A detection method for a vehicle-mounted track inspection system

Abstract

This invention discloses a detection method for an onboard track inspection system, relating to the field of track technology. The method involves setting an inspection starting point on the track to be inspected and pre-collecting complete track images. Multiple fastener images are generated by cropping the images using fasteners as the smallest unit, thus obtaining historical track data. This historical track data is simultaneously deployed on a cloud server and a local inspection side. After the onboard track inspection system acquires real-time track images, it matches the inspection starting point and similarly crops the images into multiple real-time fastener images. Each real-time fastener image is then divided into regions. The local inspection side's historical track data is used to perform pixel-level comparisons and filter abnormal region images in different regions, generating incremental data which is then transmitted to the cloud server. The historical track data is then used to perform refined detection on the abnormal region images, outputting the defect detection results. This invention solves the problems of insufficient local detection performance, data accumulation, and detection delays under high-speed operation of electric trains, enabling real-time track defect inspection.

Description

Technical Field

[0001] This invention relates to the field of track inspection technology, and in particular to a detection method for a vehicle-mounted track inspection system. Background Technology

[0002] Existing intelligent inspection systems for urban rail transit rely heavily on inspection robots. However, these robots can only be used during off-peak hours, leading to delays and reduced utilization in real-time defect detection. To address this, a system has been developed that mounts inspection equipment onto operating electric trains to achieve real-time track inspection. Compared to robots, this onboard system offers faster and more accurate defect detection. However, in practical application, it has been found that electric trains travel at extremely high speeds, reaching up to 160 km / h. While ensuring data quality, this results in a large volume of track data being collected per unit time for processing. Due to hardware power consumption and data transmission limitations, the local onboard inspection system's detection performance is insufficient, causing data accumulation and detection delays, failing to achieve real-time detection and offering little benefit for guiding track operation. Therefore, resolving the issue of real-time detection failure caused by the high speed of electric trains is a critical problem that urgently needs to be addressed. Summary of the Invention

[0003] In view of this, this application provides a detection method for an on-board track inspection system to address the shortcomings of the existing technology.

[0004] The first aspect of this application provides a detection method for an on-board track inspection system, including:

[0005] Set an inspection starting point on the line to be inspected, pre-collect a complete track image of the line to be inspected, and cut the complete track image into multiple fastener images with fasteners as the smallest unit to obtain track history data.

[0006] The historical track data is deployed simultaneously on both the cloud service side and the local inspection side.

[0007] Real-time images of the track are collected using an onboard track inspection system;

[0008] Match the location corresponding to the inspection starting point in the real-time track image, and crop the real-time track image into multiple real-time fastener images with fasteners as the smallest unit.

[0009] Each real-time image of a fastener is divided into regions. The historical track data from the local inspection side is called to perform pixel-level comparison of different regions in each real-time image of a fastener. Regions with pixel differences are selected and recorded as abnormal region images.

[0010] The extracted abnormal region images are cropped, and corresponding incremental data is generated based on the cropped abnormal region images;

[0011] The generated incremental data is transmitted to the cloud service side, and the historical track data on the cloud service side is called to perform fine-grained detection on the incremental data, and the final track defect detection result is output.

[0012] In one possible implementation of the first aspect, the complete track image is cropped into multiple fastener images, with each fastener as the smallest unit, to obtain track history data including:

[0013] Match the position corresponding to the inspection starting point in the complete track image, and obtain the two-dimensional texture image in the track image corresponding to the inspection starting point;

[0014] Based on the two-dimensional texture image, the boundary of the fastener body near the inspection starting point is determined, and based on the boundary of the corresponding fastener body, the boundary of the corresponding fastener image is determined, thus completing the cropping of the fastener image at the inspection starting point.

[0015] Based on prior data, the fixed spacing between adjacent fasteners on the track is obtained and converted into the image spacing between adjacent fasteners in the track image;

[0016] Using the fastener image at the inspection starting point as the starting point and the image spacing as the reference, the remaining track images are cropped to obtain multiple fastener images, including the fastener image at the inspection starting point.

[0017] One possible implementation of the first aspect also includes:

[0018] Obtain all fastener images after cropping the complete track image, and number and store all fastener images starting from the fastener image at the inspection start point.

