surface as foreign matter on the track bed because the grayscale and shape contours of the foreign matter on the track bed and the missing cement surface on the track bed surface are similar in the collected track bed image. This will lead to erroneous track quality warnings and reduce the accuracy of visual inspection of track quality. Summary of the invention

[0004] The present invention provides a track quality visual inspection method of a fully automatic track inspection system, so as to solve the problem that the existing cement surface loss is misjudged as a foreign body on the track bed, resulting in low accuracy of track quality visual inspection.

[0005] The track quality visual inspection method of a fully automatic track inspection system of the present invention adopts the following technical solution:

[0006] One embodiment of the present invention provides a track quality visual inspection method of a fully automatic track inspection system, the method comprising the following steps:

[0007] Acquire several roadbed images of locations to be inspected;

[0008] The suspected foreign body area in the track bed image is selected by a pre-trained neural network model; the suspected foreign body area of ​​each track bed image is clustered to obtain the boundary chain code sequence of each connected domain in the suspected foreign body area of ​​each track bed image; according to the similarity of the boundary chain code sequences of every two connected domains in the suspected foreign body area of ​​each track bed image, the contour matching degree of every two connected domains in the suspected foreign body area of ​​each track bed image is obtained; according to the contour matching degree of every two connected domains in the suspected foreign body area of ​​each track bed image and the brightness of all pixels in each connected domain in the suspected foreign body area of ​​each track bed image, all shadow areas of the suspected foreign body area of ​​each track bed image are obtained;

[0009] According to the contour similarity of each shadow area in different roadbed images, the stability of the shadow area of ​​each two roadbed images is obtained; according to the number of pixels in all shadow areas in the suspected foreign body area of ​​each roadbed image, the shadow area ratio of each roadbed image is obtained; according to the stability of the shadow area of ​​each roadbed image and all roadbed images other than other roadbed images, and the shadow area ratio of each roadbed image, the probability of foreign body on the roadbed at the position to be detected is obtained;

[0010] According to the probability of foreign matter on the ballast bed at the position to be inspected, the track quality visual inspection of the fully automatic track inspection system is realized.

[0011] Furthermore, the step of obtaining the boundary chain code sequence of each connected domain in the suspected foreign matter region of each roadbed image includes:

[0012] The number of clusters is preset, and K-means clustering is performed on the suspected foreign body area of ​​the target image to obtain the clustering result image. The connected domain analysis algorithm is used to obtain all connected domains in the clustering image result. The boundary pixel sequence of each connected domain is obtained by the boundary tracking algorithm. The top first pixel point in the boundary pixel sequence of each connected domain is taken as the starting point, and the boundary pixel sequence is traversed clockwise to obtain the 8-neighborhood edge chain code of the boundary pixel sequence of each connected domain, which is recorded as the boundary chain code sequence of each connected domain.

[0013] Furthermore, obtaining the contour matching degree of each two connected domains in the suspected foreign matter region of each roadbed image includes:

[0014] For the i-th connected domain and the j-th connected domain in the suspected foreign body area of ​​the target image, where the length of the boundary chain code sequence of the i-th connected domain is less than that of the boundary chain code sequence of the j-th connected domain, the first element of the boundary chain code sequence of the j-th connected domain is taken as the starting point, a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain is selected, and the result of the cumulative sum of the absolute values ​​of the differences between the boundary chain code sequence of the i-th connected domain and the corresponding elements of the comparison sequence is obtained, and the inversely proportional normalized value of the cumulative sum of the absolute values ​​is taken as the first matching value of the i-th connected domain and the j-th connected domain; the j-th connected domain is taken as the first matching value of the boundary chain code sequence of the j-th connected domain. The second element of the boundary chain code sequence of the connected domain is taken as the starting point, and a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain is selected to obtain the second matching value between the i-th connected domain and the j-th connected domain; and so on, until there is a constant c that makes the elements in the boundary chain code sequence of the j-th connected domain insufficient to select a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain when the c-th element of the boundary chain code sequence of the j-th connected domain is taken as the starting point; the maximum value of the c-1 matching values ​​between the i-th connected domain and the j-th connected domain is recorded as the contour matching degree between the i-th connected domain and the j-th connected domain.

