Railway equipment detection method, device and storage medium

Abstract

This application provides a method, apparatus, and storage medium for detecting track equipment. The detection method includes: acquiring standard 3D point cloud data of the equipment under normal operation using a linear scan camera; obtaining a standard positional relationship between the linear scan camera and the equipment under inspection based on the standard 3D point cloud data; acquiring current 3D point cloud data of the equipment under inspection; segmenting the current 3D point cloud data according to the standard positional relationship; and determining the state of the equipment under inspection based on the segmented areas. This application's technical solution obtains 3D point cloud data of the track equipment in real time, and then segments, identifies, detects, and analyzes the position of the 3D point cloud data of the track equipment using prior standard size data of the track equipment. This method enables remote real-time detection of abnormal information in the control room, improving the operation and maintenance efficiency of track equipment and reducing the time spent by maintenance personnel working at heights.

Description

Technical Field

[0001] This application relates to the field of rail transit technology, and in particular to a detection method, device and storage medium for rail equipment. Background Technology

[0002] Currently, the inspection of rail vehicles and their surrounding equipment is carried out by sending inspection personnel to the tracks for visual inspection during non-operational hours, which is slow and inefficient. Summary of the Invention

[0003] This application provides a method, apparatus, and storage medium for inspecting track equipment, in order to solve the problem in the prior art that the inspection speed is slow and the efficiency is low when inspectors conduct visual inspections on the track during non-operational hours.

[0004] The first aspect of this application provides a method for detecting track equipment, comprising:

[0005] Standard 3D point cloud data of the device under test in normal condition is acquired by a linear scan camera, and the standard positional relationship between the linear scan camera and the device under test is obtained based on the standard 3D point cloud data.

[0006] The current 3D point cloud data of the device under test is acquired, and the current 3D point cloud data is segmented according to the standard positional relationship. The state of the device under test is determined based on the segmented region.

[0007] A second aspect of this application provides a detection device for track equipment, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in the first aspect of this application.

[0008] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect of this application.

[0009] This application provides a detection method, equipment, and storage medium for track equipment. The detection method includes: acquiring standard 3D point cloud data of the equipment under test in a normal state using a linear scan camera; obtaining a standard positional relationship between the linear scan camera and the equipment under test based on the standard 3D point cloud data; acquiring the current 3D point cloud data of the equipment under test; segmenting the current 3D point cloud data according to the standard positional relationship; and determining the state of the equipment under test based on the segmented areas. This application's technical solution, by installing a linear scan camera and matching left and right wheel speed encoders on a track vehicle, obtains real-time 3D model data of the track equipment. Then, it uses prior standard dimension data of the track equipment to segment the 3D data, perform target recognition and detection, and analyze and determine the position. This method can automatically and in real-time detect abnormal information of track equipment, improving the operation and maintenance efficiency of track equipment. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of a detection method for track equipment according to one embodiment of this application;

[0012] Figure 2 This is a schematic cross-sectional view of the track in one embodiment of this application;

[0013] Figure 3 This is a top view of the track in one embodiment of this application;

[0014] Figure 4 This is a flowchart of step S10 in a detection method for rail vehicles according to an embodiment of this application;

[0015] Figure 5 This is a flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0016] Figure 6 This is another flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0017] Figure 7 This is another flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0018] Figure 8This is another flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0019] Figure 9 This is a flowchart of step S212 in a detection method for rail vehicles according to an embodiment of this application;

[0020] Figure 10 This is another flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0021] Figure 11 This is another flowchart of step S20 in a detection method for rail vehicles according to an embodiment of this application;

[0022] Figure 12 This is a flowchart of step S221 in a detection method for rail vehicles according to an embodiment of this application. Detailed Implementation

[0023] 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, 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.