[0019] In one possible implementation of the first aspect, dividing each real-time image of a fastener into regions includes dividing each real-time image of a fastener into a fastener body region, a track bed region, and a rail region.

[0020] One possible implementation of the first aspect also includes:

[0021] Based on prior data, the actual size parameters of the fastener body and rail are pre-collected and combined with the image resolution in the real-time track image to be converted into the image size parameters of the fastener body and rail in the real-time track image.

[0022] Based on the real-time orbital image, obtain the corresponding two-dimensional texture image and three-dimensional depth image;

[0023] Based on the corresponding two-dimensional texture image, the contour features of the fastener body and the rail are identified, and the initial area of ​​the fastener body and the initial area of ​​the rail are determined.

[0024] Based on the corresponding three-dimensional depth image, by setting a depth threshold, the initial area of ​​the fastener body and the initial area of ​​the rail are calibrated to obtain the fastener body area and the rail area.

[0025] In a single real-time image of a fastener, the corresponding track bed area is determined based on the defined fastener body area and the defined rail area.

[0026] In one possible implementation of the first aspect, calling the historical track data from the local inspection side to perform pixel-level comparison of different regions in each real-time image of the fastener includes:

[0027] The historical track data from the local inspection side is called to perform a first-pixel level comparison of the rail area in each real-time image of the fastener;

[0028] And perform a second pixel-level comparison on the fastener body area in each real-time image of the fastener;

[0029] And perform a third-pixel level comparison on the track bed area in each real-time image of the fastener;

[0030] The pixel precision decreases sequentially in the first pixel-level comparison, the second pixel-level comparison, and the third pixel-level comparison.

[0031] In one possible implementation of the first aspect, generating corresponding incremental data based on the cropped abnormal region image includes:

[0032] Based on the cropped abnormal area image, obtain the position coordinates of the abnormal area in the corresponding fastener image, the number of the corresponding fastener image, and the type of the abnormal area;

[0033] The abnormal area image, the location coordinates of the abnormal area, the number of the real-time image of the fastener corresponding to the abnormal area image, and the type of the abnormal area are integrated to generate corresponding incremental data.

[0034] In one possible implementation of the first aspect, refining the incremental data by invoking historical track data from the cloud service includes:

[0035] The incremental data is received and parsed to obtain the abnormal area image, the location coordinates of the abnormal area, the number of the real-time image of the fastener corresponding to the abnormal area image, and the type of the abnormal area;

[0036] Based on the number of the real-time image of the fastener corresponding to the abnormal area image, the fastener image with the same number is retrieved from the historical track data on the cloud service side as the detection image of the abnormal area image.

[0037] In one possible implementation of the first aspect, it further includes: based on the location coordinates of the abnormal region, locating and restoring the corresponding abnormal region image to the corresponding detection image.

[0038] One possible implementation of the first aspect also includes:

[0039] Based on the type of the abnormal region, the corresponding refined detection operator is invoked to perform defect detection on the detection image containing the abnormal region image, and the track defect detection result is output.

[0040] In one possible implementation of the first aspect, it further includes: determining whether there are multiple stations on the line to be inspected; if so, setting an inspection starting point based on the starting and ending positions of each station on the line; if not, manually setting relevant markers at fixed distances as inspection starting points; dividing the line to be inspected into segments, with all inspection starting points set before generating historical track data.

[0041] One possible implementation of the first aspect also includes:

[0042] When collecting real-time track images using the vehicle-mounted track inspection system, the system matches the location corresponding to the inspection starting point in the real-time track image as the starting point. It then matches and aligns each real-time image of a fastener collected before and after the starting point with the corresponding fastener image in the historical track data. If a match fails, the system determines that the currently collected data is abnormal, skips the current fastener, and proceeds to the next fastener's real-time image for matching. If a match is successful, the system sequentially performs matching alignment, incremental detection, incremental upload, and cloud detection.