[0015] Furthermore, the step of obtaining all shadow areas of the suspected foreign matter area of ​​each track bed image includes:

[0016] A shadow threshold is preset, and when the probability of the shadow area of ​​the j-th connected domain in the suspected foreign body area of ​​the target image is greater than the shadow threshold, the j-th connected domain in the suspected foreign body area of ​​the target image is recorded as the shadow area.

[0017] Furthermore, the specific method for obtaining the shadow area probability is:

[0018] The probability of the shadow area of ​​the jth connected domain in the suspected foreign body area of ​​the target image The calculation method is:

[0019]

[0020] Where: Indicates the suspected foreign body area of ​​the target image, It represents the contour matching degree between the jth connected domain in the suspected foreign body area of ​​the target image and the main area in the suspected foreign body area of ​​the target image. Represents the number of pixels in the jth connected domain in the suspected foreign body area of ​​the target image, Represents the gray value of the zth pixel in the jth connected domain in the suspected foreign body area of ​​the target image, Represents an exponential function with a natural constant as its base.

[0021] Furthermore, obtaining the stability of the shadow area of ​​every two roadbed images includes:

[0022] For all shadow areas in the w-th roadbed image and the y-th roadbed image, the roadbed image with the smallest number of shadow areas in the w-th roadbed image and the y-th roadbed image is recorded as the matched roadbed image, the shadow areas in the matched roadbed image are recorded as the matched shadow areas, the other roadbed image is recorded as the to-be-matched roadbed image, and the shadow areas in the to-be-matched roadbed image are recorded as the to-be-matched shadow areas;

[0023] Obtain the contour matching degree of all matched shadow areas in the matched roadbed image and each to-be-matched shadow area in the roadbed image to be matched. When all matched shadow areas in the matched roadbed image are matched in the roadbed image to be matched, the average contour matching value of the to-be-matched shadow areas that satisfy the matching is the largest, and each to-be-matched shadow area has only one matched shadow area, the average contour matching value is recorded as the shadow area stability of the w-th roadbed image and the y-th roadbed image.

[0024] Furthermore, obtaining the shadow area ratio of each track bed image includes:

[0025] The ratio of the number of pixels in all shadow areas in the suspected foreign object area of ​​each ballast image to the number of pixels in the suspected foreign object area of ​​the same ballast image is recorded as the shadow area ratio of each ballast image.

[0026] Furthermore, the obtaining of the probability of foreign matter on the ballast bed at the position to be detected includes:

[0027] The average value of the ballast bed foreign object evaluation of all ballast bed images at the position to be detected is recorded as the ballast bed foreign object probability at the position to be detected.

[0028] Furthermore, the specific acquisition method of the ballast foreign body evaluation is:

[0029] The calculation method of the ballast foreign body evaluation of the w-th ballast image is:

[0030]

[0031] Where: represents the average value of the stability of the shadow area of ​​the w-th trackbed image and all trackbed images except the w-th trackbed image, Indicates the number of roadbed images, represents the number of roadbed images except the w-th roadbed image, represents the shadow area ratio of the hth track bed image except the wth track bed image, represents the shadow area ratio of the w-th trackbed image, represents the absolute value function, Represents the Sigmoid function.

[0032] Furthermore, the track quality visual inspection of the fully automatic track inspection system is realized according to the probability of foreign matter on the track bed at the position to be inspected, including:

[0033] A foreign body threshold is preset. When the probability of foreign bodies on the ballast bed at the position to be detected is greater than the foreign body threshold, the fully automatic track inspection system records and issues warning information, and relevant management personnel are sent to deal with the foreign bodies on the ballast bed, thus realizing visual inspection of track quality by the fully automatic track inspection system.