[0024] Embodiment 1 of this application provides a detection method for track equipment, which can be applied to track vehicles. The method uses a prior model of the equipment under test to perform point cloud segmentation and three-dimensional recognition on its three-dimensional point cloud data to make a judgment. This method is used to analyze whether there are foreign objects left on the running surface, whether the trough is warped, and whether the axle and transponder have moved.

[0025] In an embodiment, such as Figure 1 As shown, a detection method for track equipment is provided, comprising:

[0026] Step S10. Obtain standard 3D point cloud data of the device under test in normal condition using a linear scan camera, and obtain the standard positional relationship between the linear scan camera and the device under test based on the standard 3D point cloud data.

[0027] Among them, the three-dimensional point cloud data of the normal device under inspection obtained by the linear scan camera can be used to determine the standard positional relationship between the linear scan camera and the device under inspection based on the coordinate relationship between the linear scan camera and the three-dimensional point cloud data. Based on the standard positional relationship, the current three-dimensional point cloud data is segmented into normal and abnormal regions. The status of the device under inspection is judged based on the normal and abnormal regions.

[0028] Step S20. Obtain the current 3D point cloud data of the device under test, segment the current 3D point cloud data according to the standard positional relationship, and determine the status of the device under test based on the segmented area.

[0029] Among them, the equipment being tested is equipment located near the track, such as... Figure 2 This is a schematic diagram of the track cross-section, including the roadbed surface 10, track 20, running surface 30, cable tray 40, and pedestrian crossing 50, as shown. Figure 3 This is a top-down view of the track, including the axle counter 60 and the transponder 70. The 3D point cloud data of the inspected equipment is acquired through wheel speed encoders and linear scan cameras mounted on the wheels.

[0030] As one implementation method, such as Figure 4 As shown, step S20 includes:

[0031] Step S101. When the rail vehicle moves, the position and pose of the rail vehicle are obtained through the wheel speed encoder, and the single-beam point cloud data of the device under test is obtained through the linear scan camera.

[0032] Among them, wheel speed encoders are installed on the left and right wheels of the vehicle, and a linear scan camera is installed at the front end to perform line scanning for different devices.

[0033] The angular velocity V of the left wheel can be obtained through the left and right wheel speed encoders. l Angular velocity V of the right wheel t Given the known wheel track L, a local two-dimensional coordinate system is established, with the vehicle's forward direction as the positive x-axis and the vehicle's lateral left direction as the positive y-axis. The trajectory of the rail vehicle is calculated as follows:

[0034] Changes in vehicle angular velocity:

[0035] Average speed of left and right wheels:

[0036] Attitude quantity: θ t+1 =θ t +ω*Δt (3)

[0037] Displacement: x t+1 =x t +v*Δt*cosθ t (4)

[0038] y t+1 =y t +v*Δt*sinθ t (5)

[0039] Where θ0 = 0, x0 = 0, y0 = 0.

[0040] Step S102. Synchronize the working time of the wheel speed encoder and the linear scan camera, and acquire 3D point cloud data based on the pose and single-beam point cloud data.

[0041] Among them, point cloud data of a single line beam range can be obtained through a line scan camera. When the rail vehicle is moving, the movement displacement of the rail vehicle can be obtained based on the above pose and synchronous working time. The precise three-dimensional point cloud data of the scanned object is obtained by stitching together the movement displacement of the rail vehicle and the obtained point cloud data of multiple single line beam ranges.

[0042] The technical effect of this step is that the above process can obtain the local precise position information of the vehicle on a two-dimensional surface. By using the trajectory estimation method of two wheel speed encoders to infer the longitudinal information of the point cloud data of the line scan camera, more accurate three-dimensional point cloud data can be obtained. Especially for the curved area, it can effectively reflect the real model information and avoid the problem of distortion that will occur when the point cloud data is stitched together in the curved area using only one wheel speed encoder.