[0043] The beneficial effects are as follows: This invention discloses a detection method for an onboard track inspection system. First, an inspection starting point is set on the line to be inspected, and a complete track image of the line is pre-acquired. Multiple fastener images are generated by cropping the image using fasteners as the smallest unit, thus obtaining historical track data. This historical track data is simultaneously deployed on both the cloud service side and the local inspection side. After the onboard track inspection system acquires real-time track images, it matches the inspection starting point and similarly crops the image into multiple real-time fastener images. Each real-time fastener image is then divided into regions. The historical track data from the local inspection side is then used to perform pixel-level comparison and filtering of abnormal region images for different regions. Based on the abnormal region images, incremental data is generated and transmitted to the cloud service side. The cloud service side then uses the historical track data to perform refined detection of the abnormal region images and outputs the defect detection results. This invention solves the problems of insufficient local detection performance, data accumulation, and detection delay under high-speed operation of electric trains, achieving real-time track defect inspection. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of a detection method for a vehicle-mounted track inspection system provided in an embodiment of this application. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0048] Example

[0049] In existing technologies, there exists an inspection system that mounts inspection equipment onto an operating electric train to achieve real-time track inspection. Compared to inspection robots, this vehicle-mounted track inspection system can achieve faster and more accurate detection of track defects. However, in practical applications, it has been found that electric trains operate at extremely high speeds, reaching up to 160 km / h. While ensuring data quality, a large amount of track data is collected per unit time for the inspection system to process. Due to limitations in hardware power consumption and data transmission performance, the local vehicle-mounted inspection system's detection performance is insufficient, resulting in data accumulation, detection delays, and failing to achieve real-time detection.

[0050] Therefore, this application provides a detection method for an on-board track inspection system, such as... Figure 1 As shown, it includes:

[0051] Set an inspection starting point on the line to be inspected, pre-collect a complete track image of the line to be inspected, and cut the complete track image into multiple fastener images with fasteners as the smallest unit to obtain track history data.

[0052] The historical track data is deployed simultaneously on both the cloud service side and the local inspection side.

[0053] Real-time images of the track are collected using an onboard track inspection system;

[0054] Match the location corresponding to the inspection starting point in the real-time track image, and crop the real-time track image into multiple real-time fastener images with fasteners as the smallest unit.

[0055] Each real-time image of a fastener is divided into regions. The historical track data from the local inspection side is called to perform pixel-level comparison of different regions in each real-time image of a fastener. Regions with pixel differences are selected and recorded as abnormal region images.

[0056] The extracted abnormal region images are cropped, and corresponding incremental data is generated based on the cropped abnormal region images;

[0057] The generated incremental data is transmitted to the cloud service side, and the historical track data on the cloud service side is called to perform fine-grained detection on the incremental data, and the final track defect detection result is output.

[0058] Before explaining the working principle of this embodiment, a brief description of the existing vehicle-mounted track inspection system is given, including a tram, an imaging module, a communication module, a data acquisition and processing module, a system control module, and a power supply module. The tram serves as the carrier for the inspection system, providing the basis for mobile inspection. The imaging module scans vertically along the track's travel direction to acquire track images containing two-dimensional texture images and three-dimensional depth images. The communication module is used for data transmission. The data acquisition and processing module is used to acquire and process the image information from the imaging module. The system control module is an embedded computing platform that plays a central control role. The power supply module provides power to the entire system. While the vehicle-mounted track inspection system can achieve real-time and accurate detection of track defects, it suffers from insufficient local detection performance, data accumulation, and detection delays under high-speed tram operation. To address these issues, this embodiment adopts a detection scheme combining local incremental detection and cloud-based refined detection, specifically:

[0059] I. Data Preprocessing:

[0060] Based on the characteristics of the line to be inspected, the inspection starting point is set; then, complete track images of the line to be inspected are pre-collected, and the complete track images are cropped into multiple fastener images with fasteners as the smallest unit, which are used as historical track data, and then deployed on the cloud service side and the local inspection side respectively.