[0034] The beneficial effects of the technical solution of the present invention are as follows: according to the similarity of the boundary chain code sequences of each two connected domains in the suspected foreign object area of ​​each roadbed image, the contour matching degree of each two connected domains in the suspected foreign object area of ​​each roadbed image is obtained, the edge features of each connected domain are represented by chain codes, and the contour matching degree of each two connected domains is accurately reflected by the difference between the edge chain codes of the connected domains; according to the contour matching degree of each two connected domains in the suspected foreign object area of ​​each roadbed image and the brightness of all pixel points in each connected domain in the suspected foreign object area of ​​each roadbed image, all shadow areas of the suspected foreign object area of ​​each roadbed image are obtained; combining the contour matching degree with the brightness of all pixel points in the connected domain, the contour matching degree of each two connected domains in the suspected foreign object area of ​​each roadbed image is obtained. Brightness reflects the probability that each connected domain is a shadow area, and then the shadow area is screened; according to the contour similarity of each shadow area in different roadbed images, the stability of the shadow area of ​​each two roadbed images is obtained, and the shadow area obtained is compared with different roadbed images to ensure that the shadow area is more consistent with the real shadow area; according to the stability of the shadow area of ​​each roadbed image and other roadbed images, as well as the proportion of the shadow area of ​​each roadbed image, the probability of roadbed foreign matter at the position to be detected is obtained, and the roadbed foreign matter and cement surface loss that may appear in the suspected foreign matter area are distinguished by the characteristics of the shadow, so as to avoid misjudgment in the visual inspection of track quality and improve the accuracy of the visual inspection of track quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A flowchart of the steps of a track quality visual inspection method of a fully automatic track inspection system of the present invention;

[0037] Figure 2 Schematic diagram of the result of selecting the suspected foreign body area by the pre-trained neural network;

[0038] Figure 3 This is a partial enlarged schematic diagram of the suspected foreign body area result;

[0039] Figure 4 Schematic diagram of K-means clustering results for suspected foreign body areas. DETAILED DESCRIPTION

[0040] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the track quality visual detection method of a fully automatic track inspection system proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0041] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0042] The specific scheme of the track quality visual inspection method of a fully automatic track inspection system provided by the present invention is described in detail below with reference to the accompanying drawings.

[0043] See also Figure 1 , which shows a flowchart of a method for visually detecting track quality of a fully automatic track inspection system provided by an embodiment of the present invention, the method comprising the following steps:

[0044] Step S001, obtaining a plurality of roadbed images at positions to be detected.

[0045] The purpose of this embodiment is to perform fully automatic track inspection to identify foreign objects on the track bed, so the track bed image is first collected; using the preset camera module installed on the comprehensive inspection vehicle, each camera in the camera module takes an image of the track bed at the position to be inspected in the track, and all the track bed images at the position to be inspected are obtained.

[0046] It should be noted that the comprehensive inspection vehicle refers to a special vehicle with multiple inspection functions, which is usually used for track inspection and evaluation of the operating status and safety performance of the rail transit system. The preset camera modules in this embodiment are 5 linearly arranged camera modules installed on the inspection beam of the comprehensive inspection vehicle. Other camera modules may be set in other embodiments, and this embodiment does not make specific limitations.

[0047] Step S002: Obtain the boundary chain code sequence of each connected domain in the suspected foreign object area of ​​each roadbed image; obtain the contour matching degree of every two connected domains in the suspected foreign object area of ​​each roadbed image; and obtain all shadow areas of the suspected foreign object area of ​​each roadbed image.

[0048] It should be noted that the suspected foreign body area can be obtained by preliminary identification through the pre-trained neural network model. Since the results of the suspected foreign body area obtained by the preliminary identification of the neural network model are not accurate, it may be a foreign body on the roadbed or a missing cement surface, so further analysis is needed; since foreign bodies on the roadbed are usually small stones, parts, garbage and other objects, the foreign bodies on the roadbed will project a shadow near themselves in the roadbed image, and the missing cement surface is a concave area and will not have a shadow projected by itself, so the shadow area in the suspected foreign body area is analyzed; the suspected foreign body area described and operated in this embodiment is a rectangular area with suspected foreign bodies as the main body. In addition to suspected foreign bodies, it also includes the roadbed ground and the shadow area. Several clusters of the suspected foreign body area are obtained by clustering, and the connected domain and edge chain code corresponding to each cluster are obtained for subsequent analysis.

[0049] Specifically, any roadbed image is recorded as the target image, and the roadbed foreign body region of the target image is segmented through the pre-trained neural network model to obtain the suspected foreign body region of the target image. Figure 2 , which shows the pre-trained neural network model preliminarily identifying the target image and selecting the suspected foreign body area. The area framed by the rectangular bounding box indicated by the arrow is the suspected foreign body area. Please refer to Figure 3 , which shows Figure 2 The local magnification effect of the suspected foreign body area result in the figure; the method for obtaining the pre-trained neural network model in this embodiment is: manually selecting the foreign body area of ​​the roadbed from 1000 roadbed foreign body images to obtain a data set, and using the data set to train the convolutional neural network model to obtain a pre-trained neural network model; it should be noted that this embodiment only provides a method for obtaining a pre-trained neural network model, and other embodiments may select other models or methods to obtain the pre-trained neural network model, which is specifically limited in this embodiment;