[0043] As one implementation method, when the device being tested is a cable tray, such as Figure 5 As shown, steps S10 and S20 include:

[0044] Step S201. Obtain standard 3D point cloud data of a normal cable groove using a linear scanning camera, and determine the first standard positional relationship between the linear scanning camera and the cable groove based on the standard 3D point cloud data.

[0045] The process involves obtaining 3D point cloud data of a normal cable groove through scanning, determining the positional relationship between the camera position and the cable groove surface, knowing the width of the cable groove surface, and obtaining the distance from the linear scanning camera to the edge of the cable groove surface based on the installation position of the linear scanning camera. This yields prior information, i.e., the first standard positional relationship. Since the rail vehicle is in motion and there are measurement errors, a tolerance is added to the first standard positional relationship. This tolerance is obtained through prior scanning and is typically the maximum value of the error in the full-line scan.

[0046] Step S202. Obtain the current three-dimensional point cloud data of the groove, and divide the current three-dimensional point cloud data into groove region and non-groove region according to the first standard positional relationship.

[0047] Specifically, the three-dimensional point cloud data of the trough is segmented by the first standard positional relationship, thereby separating the point cloud data of the trough and forming the trough region and the non-trough region.

[0048] Step S203. When the number of point cloud data in the non-groove area reaches a preset number, the groove is determined to be abnormal.

[0049] In theory, there is no point cloud data in the non-groove area. When a certain amount of point cloud data is detected in the non-groove area, the groove is considered abnormal, for example, the groove is bulging.

[0050] The technical advantage of this implementation method is that by using the standard size data of the prior cable trough, the real-time acquired three-dimensional data is divided into cable trough area and non-cable trough area, and the point cloud quantity is analyzed and judged in the non-cable trough area. This method can realize the real-time detection of abnormal information in the cable trough area in the dispatch room, thereby improving the operation and maintenance efficiency of track equipment.

[0051] As one implementation method, when the device being tested is a track running surface, such as Figure 6 As shown, steps S10 and S20 include:

[0052] Step S204. Obtain standard three-dimensional point cloud data of the normal track running surface using a linear scan camera, and determine the second standard positional relationship between the linear scan camera and the track running surface based on the standard three-dimensional point cloud data.

[0053] The process involves obtaining 3D point cloud data of the normal track running surface through scanning, determining the positional relationship between the camera position and the running surface, knowing the width of the normal track running surface, and obtaining the distance from the linear scan camera to the edge of the normal track running surface based on the installation position of the linear scan camera, thus obtaining prior information, i.e., the second standard positional relationship. Since the track vehicle is in motion and there are measurement errors, a tolerance is added to the second standard positional relationship. This tolerance is obtained through prior scanning and is usually the maximum value of the error in the full-line scan.

[0054] Step S205. Obtain the current three-dimensional point cloud data of the track running surface, and divide the current three-dimensional point cloud data into running surface region and non-running surface region according to the second standard positional relationship.

[0055] Specifically, the point cloud data of the traveling surface is segmented using the second standard positional relationship, thereby forming the grooved area and the non-groove area.

[0056] Step S206. When the number of point cloud data in the non-walking surface area reaches a preset number, the walking surface is determined to be abnormal.

[0057] In theory, there is no point cloud data in the non-running surface area. When a certain amount of point cloud data is detected in the non-running surface area, the running surface is considered abnormal, for example, there is a foreign object on the running surface.

[0058] The technical advantage of this implementation method is that by using the standard size data of the normal track running surface in the prior knowledge, the real-time acquired three-dimensional data is divided into running surface area and non-running surface area. The point cloud quantity is analyzed and judged in the non-running surface area. This method can realize the real-time detection of abnormal information in the running surface area in the dispatch room, thereby improving the operation and maintenance efficiency of track equipment.

[0059] As one implementation method, when the device being tested is a shaft counter, such as Figure 7 As shown, step S20 includes:

[0060] Step S207. Obtain standard three-dimensional point cloud data of a normal axle counter using a linear scan camera, and determine the third standard positional relationship between the linear scan camera and the axle counter based on the standard three-dimensional point cloud data and the theoretical installation position of the axle counter.