[0061] The reason why fasteners are used as the smallest unit for cutting in this embodiment stems from their inherent characteristics in the track structure. Track fasteners are evenly distributed along the rail's extension direction, and the geometric shape of each fastener is highly regular and easily recognizable. The specific steps for cutting the complete track image into multiple fastener images using fasteners as the smallest unit are as follows:

[0062] The inspection starting point is matched to the corresponding position in the complete track image, and the corresponding two-dimensional texture image is obtained. Due to the high recognizability and multiple feature attributes of the fastener itself, the boundary of the fastener body at the inspection starting point is determined using the two-dimensional texture image. After determining the boundary of the fastener body, the boundary is expanded outward by a preset distance to determine the boundary of the fastener image. The boundary of the fastener image is used as the reference size of the corresponding fastener image to complete the cropping of the fastener image at the inspection starting point. After obtaining the fastener image at the inspection starting point, the fixed spacing between adjacent fasteners on the track is obtained through prior data and simultaneously converted into the image spacing in the track image. Then, using the fastener image at the inspection starting point as the starting point and the image spacing as the reference, the remaining track image is cropped, thereby cropping the complete track image into multiple fastener images.

[0063] Furthermore, the fastener images are numbered starting from the inspection start point along the track extension direction. For example, the fastener image at the inspection start point is numbered 0, the first fastener image along the extension direction is numbered 1, and so on, with the nth fastener image along the extension direction being numbered n. Subsequent comparisons, whether local or cloud-based, only require matching the corresponding fastener image numbers to quickly locate historical data at the same physical position, avoiding positional errors. It should be noted that the processing principle for real-time track images is the same as that for historical track images, and will not be elaborated further.

[0064] II. Local incremental detection:

[0065] Real-time track images are acquired through an onboard track inspection system. The images are then cropped into multiple real-time fastener images, with each fastener as the smallest unit. Each fastener image is further divided into regions, and historical track data from the local inspection side is used to perform pixel-level comparisons of different regions within each image. This pixel-level comparison of the fastener images on the local inspection side is necessary to address the conflict between the hardware power consumption limitations of the onboard track system and the need for real-time processing of high-speed acquired data. In this embodiment, based on the functional importance and severity of track component defects, each fastener image is divided into a fastener body region, a rail region, and a track bed region. The division logic is as follows:

[0066] Since the rails and fasteners have fixed dimensional parameters, this embodiment first uses prior data to pre-collect the actual dimensional parameters of the fasteners and rails on the line to be inspected, and converts them into corresponding image dimensional parameters based on the image resolution in the real-time track images. Then, based on the real-time track images, it acquires the corresponding two-dimensional texture images and three-dimensional depth images of the fasteners. The two-dimensional texture images are used to identify the contour features of the fasteners and rails, determining their boundaries and thus defining their initial regions. Next, the three-dimensional depth information in the three-dimensional depth images is used to distinguish the spatial height of different components. This is achieved by setting depth thresholds to calibrate the initial regions of the fasteners and rails (e.g., depth values ​​greater than X correspond to the rail, Y < depth value < X corresponds to the fastener), further ensuring that the region division does not cross component types as much as possible. Using the calibrated rail and fastener regions, the corresponding track bed region is determined in a single real-time fastener image (subtraction processing is performed), thus dividing the single real-time fastener image into a fastener region, a rail region, and a track bed region.

[0067] The system calls upon historical track data from the local inspection side to perform pixel-level comparisons of different areas in each real-time image of the fasteners. Specifically:

[0068] The first pixel-level comparison is performed on the rail area. If a full pixel comparison is used, the rail surface texture is regular, and defects (such as cracks) are mostly manifested as sudden changes in local pixel grayscale or abnormal depth values. To ensure that no defects with a low pixel ratio, such as minor cracks and minor wear, are missed, a full pixel comparison is used. The second pixel-level comparison is performed on the fastener body area. If a key feature pixel comparison is used, the fastener body has a fixed geometric shape, and defects are mostly manifested as missing bolts or fastener deformation. Therefore, a full pixel comparison is not necessary. Only the key feature pixels of the fastener (such as bolt center pixels and fastener edge pixels) are compared with high precision. The third pixel-level comparison is performed on the track bed area. If a block comparison is used, the track bed is composed of gravel, and under normal conditions, the texture is irregularly distributed. Defects are mostly manifested as overall changes in the texture density / color of the blocks. Therefore, a pixel-by-pixel comparison is not necessary. Instead, the track bed area is divided into small blocks, and the average grayscale value or texture density and other statistical characteristics of each block are compared. If the difference between the images of each region and the corresponding historical orbit data exceeds a set threshold (which is used to determine whether there is a difference in pixels), then the corresponding region image is determined to be an abnormal region image, thereby greatly reducing the amount of computation.