[0050] Preset the number of clusters and perform K-means clustering on the suspected foreign body area of ​​the target image to obtain the clustering result image. Please refer to Figure 4 , which shows Figure 2 The K-means clustering result image of the suspected foreign body area in the clustering image is obtained by the connected domain analysis algorithm, and the boundary pixel sequence of each connected domain is obtained by the boundary tracking algorithm. The top first pixel point in the boundary pixel sequence of each connected domain is taken as the starting point, and the boundary pixel sequence is traversed clockwise to obtain the 8-neighborhood edge chain code of the boundary pixel sequence of each connected domain, which is recorded as the boundary chain code sequence of each connected domain; it should be noted that the connected domain analysis algorithm, the boundary tracking algorithm, and the 8-neighborhood edge chain code are all well-known technologies, which are not described in detail in this embodiment; the preset number of clusters in this embodiment is 4, and other embodiments can be set to other values, which are not specifically limited in this embodiment.

[0051] It should be further explained that, since the shadow of an object is always attached to the edge of the object, the edge chain code of the shadow area is similar to the edge chain code contour of the suspected foreign object; the difference in the edge chain code contours of each two connected domains is analyzed to obtain the contour matching degree of each two connected domains; the smaller the difference in the edge chain code, the more similar the contours are; the greater the contour matching degree between other connected domains except the connected domain corresponding to the suspected foreign object and the connected domain corresponding to the suspected foreign object, the more likely the other connected domains except the connected domain corresponding to the suspected foreign object are the shadow areas generated by the projection of the suspected foreign object, and the grayscale of each connected domain is further combined to obtain the probability of each connected domain being a shadow area, and the shadow area is screened to obtain.

[0052] Specifically, the boundary chain code sequences of every two connected domains in the suspected foreign body area of ​​the target image are compared to obtain the contour matching degree of every two connected domains. The contour matching degree of every two connected domains is calculated as follows: for the i-th connected domain and the j-th connected domain in the suspected foreign body area of ​​the target image, wherein the length of the boundary chain code sequence of the i-th connected domain is less than the length of the boundary chain code sequence of the j-th connected domain, the first element of the boundary chain code sequence of the j-th connected domain is taken as the starting point, a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain is selected, and the result of the cumulative sum of the absolute values ​​of the differences between the boundary chain code sequence of the i-th connected domain and the corresponding elements of the comparison sequence is obtained, and the inverse ratio of the result of the cumulative sum of the absolute values ​​is calculated. Take the normalized value as the first matching value between the ith connected domain and the jth connected domain; take the second element of the boundary chain code sequence of the jth connected domain as the starting point, select a comparison sequence with the same length as the boundary chain code sequence of the ith connected domain, and obtain the second matching value between the ith connected domain and the jth connected domain; and so on, until there is a constant c so that when the cth element of the boundary chain code sequence of the jth connected domain is taken as the starting point, the elements in the boundary chain code sequence of the jth connected domain are not enough to select a comparison sequence with the same length as the boundary chain code sequence of the ith connected domain; the maximum value of the c-1 matching values ​​between the ith connected domain and the jth connected domain is recorded as the contour matching degree between the ith connected domain and the jth connected domain.

[0053] It should be noted that this embodiment adopts The model is used to present the inverse proportional relationship and normalization processing, where The input of this model can be replaced by other models with the same purpose during implementation, and this embodiment does not make any specific limitation.

[0054] Furthermore, the connected domain with the largest number of pixels in the suspected foreign body area of ​​the target image is recorded as the main area, and the shadow area probability of the jth connected domain in the suspected foreign body area of ​​the target image is The calculation method is:

[0055]

[0056] Where: Indicates the suspected foreign body area of ​​the target image, It represents the contour matching degree between the jth connected domain in the suspected foreign body area of ​​the target image and the main area in the suspected foreign body area of ​​the target image. Represents the number of pixels in the jth connected domain in the suspected foreign body area of ​​the target image, Represents the gray value of the zth pixel in the jth connected domain in the suspected foreign body area of ​​the target image, represents an exponential function with a natural constant as base;

[0057] It should be noted that The larger the value of is, the greater the matching degree between the jth connected domain in the suspected foreign body area of ​​the target image and the contour of the main area in the suspected foreign body area of ​​the target image is, the more similar the contour of the jth connected domain is to the main area, and the greater the probability of the shadow area of ​​the jth connected domain in the suspected foreign body area of ​​the target image is. It represents the mean gray value of all pixels in the jth connected domain in the suspected foreign body area of ​​the target image. The smaller the value, the more likely the jth connected domain is a shadow, and the greater the probability that the jth connected domain in the suspected foreign body area of ​​the target image is a shadow area.