[0061] The process involves obtaining 3D point cloud data of the normal axle counter through scanning, determining the positional relationship between the camera and the axle counter, knowing the length and width of the axle counter, and obtaining the distance from the linear scan camera to the edge of the normal axle counter based on the installation position of the linear scan camera. This yields prior information, i.e., the third standard positional relationship. Since the rail vehicle is in motion and there are measurement errors, a tolerance is added to the third standard positional relationship. This tolerance is obtained through prior scanning and is usually the maximum value of the error in the full-line scan.

[0062] Step S208. Obtain the current three-dimensional point cloud data of the axle counter, and divide the current three-dimensional point cloud data into the axle counter region and the non-axle counter region according to the third standard positional relationship.

[0063] Specifically, the point cloud data of the axle counter is segmented using the third standard positional relationship, thereby forming the axle counter region and the non-axle counter region.

[0064] Step S209. When the number of point cloud data in the non-axis counter area reaches a preset number, determine that the axis counter is abnormal.

[0065] In theory, there is no point cloud data in the non-axis counting area. When a certain amount of point cloud data is detected in the non-axis counting area, the axis counting is considered abnormal, for example, the axis counting has shifted.

[0066] Furthermore, such as Figure 8 As shown, the scanned 3D point cloud data of the axle counter is segmented into axle counter region and a non-axle counter region based on the third standard positional relationship. This process also includes:

[0067] Step S210. Perform deep learning training and verification based on the axle counter features in the standard 3D point cloud data to obtain the target detection model of the axle counter.

[0068] In this process, the point cloud data within the axle counting area is used for 3D detection and recognition. Since the axle counter is a non-continuous object relative to the groove and running surface, it is necessary to further identify its specific location. The axle counter and its corresponding reference points are manually labeled using the collected normal 3D data. For example, the four corner points of the axle counter or the bolts on the axle counter are used as reference points. The axle counting features are then trained and validated using a convolutional network model to obtain the target detection model.

[0069] Step S211. Based on the target detection model, obtain the positional relationship of at least two reference points of the axle counter in the axle counter area in real time.

[0070] The axle counter has four corner points, and a bolt is placed near each corner point. Any two points can be selected as reference points. The coordinates of any two reference points are detected by the target detection model, and the distance is obtained based on the coordinates of the reference points.

[0071] Step S212. Compare the positional relationship of at least two reference points of the axle counter with the corresponding positional relationship of a normal axle counter, and determine whether the axle counter has shifted based on the comparison results.

[0072] The method involves comparing the distance between two reference points with the normal distance. If the comparison result exceeds the error range, it is determined that the axle counter has shifted.

[0073] As one implementation method, such as Figure 9 As shown, the axle counter includes four corner points, with a bolt near each corner point. Step S212 includes:

[0074] Step S213. Obtain the coordinates of the nearest neighbor corner point and the nearest neighbor bolt using the nearest neighbor search method.

[0075] The training target detection model obtains the coordinates of the corner points of the axle counter and the corresponding four fixed bolts in real time. Due to the existence of annotation error, the position output by the model detection inference is not necessarily the four corner points of the axle counter under the view. Therefore, the coordinates of the nearest corner point are obtained by the nearest neighbor search, and the point cloud data of the nearest neighbor screw corresponding to the corner point is confirmed. The average value is then used to obtain the three-dimensional coordinate information of the bolt.

[0076] Step S214. Obtain the Euclidean distance between the nearest neighbor corner point and the nearest neighbor bolt based on the coordinates of the nearest neighbor corner point and the nearest neighbor bolt.

[0077] Step S215. Compare the Euclidean distance with the corresponding distance of a normal axle counter. If the comparison result is within the redundancy threshold range, the axle counter is determined to be normal. If the comparison result is outside the redundancy threshold range, the axle counter is determined to have shifted.