[0069] The extracted abnormal region images are cropped, and corresponding incremental data is generated based on the cropped abnormal region images. Specifically, the extracted abnormal regions are cropped, and according to the location and size of the abnormal regions, only a partial image of the abnormal region is included in the cropping area of ​​the current real-time image of the fastener. The cropped abnormal region images, the position coordinates of the abnormal regions in the corresponding real-time image of the fastener, the number of the corresponding fastener image, and the type of the abnormal regions are integrated to generate incremental data. The amount of data is much smaller than the original collected data, thus improving data processing efficiency.

[0070] III. Cloud-based refined detection:

[0071] Cloud-based refined detection is performed by the cloud server. Its core objective is to accurately determine whether diseases have occurred in the anomaly areas identified through a preliminary comparison of incremental and historical data, and to obtain disease parameters. Specifically:

[0072] The cloud service receives incremental data and parses it to extract images, location coordinates, numbers, and region types of abnormal areas.

[0073] Based on the corresponding fastener image number, retrieve the corresponding fastener image from the historical track data on the cloud service side as the detection image;

[0074] Based on the location coordinates of the abnormal region image, the abnormal region image is located and restored to the corresponding position in the detection image, restoring the abnormal region in the real track scene and avoiding misjudgment caused by isolated analysis of a single abnormal region.

[0075] Based on the type of abnormal area in the incremental data, the corresponding refined detection operator is invoked. Since the defect characteristics of different types of track components are different, different detection algorithms are required. For example, fastener detection needs to focus on missing / loose bolts, while rail detection needs to focus on cracks / wear. After invoking different refined detection operators to detect the corresponding abnormal area images, the defect detection results are output (including whether there is a defect, the type of defect, the location of the defect, and the defect parameters).

[0076] Furthermore, if the abnormal area image in the incremental data is the fastener body area, a fastener-specific detection operator (such as a bolt feature matching algorithm) is called to accurately identify whether the number and position of bolts are consistent with historical images; if the abnormal area image is the rail area, a rail detection operator is called to determine whether there are cracks or other defects on the rail surface; if the abnormal area image is the ballast area, a ballast density detection operator is called to determine whether there is abnormal loss of ballast gravel.

[0077] This embodiment achieves real-time, high-precision track inspection by combining cloud monitoring and incremental detection. This separate detection algorithm significantly reduces the computational load on the local side of the electric train, resulting in reduced hardware power consumption. The local inspection side only needs to preprocess the collected data (cutting images by fastener, simple anomaly detection, and extracting abnormal areas to generate incremental data), without processing the full data volume, greatly reducing local data processing workload. The incremental data volume is much smaller than the original collected data volume, reducing transmission pressure and allowing for rapid uploading to the cloud service side. The cloud service side, relying on its stronger computing power, performs refined comparison and detection of incremental data and historical data, quickly outputting defect results and feeding them back to the onboard track system. Ultimately, this achieves real-time, high-precision track inspection under high-speed electric train operation, avoiding data accumulation and detection delays, enabling timely guidance of line operation with the inspection results, and enhancing the practical application value of the inspection.

[0078] This embodiment ensures the accuracy and speed of detection through historical data support, regional comparison, and a hierarchical detection mechanism. Complete historical data of the line is pre-collected and stored according to fastener components. Both local comparison and cloud detection are based on current data and corresponding historical data. Locally, a hierarchical pixel-level comparison (regional differentiated comparison strategy) is employed to extract corresponding abnormal areas. This significantly reduces the amount of data while maintaining high-precision comparison for core areas, avoiding the omission of subtle defects. The cloud uses dedicated refined detection operators based on the type of abnormal area to specifically analyze defect types and parameters, mitigating the problem of insufficient detection accuracy in general algorithms. Ultimately, high-precision defect identification and accurate parameter measurement are achieved, ensuring the reliability of detection results and providing accurate data for track maintenance.

[0079] The process involves determining whether the line to be inspected has multiple stations. If so, an inspection starting point is set based on the start and end points of each station's section. If not, relevant markers are manually set at fixed intervals as inspection starting points. The line to be inspected is then segmented, and all inspection starting points are set before generating historical track data. Examples include steps at station entrances and exits, manually set markers, or, for large railways, manually set markers at fixed intervals (2km) to segment the entire line to be inspected.