[0058] A shadow threshold is preset. When the probability of the shadow area of ​​the j-th connected domain in the suspected foreign object area of ​​the target image is greater than the shadow threshold, the j-th connected domain in the suspected foreign object area of ​​the target image is recorded as the shadow area. Similarly, all shadow areas in the suspected foreign object areas of all roadbed images are obtained. The shadow threshold is preset to 0.8 in this embodiment, and other embodiments may be set to other values, which are not specifically limited in this embodiment.

[0059] Step S003, obtaining the stability of the shadow area between every two roadbed images; obtaining the shadow area ratio of each roadbed image; and obtaining the probability of foreign matter on the roadbed at the position to be detected.

[0060] It should be noted that in order to avoid inaccurate shadow judgment caused by errors, the similarity of shadow areas in different roadbed images is compared to analyze the possibility that the detected shadow area is a real shadow area; the contour similarity of each shadow area in every two roadbed images is compared; the greater the contour similarity between each roadbed image and the corresponding shadow area in other roadbed images, the more likely the shadow area of ​​the roadbed image is a real shadow area; further, the shadow is used to distinguish between roadbed foreign matter and the absence of cement surface, and the probability of roadbed foreign matter in the corresponding roadbed image is obtained.

[0061] Specifically, for all shadow areas in the w-th roadbed image and the y-th roadbed image, the roadbed image with the smallest number of shadow areas in the w-th roadbed image and the y-th roadbed image is recorded as the matched roadbed image, the shadow areas in the matched roadbed image are recorded as the matched shadow areas, the other roadbed image is recorded as the to-be-matched roadbed image, and the shadow areas in the to-be-matched roadbed image are recorded as the to-be-matched shadow areas;

[0062] Obtaining the contour matching degree of all matched shadow areas in the matched ballast image and each shadow area to be matched in the to-be-matched ballast image; when all matched shadow areas in the matched ballast image are matched in the to-be-matched ballast image, the average contour matching value of the shadow areas to be matched that satisfy the matching is the largest, and each shadow area to be matched has and only matches one matched shadow area, the average contour matching value is recorded as the shadow area stability of the w-th ballast image and the y-th ballast image;

[0063] Furthermore, the ratio of the number of pixels in all shadow areas in the suspected foreign matter area of ​​each ballast image to the number of pixels in the suspected foreign matter area of ​​the same ballast image is recorded as the shadow area ratio of each ballast image.

[0064] The calculation method of the ballast foreign body evaluation of the w-th ballast image is:

[0065]

[0066] Where: represents the average value of the stability of the shadow area of ​​the w-th trackbed image and all trackbed images except the w-th trackbed image, represents the number of roadbed images, in this embodiment, m=5, represents the number of roadbed images except the w-th roadbed image, represents the shadow area ratio of the hth track bed image except the wth track bed image, represents the shadow area ratio of the w-th trackbed image, represents the absolute value function, Represents the Sigmoid function;

[0067] It should be noted that The larger the value of is, the more likely the shadow area in the w-th ballast image is to be a real shadow area, and the greater the ballast foreign matter evaluation of the w-th ballast image is; It represents the reciprocal of the average value of the difference between the proportion of the shadow area of ​​the w-th roadbed image and all roadbed images except the w-th roadbed image. The larger the value is, the more similar the shadow areas of the w-th roadbed image and all roadbed images except the w-th roadbed image are, and the greater the roadbed foreign matter evaluation of the w-th roadbed image is.

[0068] Similarly, the roadbed foreign object evaluation of all roadbed images is obtained, and the average value of the roadbed foreign object evaluation of all roadbed images at the position to be detected is recorded as the roadbed foreign object probability at the position to be detected.

[0069] Step S004: According to the probability of foreign matter on the ballast bed at the position to be detected, the track quality visual inspection of the fully automatic track inspection system is implemented.