[0078] The process involves calculating the Euclidean distance between the corner point of the axle counter and the corresponding screw in three dimensions, and then comparing it with the prior normal distance. If the distance exceeds the redundancy threshold, the axle counter is considered to have an offset.

[0079] The technical advantages of this embodiment are as follows: Compared with the trough and the running surface, the axle counter is a non-continuous device. By training the target detection model, the four corner points of the axle counter and other marker points on the axle counter are obtained in real time. The distance between the corner points and the marker points is obtained and compared with the prior distance, thereby determining the abnormal information of the axle counter and improving the operation and maintenance efficiency of the track equipment.

[0080] As one implementation method, when the device being detected is a transponder, such as Figure 10 As shown, steps S10 and S20 include:

[0081] Step S216. Acquire standard three-dimensional point cloud data of a normal transponder using a linear scan camera, and determine the fourth standard positional relationship between the linear scan camera and the transponder based on the standard three-dimensional point cloud data and the installation position of the transponder.

[0082] The process involves obtaining 3D point cloud data of a normal transponder through scanning, determining the positional relationship between the camera and the transponder, knowing the length and width of the transponder, and obtaining the distance from the linear scan camera to the vicinity of the normal transponder based on the installation position of the linear scan camera, thus obtaining prior information, namely the fourth standard positional relationship. Since the rail vehicle is in motion and there are measurement errors, a tolerance is added to the fourth standard positional relationship. This tolerance is obtained through prior scanning and is usually the maximum value of the error in the full line scan.

[0083] Step S217. Obtain the current three-dimensional point cloud data of the transponder, and segment the current three-dimensional point cloud data according to the fourth standard positional relationship to obtain the transponder region and the non-transponder region.

[0084] Specifically, the point cloud data of the axle counter is segmented using the fourth standard positional relationship, thereby forming the transponder region and the non-transponder region.

[0085] Step S218. When the number of point cloud data in the non-responder area reaches a preset number, the transponder is determined to be abnormal.

[0086] In theory, there is no point cloud data in the non-transponder area. When a certain amount of point cloud data is detected in the non-transponder area, the transponder is considered to be abnormal, for example, the transponder has shifted.

[0087] Furthermore, the transponder includes a first side and a second side, the first side being arranged along the length direction of the track, and the second side being perpendicular to the first side; the scanned three-dimensional point cloud data of the transponder is segmented according to a fourth standard positional relationship to obtain a transponder region and a non-transponder region, and the method further includes:

[0088] Within the transponder area, a first marker point is reserved that is positioned opposite the first side and a second marker point that is positioned opposite the second side.

[0089] The transponder is installed at the center of the inner track bottom surface of the Skybus track. Since there is no standard prior position for the transponder, a standard alignment mark needs to be added. Therefore, the corner of the trough or the corner of the track side is used as the first mark point at the first side to determine whether the transponder is offset in the direction perpendicular to the track. A manual mark is set near the second side as the second mark point to determine whether the transponder is offset in the direction parallel to the track.

[0090] Furthermore, the scanned 3D point cloud data of the transponder is segmented according to the fourth standard positional relationship to obtain the transponder region and the non-transponder region. This process also includes:

[0091] Step S219. Perform deep learning training and verification based on the transponder features in the standard 3D point cloud data to obtain the target detection model of the transponder.

[0092] The process involves pre-collecting transponder data on the line, manually labeling the transponder data, and then training and validating it using a target detection convolutional network model to obtain a transponder target detection model.

[0093] Step S220. Based on the target detection model, identify and obtain the positional relationship between the transponder and the two marker points in the transponder area in real time.

[0094] The transponder has four corner points. The first marker point near the first side is either the corner point of the groove or the corner point of the track side. The second marker point near the second side is a manual marker point. The coordinates of the corner points of the transponder and the coordinates of the two marker points are detected by the target detection model, and the distance is obtained based on the coordinates of the marker points.