[0080] In the process of acquiring real-time track images, there may be instances of missed acquisitions due to hardware failures or environmental influences. This embodiment addresses this by matching the inspection starting point and aligning each fastener image acquired before and after the starting point with the corresponding fastener image in the track's historical data. If a match is not found, the current acquired data is deemed abnormal, and the current fastener is skipped until the next fastener image is matched. If a match is found, the matching alignment, incremental detection, incremental upload, and cloud detection are performed sequentially.

[0081] It should be noted that, based on prior data, the fixed spacing between adjacent fasteners on the track is obtained and converted into image spacing between adjacent fasteners in the track image; and based on prior data, the actual dimensional parameters of the fastener body and rail are pre-collected and combined with the image resolution in the real-time track image to convert into image dimensional parameters of the fastener body and rail in the real-time track image. The prior data refers to the parameters in the track line to be inspected that are clearly defined and fixed by track design, construction specifications, or industry standards. Specifically, the fixed spacing between adjacent fasteners on the track obtained based on prior data refers to the inherent, standardized physical spacing parameters of adjacent fasteners that are followed during actual track fastener installation; the actual dimensional parameters of the fastener body and rail pre-collected based on prior data refer to the standardized physical dimensional parameters of the fastener body and rail used in the track to be inspected, which are clearly specified by track design and industry standards.

[0082] It should be noted that in this embodiment, the types of abnormal regions are classified based on region attributes, with three categories, which perfectly match the pre-classification results of the real-time fastener images during the local incremental detection stage. The fastener body region in the real-time fastener image corresponds to the fastener body region abnormality type, the rail region corresponds to the rail region abnormality type, and the track bed region corresponds to the track bed region abnormality type.

[0083] In some embodiments, the complete track image is cropped into multiple fastener images, with fasteners as the smallest unit, to obtain track history data including:

[0084] Match the position corresponding to the inspection starting point in the complete track image, and obtain the two-dimensional texture image in the track image corresponding to the inspection starting point;

[0085] Based on the two-dimensional texture image, the boundary of the fastener body near the inspection starting point is determined, and based on the boundary of the corresponding fastener body, the boundary of the corresponding fastener image is determined, thus completing the cropping of the fastener image at the inspection starting point.

[0086] Based on prior data, the fixed spacing between adjacent fasteners on the track is obtained and converted into the image spacing between adjacent fasteners in the track image;

[0087] Using the fastener image at the inspection starting point as the starting point and the image spacing as the reference, the remaining track images are cropped to obtain multiple fastener images, including the fastener image at the inspection starting point.

[0088] In some embodiments, it also includes:

[0089] Obtain all fastener images after cropping the complete track image, and number and store all fastener images starting from the fastener image at the inspection start point.

[0090] In some embodiments, dividing each real-time image of a fastener into regions includes dividing each real-time image of a fastener into a fastener body region, a track bed region, and a rail region.

[0091] In some embodiments, it also includes:

[0092] Based on prior data, the actual size parameters of the fastener body and rail are pre-collected and combined with the image resolution in the real-time track image to be converted into the image size parameters of the fastener body and rail in the real-time track image.

[0093] Based on the real-time orbital image, obtain the corresponding two-dimensional texture image and three-dimensional depth image;

[0094] Based on the corresponding two-dimensional texture image, the contour features of the fastener body and the rail are identified, and the initial area of ​​the fastener body and the initial area of ​​the rail are determined.

[0095] Based on the corresponding three-dimensional depth image, by setting a depth threshold, the initial area of ​​the fastener body and the initial area of ​​the rail are calibrated to obtain the fastener body area and the rail area.

[0096] In a single real-time image of a fastener, the corresponding track bed area is determined based on the defined fastener body area and the defined rail area.