[0070] A foreign body threshold is preset. When the probability of a foreign body on the ballast bed at the position to be detected is greater than the foreign body threshold, it indicates that a foreign body on the ballast bed exists at the position to be detected. This is recorded and a warning message is given in the fully automatic track inspection system. Relevant management personnel are sent to handle the foreign body on the ballast bed, thereby realizing visual inspection of track quality by the fully automatic track inspection system. In this embodiment, the foreign body threshold is preset to 0.7. Other embodiments may be set to other values, which are not specifically limited in this embodiment.

[0071] 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 in the protection scope of the present invention.

Claims

1. A visual inspection method for track quality of a fully automatic track inspection system. It is characterized in that The method comprises the following steps: Acquire several roadbed images of locations to be inspected; The suspected foreign body area in the track bed image is selected by a pre-trained neural network model; the suspected foreign body area of ​​each track bed image is clustered to obtain the boundary chain code sequence of each connected domain in the suspected foreign body area of ​​each track bed image; according to the similarity of the boundary chain code sequences of every two connected domains in the suspected foreign body area of ​​each track bed image, the contour matching degree of every two connected domains in the suspected foreign body area of ​​each track bed image is obtained; according to the contour matching degree of every two connected domains in the suspected foreign body area of ​​each track bed image and the brightness of all pixels in each connected domain in the suspected foreign body area of ​​each track bed image, all shadow areas of the suspected foreign body area of ​​each track bed image are obtained; According to the contour similarity of each shadow area in different roadbed images, the stability of the shadow area of ​​each two roadbed images is obtained; according to the number of pixels of all shadow areas in the suspected foreign body area of ​​each roadbed image, the shadow area ratio of each roadbed image is obtained; according to the stability of the shadow area of ​​each roadbed image and other roadbed images, and the shadow area ratio of each roadbed image, the probability of foreign body on the roadbed at the position to be detected is obtained; According to the probability of foreign matter on the ballast bed at the position to be detected, the track quality visual inspection of the fully automatic track inspection system is realized; The step of obtaining the stability of the shadow area of ​​every two roadbed images includes: For all shadow areas in the w-th roadbed image and the y-th roadbed image, the roadbed image with the smallest number of shadow areas in the w-th roadbed image and the y-th roadbed image is recorded as the matched roadbed image, the shadow areas in the matched roadbed image are recorded as the matched shadow areas, the other roadbed image is recorded as the to-be-matched roadbed image, and the shadow areas in the to-be-matched roadbed image are recorded as the to-be-matched shadow areas; Obtain the contour matching degree of all matched shadow areas in the matched roadbed image and each to-be-matched shadow area in the roadbed image to be matched. When all matched shadow areas in the matched roadbed image are matched in the roadbed image to be matched, the average contour matching value of the to-be-matched shadow areas that satisfy the matching is the largest, and each to-be-matched shadow area has only one matched shadow area, the average contour matching value is recorded as the shadow area stability of the w-th roadbed image and the y-th roadbed image.

2. According to claim 1, a track quality visual inspection method for a fully automatic track inspection system, It is characterized in that The step of obtaining the boundary chain code sequence of each connected domain in the suspected foreign body region of each track bed image includes: The number of clusters is preset, and K-means clustering is performed on the suspected foreign body area of ​​the target image to obtain the clustering result image. The connected domain analysis algorithm is used to obtain all connected domains in the clustering image result. The boundary pixel sequence of each connected domain is obtained by the boundary tracking algorithm. The top first pixel point in the boundary pixel sequence of each connected domain is taken as the starting point, and the boundary pixel sequence is traversed clockwise to obtain the 8-neighborhood edge chain code of the boundary pixel sequence of each connected domain, which is recorded as the boundary chain code sequence of each connected domain.