[0095] Step S221. Compare the obtained positional relationship between the transponder and the two marker points with the corresponding positional relationship in a normal transponder, and determine whether the transponder has shifted based on the comparison result.

[0096] Furthermore, such as Figure 11 As shown, step S221 includes:

[0097] Step S222. Obtain the coordinates of the nearest neighbor corner point using the nearest neighbor search method. Obtain the horizontal Euclidean distance based on the coordinates of the nearest neighbor corner point and the coordinates of the first marker point. Obtain the vertical Euclidean distance based on the coordinates of the nearest neighbor corner point and the coordinates of the second marker point.

[0098] Step S223. Compare the lateral Euclidean distance with the corresponding normal distance in the transponder. When the comparison result is within the redundancy threshold range, the transponder is determined to be normal. When the comparison result is outside the redundancy threshold range, the transponder is determined to have lateral offset.

[0099] Step S224. Compare the longitudinal Euclidean distance with the corresponding normal distance in the transponder. When the comparison result is within the redundancy threshold range, the transponder is determined to be normal. When the comparison result is outside the redundancy threshold range, the transponder is determined to have experienced longitudinal offset.

[0100] The process involves performing 3D point cloud target recognition on the segmented transponder area data to identify the positions of the four corner points of the transponder. A nearest neighbor search is then performed on the 3D point cloud data in a planar dimension to find the 3D coordinate information of the four corner points of the transponder. The distances between the corner points and the first and second marker points are also obtained. Through the above process, it is possible to detect whether the difference between the transponder position and the theoretical value exceeds the redundancy range. If it exceeds the redundancy range, it is determined to be a lateral or longitudinal offset, and the information is reported to the maintenance and support center for processing.

[0101] The technical advantages of this embodiment are as follows: Compared with the trough and the running surface, the transponder is a non-continuous device. Compared with the axle counter, the transponder needs to be marked. By training the target detection model, the four corner points of the transponder and other marked points near the transponder are obtained in real time. The distance between the corner points and the marked points is obtained and compared with the prior distance, thereby determining the abnormal information of the transponder and improving the operation and maintenance efficiency of the track equipment.

[0102] This application provides a detection method, computer equipment, and storage medium for track equipment. The detection method includes: acquiring three-dimensional point cloud data of the equipment under test; acquiring the standard positional relationship between the scanning camera and the equipment under test; segmenting the three-dimensional point cloud data according to the standard positional relationship; and determining the state of the equipment under test based on the segmented areas. This application's technical solution, by installing a line-scanning camera and matching left and right wheel speed encoders on the track vehicle, obtains real-time three-dimensional model data of the track equipment. Then, it uses prior standard dimension data of the track equipment to segment the three-dimensional data, perform target recognition and detection, and analyze and determine the position. This method enables real-time detection of abnormal information in the dispatch room, improving the operation and maintenance efficiency of track equipment and reducing the time maintenance personnel spend working at heights.

[0103] In one embodiment, a detection device for track equipment is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the detection method described in the above embodiment.

[0104] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the detection method described in the above embodiment.

[0105] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0107] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for detecting track equipment, characterized in that, include: Standard 3D point cloud data of the device under test in normal condition is acquired by a linear scan camera, and the standard positional relationship between the linear scan camera and the device under test is obtained based on the standard 3D point cloud data; the 3D point cloud data of the device under test is acquired by a wheel speed encoder and a linear scan camera installed on the wheel. Acquire the current 3D point cloud data of the device under test, segment the current 3D point cloud data into points according to the standard positional relationship, and determine the state of the device under test based on the segmented area; The steps of acquiring the current 3D point cloud data of the device under test, segmenting the current 3D point cloud data according to the standard positional relationship, and determining the state of the device under test based on the segmented regions include: The current 3D point cloud data is segmented into normal and abnormal regions based on the standard positional relationship. When the number of point cloud data in the abnormal region reaches a preset number, the device being tested is determined to be abnormal.

