Turnout identification method, electronic device and storage medium

Abstract

The embodiments of the present application provide a method for identifying a turnout, an electronic device, and a storage medium, wherein the method for identifying a turnout includes: collecting three-dimensional point cloud data corresponding to the turnout in real time by a laser radar disposed at the front end of a data collection vehicle; determining point cloud data corresponding to the rail area according to the rail height and the three-dimensional point cloud data; and determining the position of the rail according to the point cloud data corresponding to the rail area. The method for identifying a turnout, an electronic device, and a storage medium provided in the embodiments of the present application detect and identify the turnout by a laser radar. The detection method is simple and easy to understand, has low requirements on the quality of the point cloud, has a fast computing speed, and can realize efficient positioning of the core components of the turnout.

Description

Technical Field

[0001] The present application relates to railway equipment detection technology, and in particular to a turnout identification method, electronic equipment and storage medium. Background Art

[0002] The realization of automatic identification and detection of turnout components is of great significance to the intelligentization of track maintenance machinery. At present, the operation of turnout tamping vehicles often relies on manual pickaxes, which requires the operator to be highly concentrated and multiple people to cooperate to complete. There is a risk of the tamping pick being mistakenly inserted into the rails or sleepers during the operation. Therefore, the automatic detection of rails and sleepers in the turnout area provides important support for the automatic tamping of turnout tamping vehicles.

[0003] At present, the main methods for detecting turnouts are customized sensor detection, two-dimensional image-based detection, and three-dimensional point cloud-based detection.

[0004] Based on the customized sensor detection method, special sensors are developed for the turnout structure for detection. For example, since most rails and fasteners are made of steel, magnetic induction sensors can be used to detect the positions of rails and fasteners. However, due to the complex structure of the turnout area, the general magnetic induction sensors need to be improved for special areas such as switches and guardrails in order to obtain the accurate positions of rails and sleepers in the turnout area. In addition, the number of sensors required is large and the layout is dense to cover the entire turnout area. This solution is costly, difficult to install and wire, and cumbersome to maintain in the later stage.

[0005] The detection method based on two-dimensional images often uses a camera to collect turnout images and detects rails and sleepers through image processing technology. This method is easy to understand, simple to operate, and requires a small amount of calculated data, but the camera is easily affected by external environmental interference, such as light intensity, rain, snow, fog, etc. The operating environment of railway maintenance vehicles is relatively harsh, the quality of camera-collected data is low, and the image only has two-dimensional information, and the data information is relatively simple, which has a greater impact on subsequent detection.

[0006] Most detection methods based on three-dimensional point clouds use laser radar to scan the surface of the turnout, collect point cloud data, and use various algorithms such as point cloud matching or target detection to detect the position of the rails and sleepers. Laser radar has strong anti-interference ability and the point cloud contains rich information, so point cloud-based detection methods are often better than image-based detection methods. At present, high-precision track inspection vehicles often use laser radar and inertial navigation systems to collect high-precision turnout point cloud data, perform post-processing, and generate turnout detection results. This method is costly, has a large amount of computational data, and takes a long time to post-process. Summary of the invention

[0007] In order to solve one of the above technical defects, a turnout identification method, an electronic device and a storage medium are provided in an embodiment of the present application.

[0008] According to a first aspect of an embodiment of the present application, a turnout identification method is provided, comprising:

[0009] The laser radar installed at the front end of the data collection vehicle collects the three-dimensional point cloud data corresponding to the turnout in real time;

[0010] Determine the point cloud data corresponding to the track area according to the rail height and the three-dimensional point cloud data;

[0011] The position of the rails is determined based on the point cloud data corresponding to the track area.

[0012] According to a second aspect of an embodiment of the present application, a turnout identification device is provided, comprising:

[0013] Memory;

[0014] Processor; and

[0015] Computer programs;

[0016] The computer program is stored in the memory and is configured to be executed by the processor to implement the turnout identification method as described above.

[0017] According to a third aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored; the computer program is executed by a processor to implement the turnout identification method as described above.