[0097] In some embodiments, calling the historical track data from the local inspection side to perform pixel-level comparison of different regions in each real-time image of the fastener includes:

[0098] The historical track data from the local inspection side is called to perform a first-pixel level comparison of the rail area in each real-time image of the fastener;

[0099] And perform a second pixel-level comparison on the fastener body area in each real-time image of the fastener;

[0100] And perform a third-pixel level comparison on the track bed area in each real-time image of the fastener;

[0101] The pixel precision decreases sequentially in the first pixel-level comparison, the second pixel-level comparison, and the third pixel-level comparison.

[0102] In some embodiments, generating corresponding incremental data based on the cropped abnormal region image includes:

[0103] Based on the cropped abnormal area image, obtain the position coordinates of the abnormal area in the corresponding fastener image, the number of the corresponding fastener image, and the type of the abnormal area;

[0104] The abnormal area image, the location coordinates of the abnormal area, the number of the real-time image of the fastener corresponding to the abnormal area image, and the type of the abnormal area are integrated to generate corresponding incremental data.

[0105] In some embodiments, retrieving historical track data from the cloud service side to perform refined analysis of the incremental data includes:

[0106] The incremental data is received and parsed to obtain the abnormal area image, the location coordinates of the abnormal area, the number of the real-time image of the fastener corresponding to the abnormal area image, and the type of the abnormal area;

[0107] Based on the number of the real-time image of the fastener corresponding to the abnormal area image, the fastener image with the same number is retrieved from the historical track data on the cloud service side as the detection image of the abnormal area image.

[0108] In some embodiments, the method further includes: based on the location coordinates of the abnormal region, locating and restoring the corresponding abnormal region image to the corresponding detection image.

[0109] In some embodiments, the method further includes: based on the type of the abnormal region, calling the corresponding refined detection operator to perform disease detection on the detection image containing the abnormal region image, and outputting the track disease detection result.

[0110] In some embodiments, the method further includes: determining whether there are multiple stations on the line to be inspected; if so, setting an inspection starting point based on the start and end positions of each station on the line; if not, manually setting relevant markers at fixed distances as inspection starting points; dividing the line to be inspected into segments, with all inspection starting points set before generating historical track data.

[0111] In some embodiments, it also includes:

[0112] When collecting real-time track images using the vehicle-mounted track inspection system, the system matches the location corresponding to the inspection starting point in the real-time track image as the starting point. It then matches and aligns each real-time image of a fastener collected before and after the starting point with the corresponding fastener image in the historical track data. If a match fails, the system determines that the currently collected data is abnormal, skips the current fastener, and proceeds to the next fastener's real-time image for matching. If a match is successful, the system sequentially performs matching alignment, incremental detection, incremental upload, and cloud detection.

[0113] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computing software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0115] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A detection method of a vehicle-mounted track inspection system, characterized in that, The method comprises the following steps: Setting a starting point of inspection in a to-be-inspected line, pre-acquiring a complete track image of the to-be-inspected line, and cutting the complete track image into a plurality of fastener images in the minimum unit of fastener to obtain track historical data; Deploying the track historical data on a cloud service side and a local inspection side at the same time; Acquiring a real-time track image through a vehicle-mounted track inspection system; Matching a position corresponding to the starting point of inspection in the real-time track image, and cutting the real-time track image into a plurality of fastener real-time images in the minimum unit of fastener; Dividing each fastener real-time image into regions, calling the track historical data of the local inspection side to perform pixel-level comparison on different regions in each fastener real-time image, screening out regions with pixel differences, and recording the regions as abnormal region images; Cutting the extracted abnormal region images, and generating corresponding incremental data based on the cut abnormal region images; Transmitting the generated incremental data to the cloud service side, calling the track historical data of the cloud service side to perform fine detection on the incremental data, and outputting a final track disease detection result; The region division of each fastener real-time image comprises the following steps: Dividing each fastener real-time image into a fastener body region, a track bed region and a steel rail region, specifically: Based on prior data, pre-acquiring actual size parameters of the fastener body and the steel rail, and combining the image resolution in the real-time track image to convert the image size parameters of the fastener body and the steel rail in the real-time track image; Based on the real-time track image, obtaining a corresponding two-dimensional texture image and a three-dimensional depth image; Based on the corresponding two-dimensional texture image, identifying the contour features of the fastener body and the contour features of the steel rail, and determining the initial region of the fastener body and the initial region of the steel rail; Based on the corresponding three-dimensional depth image, calibrating the initial region of the fastener body and the initial region of the steel rail by setting a depth threshold to obtain the fastener body region and the steel rail region; In a single fastener real-time image, based on the determined fastener body region and the determined steel rail region, the corresponding track bed region is determined; The pixel-level comparison of different regions in each fastener real-time image by calling the track historical data of the local inspection side comprises the following steps: Calling the track historical data of the local inspection side to perform first pixel-level comparison on the steel rail region in each fastener real-time image; Performing second pixel-level comparison on the fastener body region in each fastener real-time image; Performing third pixel-level comparison on the track bed region in each fastener real-time image; 2. The detection method of the vehicle-mounted track inspection system according to claim 1, characterized in that, The first pixel-level comparison, the second pixel-level comparison and the third pixel-level comparison correspond to pixel accuracies that decrease in turn. The cutting of the complete track image into a plurality of fastener images in the minimum unit of fastener to obtain track historical data comprises the following steps: Matching a position corresponding to the starting point of inspection in the complete track image, and obtaining a two-dimensional texture image in the track image corresponding to the starting point of inspection; Based on the two-dimensional texture image, determining the boundary of the fastener body near the starting point of inspection, and determining the boundary of the corresponding fastener image based on the boundary of the corresponding fastener body to complete the cutting of the fastener image at the starting point of inspection. Based on prior data, the fixed interval between adjacent fasteners on the track is obtained and converted into the image interval between adjacent fasteners in the track image; Taking the fastener image at the inspection starting point as the starting point and the image interval as the reference, the remaining track images are cut to obtain multiple fastener images including the fastener image at the inspection starting point; All fastener images after cutting of the complete track image are obtained, and all fastener images are numbered and marked based on the fastener image at the inspection starting point.