2. The detection method as described in claim 1, characterized in that, The acquisition of the current 3D point cloud data of the device under test includes: When the rail vehicle moves, the wheel speed encoder and the linear scan camera on the track vehicle work synchronously. The wheel speed encoder obtains the position and pose of the rail vehicle, and the linear scan camera obtains the single-beam point cloud data of the device being inspected. The current 3D point cloud data is obtained based on the pose value and the single-beam point cloud data.

3. The detection method as described in claim 1, characterized in that, When the device under test is a cable tray, the step of acquiring standard 3D point cloud data of the device under test in its normal state using a linear scan camera, and obtaining the standard positional relationship between the linear scan camera and the device under test based on the standard 3D point cloud data, includes: Standard three-dimensional point cloud data of a normal wire groove is obtained by a linear scan camera, and a first standard positional relationship between the linear scan camera and the wire groove is determined based on the standard three-dimensional point cloud data. The process of acquiring the current 3D point cloud data of the device under test, segmenting the current 3D point cloud data according to the standard positional relationship, and determining the state of the device under test based on the segmented regions includes: Obtain the current three-dimensional point cloud data of the trough, and segment the current three-dimensional point cloud data according to the first standard positional relationship to obtain the trough region and the non-trough region; When the number of point cloud data in the non-groove area reaches a preset quantity, the groove is determined to be abnormal.

4. The detection method as described in claim 1, characterized in that, When the device under test is a track running surface, the step of acquiring standard 3D point cloud data of the device under test in its normal state using a linear scan camera, and obtaining the standard positional relationship between the linear scan camera and the device under test based on the standard 3D point cloud data, includes: Standard three-dimensional point cloud data of the normal track running surface is acquired by a linear scan camera, and a second standard positional relationship between the linear scan camera and the track running surface is determined based on the standard three-dimensional point cloud data. The process of acquiring the current 3D point cloud data of the device under test, segmenting the current 3D point cloud data according to the standard positional relationship, and determining the state of the device under test based on the segmented regions includes: Obtain the current three-dimensional point cloud data of the track running surface, and segment the current three-dimensional point cloud data into running surface region and non-running surface region according to the second standard positional relationship; When the number of point cloud data in the non-travel surface region reaches a preset amount, the track travel surface is determined to be abnormal.

5. The detection method as described in claim 1, characterized in that, When the device under test is an axis counter, the step of acquiring standard 3D point cloud data of the device under test in a normal state using a linear scan camera, and obtaining the standard positional relationship between the linear scan camera and the device under test based on the standard 3D point cloud data, includes: The standard three-dimensional point cloud data of a normal axle counter is acquired by a linear scan camera, and the third standard positional relationship between the linear scan camera and the axle counter is determined based on the standard three-dimensional point cloud data and the theoretical installation position of the axle counter. The process of acquiring the current 3D point cloud data of the device under test, segmenting the current 3D point cloud data according to the standard positional relationship, and determining the state of the device under test based on the segmented regions includes: Obtain the current three-dimensional point cloud data of the axle counter, and segment the current three-dimensional point cloud data according to the third standard positional relationship to obtain the axle counter region and the non-axle counter region; When the point cloud data in the non-axis counter area reaches a preset number, the axis counter is determined to be abnormal.

6. The detection method as described in claim 5, characterized in that, The step of segmenting the current 3D point cloud data according to the third standard positional relationship to obtain the axle counter region and the non-axle counter region further includes: Deep learning training and validation are performed based on the axle counter features in the standard 3D point cloud data to obtain the target detection model of the axle counter. The positional relationship of at least two reference points of the axle counter in the axle counter region is obtained in real time based on the target detection model. The positional relationship of at least two reference points of the axle counter is compared with the corresponding positional relationship of a normal axle counter, and the axle counter is determined to have shifted based on the comparison results.