3. According to claim 1, a track quality visual inspection method of a fully automatic track inspection system, It is characterized in that The step of obtaining the contour matching degree of each two connected domains in the suspected foreign body region of each ballast image includes: For the i-th connected domain and the j-th connected domain in the suspected foreign body area of ​​the target image, where the length of the boundary chain code sequence of the i-th connected domain is less than that of the boundary chain code sequence of the j-th connected domain, the first element of the boundary chain code sequence of the j-th connected domain is taken as the starting point, a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain is selected, and the result of the cumulative sum of the absolute values ​​of the differences between the boundary chain code sequence of the i-th connected domain and the corresponding elements of the comparison sequence is obtained, and the inversely proportional normalized value of the cumulative sum of the absolute values ​​is taken as the first matching value of the i-th connected domain and the j-th connected domain; the j-th connected domain is taken as the first matching value of the boundary chain code sequence of the j-th connected domain. The second element of the boundary chain code sequence of the connected domain is taken as the starting point, and a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain is selected to obtain the second matching value between the i-th connected domain and the j-th connected domain; and so on, until there is a constant c that makes the elements in the boundary chain code sequence of the j-th connected domain insufficient to select a comparison sequence of the same length as the boundary chain code sequence of the i-th connected domain when the c-th element of the boundary chain code sequence of the j-th connected domain is taken as the starting point; the maximum value of the c-1 matching values ​​between the i-th connected domain and the j-th connected domain is recorded as the contour matching degree between the i-th connected domain and the j-th connected domain.

4. According to claim 1, a track quality visual inspection method of a fully automatic track inspection system, It is characterized in that The method of obtaining all shadow areas of the suspected foreign matter area of ​​each ballast bed image includes: A shadow threshold is preset, and when the probability of the shadow area of ​​the j-th connected domain in the suspected foreign body area of ​​the target image is greater than the shadow threshold, the j-th connected domain in the suspected foreign body area of ​​the target image is recorded as the shadow area.

5. According to claim 4, a track quality visual inspection method of a fully automatic track inspection system, It is characterized in that The specific method for obtaining the shadow area probability is: The probability of the shadow area of ​​the jth connected domain in the suspected foreign body area of ​​the target image The calculation method is: Where: Indicates the suspected foreign body area of ​​the target image, It represents the contour matching degree between the jth connected domain in the suspected foreign body area of ​​the target image and the main area in the suspected foreign body area of ​​the target image. Represents the number of pixels in the jth connected domain in the suspected foreign body area of ​​the target image, Represents the gray value of the zth pixel in the jth connected domain in the suspected foreign body area of ​​the target image, Represents an exponential function with a natural constant as its base.

6. According to the method for visual inspection of track quality of a fully automatic track inspection system of claim 1, It is characterized in that The obtaining of the shadow area ratio of each track bed image includes: The ratio of the number of pixels in all shadow areas in the suspected foreign object area of ​​each ballast image to the number of pixels in the suspected foreign object area of ​​the same ballast image is recorded as the shadow area ratio of each ballast image.

7. According to claim 1, a track quality visual inspection method of a fully automatic track inspection system, It is characterized in that The obtaining of the probability of foreign matter on the ballast bed at the position to be detected includes: The average value of the ballast foreign matter evaluation of all ballast images at the position to be detected is recorded as the ballast foreign matter probability at the position to be detected; The specific acquisition method of the ballast foreign body evaluation is as follows: The calculation method of the ballast foreign body evaluation of the w-th ballast image is: Where: represents the average value of the stability of the shadow area of ​​the w-th trackbed image and all trackbed images except the w-th trackbed image, Indicates the number of roadbed images, represents the number of roadbed images except the w-th roadbed image, represents the shadow area ratio of the hth track bed image except the wth track bed image, represents the shadow area ratio of the w-th trackbed image, represents the absolute value function, Represents the Sigmoid function.

8. According to the track quality visual inspection method of the fully automatic track inspection system of claim 1, It is characterized in that The method of realizing the track quality visual inspection of the fully automatic track inspection system according to the probability of foreign matter on the track bed at the position to be inspected includes: A foreign body threshold is preset. When the probability of foreign bodies on the ballast bed at the position to be detected is greater than the foreign body threshold, the fully automatic track inspection system records and issues warning information, and relevant management personnel are sent to deal with the foreign bodies on the ballast bed, thus realizing visual inspection of track quality by the fully automatic track inspection system.

Citation Information

Patent Citations

  • Method for detecting foreign matters on steel rail

    CN113160206A

  • Method and device for detecting metal object, terminal equipment and storage medium

    CN115343774A

  • Intelligent detection method for foreign matters on surface of PCB (Printed Circuit Board)

    CN115359043A

  • Nuclear magnetic resonance tumor region extraction method

    CN115797356A

  • Method for detecting pollution discharge in river based on computer vision

    CN116229276A