[0018] The technical solution provided in the embodiment of the present application collects the three-dimensional point cloud data corresponding to the turnout in real time through the laser radar set at the front end of the data collection vehicle; determines the point cloud data corresponding to the rail area according to the rail height and the three-dimensional point cloud data; determines the position of the rail according to the point cloud data corresponding to the rail area, detects and identifies the turnout through the laser radar, first determines the rail area, and only calculates the point cloud data of the rail area. The invention calculates a small amount of data and is simple to calculate, which effectively solves the problem that the calculation of the point cloud detection method is complex and the position of the turnout components cannot be output in a short time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 A flow chart of a turnout identification method provided in an embodiment of the present application;

[0021] Figure 2 A schematic diagram of slicing three-dimensional point cloud data in the turnout identification method provided in an embodiment of the present application;

[0022] Figure 3 A top view of a rail in a turnout provided in an embodiment of the present application;

[0023] Figure 4 A top view of the rails, sleepers and fasteners in the turnout provided in the embodiment of the present application;

[0024] Figure 5 A flowchart of another turnout identification method provided in an embodiment of the present application;

[0025] Figure 6 A flow chart of obtaining the position of the rail in the turnout identification method provided in an embodiment of the present application;

[0026] Figure 7 A flow chart for obtaining the sleeper position in the turnout identification method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the technical solutions and advantages in the embodiments of the present application more clearly understood, the exemplary embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0028] This embodiment provides a method for identifying a turnout, wherein two or more laser radars are installed at the front end of a data collection vehicle. The data collection vehicle may be a detection vehicle, a tamping vehicle, or other devices that can run in close contact with rails. The selected laser radar should have a larger acquisition angle as much as possible, covering the turnout surface to avoid blind spots. The data collected by the laser radar is the distance to the object.

[0029] like Figure 1 As shown, the turnout identification method provided in this embodiment includes:

[0030] Step 101: collect three-dimensional point cloud data corresponding to the turnout in real time through a laser radar installed at the front end of a data collection vehicle.

[0031] During the detection process, the laser radar is started and the data collection vehicle travels a distance along the track, which includes at least one complete rail sleeper. During the operation of the data collection vehicle, the laser radar collects the three-dimensional point cloud data of the turnout. The data collected by the radar is then stored and analyzed.

[0032] When two laser radars are used, the two laser radars are arranged at intervals along the width direction of the data collection vehicle, and one laser radar can be placed directly above a rail.

[0033] Depending on the detection range, multiple laser radars can be used and installed at the front end of the data collection vehicle to fully cover the detection range and avoid the blind spot at the bottom of the rail, so as to obtain more complete rail bottom data.

[0034] Furthermore, the position of at least two laser radars is calibrated, and the three-dimensional point cloud data collected by at least two laser radars are unified into the same coordinate system. For example, the origin of one of the laser radars is used as the zero point, and the three-dimensional point cloud data collected by the remaining laser radars are all converted to data relative to the zero point. For another example, when two laser radars are used, the midpoint of the two laser radars is used as the zero point, and the three-dimensional point cloud data collected by each laser radar is all converted to data relative to the zero point.

[0035] Assume that the length direction of the rail is the X direction, the length direction of the sleeper is the Y direction, and the vertical direction is the Z direction. A specific solution is: take the starting point of the data collection vehicle as the zero point of the point cloud X coordinate, and the midpoint of the two laser radar installation positions as the YOZ plane zero point. The data collection vehicle runs M meters along the X direction, and the travel offset at each moment is superimposed on the X direction to obtain M meters of turnout point cloud data.

[0036] Furthermore, operations such as noise filtering and hole repair can be performed on the turnout point cloud data to improve the quality of the point cloud data.

[0037] Step 102: Determine the point cloud data corresponding to the rail area according to the rail height and the three-dimensional point cloud data.

[0038] Since the data collection vehicle runs along the rails, the laser radar follows the data collection vehicle, so the distance between the laser radar and the rails is relative. The point cloud data in the Z direction (H -ε, H+L+ε) extracted by straight-through filtering is the point cloud data corresponding to the rail area. Among them, H is the height difference from the zero point of the specified point cloud center to the top of the rail, which is determined according to the installation height of the laser radar. L is the height difference from the top to the bottom of the rail, and ε is the tolerance error.

[0039] Step 103: Determine the position of the rail according to the point cloud data corresponding to the rail area.

[0040] This step can be done in two ways:

[0041] Method 1: Point cloud compression method

[0042] 1.1 Map the point cloud data corresponding to the rail area to multiple two-dimensional planes arranged along the height direction of the rail.