7. The detection method as described in claim 6, characterized in that, The axle counter includes four corner points, with a bolt near each corner point. The positional relationship of at least two reference points of the axle counter is compared with the corresponding positional relationship of a normal axle counter. Based on the comparison result, it is determined whether the axle counter has shifted, including: The coordinates of the nearest neighbor corner point and the nearest neighbor bolt are obtained by the nearest neighbor search method; The Euclidean distance between the nearest neighbor corner point and the nearest neighbor bolt is obtained based on the coordinates of the nearest neighbor corner point and the nearest neighbor bolt. The Euclidean distance is compared with the corresponding distance of a normal axle counter. When the comparison result is within the redundancy threshold range, the axle counter is determined to be normal. When the comparison result is outside the redundancy threshold range, the axle counter is determined to have deviated.

8. The detection method as described in claim 1, characterized in that, When the device under test is a transponder, the step of acquiring standard 3D point cloud data of the device under test in its normal state using a linear scan camera, and obtaining the standard positional relationship between the linear scan camera and the device under test based on the standard 3D point cloud data, includes: Standard three-dimensional point cloud data of a normal transponder is acquired by a linear scan camera, and a fourth standard positional relationship between the linear scan camera and the transponder is determined based on the standard three-dimensional point cloud data and the installation position of the transponder. The process of acquiring the current 3D point cloud data of the device under test, segmenting the current 3D point cloud data according to the standard positional relationship, and determining the state of the device under test based on the segmented regions includes: Obtain the current three-dimensional point cloud data of the transponder, and segment the current three-dimensional point cloud data according to the fourth standard positional relationship to obtain the transponder region and the non-transponder region; When the point cloud data in the non-responder area reaches a preset number, the transponder is determined to be abnormal.

9. The detection method as described in claim 8, characterized in that, The transponder includes a first side and a second side, the first side being arranged along the length of the track, and the second side being perpendicular to the first side; The step of segmenting the current 3D point cloud data according to the fourth standard positional relationship to obtain transponder regions and non-transponder regions also includes: Within the transponder area, a first marker point is reserved that is positioned opposite the first side and a second marker point that is positioned opposite the second side.

10. The detection method as described in claim 9, characterized in that, The step of segmenting the current 3D point cloud data according to the fourth standard positional relationship to obtain transponder regions and non-transponder regions further includes: Deep learning training and validation are performed based on the transponder features in the standard 3D point cloud data to obtain a target detection model for the transponder. The target detection model is used to identify the positional relationship between the transponder and two marker points in the transponder region in real time. The obtained positional relationship between the transponder and the two marker points is compared with the corresponding positional relationship in a normal transponder, and the transponder is determined to have shifted based on the comparison results.

11. The detection method as described in claim 10, characterized in that, The step of comparing the obtained positional relationship between the transponder and the two marker points with the corresponding positional relationship in a normal transponder, and determining whether the transponder has shifted based on the comparison result, includes: The coordinates of the nearest neighbor corner point are obtained by the nearest neighbor search method. The horizontal Euclidean distance is obtained based on the coordinates of the nearest neighbor corner point and the coordinates of the first marker point. The vertical Euclidean distance is obtained based on the coordinates of the nearest neighbor corner point and the coordinates of the second marker point. The transponder is compared with the corresponding normal distance in the transponder. If the comparison result is within the redundancy threshold range, the transponder is determined to be normal. If the comparison result is outside the redundancy threshold range, the transponder is determined to have lateral offset. The longitudinal Euclidean distance is compared with the corresponding normal distance in the transponder. If the comparison result is within the redundancy threshold range, the transponder is determined to be normal. If the comparison result is outside the redundancy threshold range, the transponder is determined to have experienced longitudinal offset.

12. A detection device for track equipment, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 11.

13. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 11.

Images