3. The detection method of the vehicle-mounted track inspection system according to claim 2, characterized in that, Based on the cut abnormal area image, the corresponding incremental data is generated, including: Based on the cut abnormal area image, the position coordinates of the abnormal area in the corresponding fastener image, the number of the corresponding fastener image, and the type of the abnormal area are obtained; The abnormal area image, the position coordinates of the abnormal area, the number of the abnormal area image corresponding to the real-time image of the fastener, and the type of the abnormal area are integrated to generate the corresponding incremental data.

4. The detection method of the vehicle-mounted track inspection system according to claim 3, characterized in that, Calling the track historical data on the cloud service side for fine detection of the incremental data includes: Receiving and analyzing the incremental data to obtain the abnormal area image, the position coordinates of the abnormal area, the number of the abnormal area image corresponding to the real-time image of the fastener, and the type of the abnormal area; Based on the number of the abnormal area image corresponding to the real-time image of the fastener, the fastener image with the same number in the track historical data on the cloud service side is called as the detection image of the abnormal area image.

5. The detection method of the vehicle-mounted track inspection system according to claim 4, characterized in that, Further comprising: Based on the position coordinates of the abnormal area, the corresponding abnormal area image is positioned and restored to the corresponding detection image.

6. The detection method of the vehicle-mounted track inspection system according to claim 5, characterized in that, Further comprising: Based on the type of the abnormal area, the detection image containing the abnormal area image is detected by calling the corresponding fine detection operator, and the track disease detection result is output.

7. The detection method of the vehicle-mounted track inspection system according to claim 1, characterized in that, Further comprising: Judging whether the to-be-inspected line has multiple stations, if yes, setting the inspection starting point according to the interval starting point and terminal position of each station of the line, and if not, setting the relevant marks as the inspection starting point with a fixed distance; The to-be-inspected line is segmented, and all inspection starting points are set before generating the track historical data.

8. The detection method of the vehicle-mounted track inspection system according to claim 7, characterized in that, Further comprising: When collecting the track real-time image by the vehicle-mounted track inspection system, the position corresponding to the inspection starting point in the track real-time image is matched as the starting point, and each fastener real-time image collected before and after the starting point is matched and aligned with the corresponding position fastener image in the track historical data, if not matched successfully, it is determined that the current collected data is abnormal, the current fastener is skipped, and the next fastener real-time image is matched; if matched successfully, the matching and alignment, incremental detection, incremental upload, and cloud detection are sequentially performed in order.

Images