[0043] One implementation method is: Figure 2As shown in the figure, the 3D point cloud data is sliced, and the point cloud in the area (H -ε, H+d) is extracted along the height direction (Z direction) of the rail, and compressed to the plane with Z direction H, that is, the rail top surface. Specifically, the point cloud data in the area (H -ε+(n-1)d, H+nd) is extracted in sequence and compressed to a 2D plane. n is a positive integer, indicating that n planes are obtained after compression, nd≥the distance L between the top and bottom of the rail, d is the unit 2D plane thickness; ε is the tolerance error, and H is the height difference from the zero point of the point cloud center to the rail top.

[0044] A specific method is: extract (H -ε, H+d), (H -ε+d ,H+2d), (H -ε+2d ,H+3d )...(H -ε+(n-1)d ,H+nd ) in sequence and compress them until nd≥L, and the last point cloud is compressed to the plane with Z direction as H+L.

[0045] 1.2 Reconstruct the point cloud data corresponding to each two-dimensional plane to determine the position of the rail.

[0046] The point cloud data corresponding to the compressed n planes are reconstructed separately, and the restored n slices are the three-dimensional representation of the rail. Among them, the output Z direction is the XY boundary of the H plane slice, which is the rail top contour position. The output Z direction is the XY boundary of the H+L plane slice, which is the rail bottom contour position. The output boundary of other compressed slices is the contour position of the current middle area of ​​the rail (rail waist), and the rail position is obtained.

[0047] Method 2: feature extraction method, such as Figure 6 As shown:

[0048] 2.1 Use straight-through filtering to extract the point cloud data corresponding to the rail top surface in the rail height direction.

[0049] Specifically, a straight-through filter is used to extract (H -ε, H + T +ε) in the Z direction, where T is the height of the rail head (the upper area, such as Figure 2 ), the point cloud obtained is all the point clouds of the rail head area, which are mapped to the XOY plane with Z direction as H, that is, the top surface of the rail.

[0050] 2.2 The point cloud data corresponding to the rail top surface are clustered using the point cloud density clustering algorithm to distinguish the point cloud data corresponding to multiple rail top surfaces.

[0051] like Figure 3As shown, the object collected by the laser radar is a turnout, and the turnout area includes four rails, which can be straight or curved. Therefore, there are multiple rail top surfaces in the rail top surface extracted in the above steps. The point cloud density clustering algorithm is used to cluster the point cloud, and multiple rail top surfaces are obtained after clustering. Each rail top surface is processed separately. The point cloud density clustering algorithm can be a commonly used algorithm in the point cloud data processing process.

[0052] 2.3 Use the point cloud edge contour extraction algorithm to calculate the point cloud data corresponding to each rail top surface and determine the contour line of each rail top surface.

[0053] Use the point cloud edge contour extraction algorithm to extract each rail top surface contour point, connect the extracted contour points according to the nearest neighbor points, form the rail top surface edge contour, and generate a binary image. Fill the area surrounded by the edge contour, and the pixel points of the filled area are 0, and save it as the rail top contour image. The pixel points of the area outside the edge contour are 255.

[0054] 2.4 Determine the center line of the rail top surface based on the contour line of the rail top surface.

[0055] One implementation method is to traverse the pixels of the binary contour map corresponding to the rail top contour and extract the center point. The length of the contour map is recorded as X and the width is recorded as Y. Specifically, use X=1,2,3,4,5,…,x to traverse. When X=i, start searching from Y=1. When it satisfies =0, = 255 and , = 0, is denoted as the left edge, where Indicates the pixel value of X=i, Y=n; when = 255, = 0 and = 255, is recorded as the right edge, ) / 2, which is the center point of the rail top when X= i.

[0056] When X= i, the number of times the left and right edges appear should be consistent. There will be multiple rails in the turnout area. In the frog area, when X= i, when the left and right edges appear multiple times, they should be classified into: areas where both the left and right edges appear once, areas where the left and right edges appear twice; areas where the left and right edges appear twice are further divided into areas where the left and right edges appear for the first time, and areas where the left and right edges appear for the second time. The center point of each area is calculated according to the above method and the center point of each area is classified into a separate category. Finally, one or more categories of center points are extracted from the contour map. For example, , ,in Represented as the first type of center point when X = i.

[0057] Then, the optimal line segment is fitted using the center points of each category to obtain the track top equation for each category.

[0058] 2.5 Determine the position of the rails based on the center line of the top surface of each rail.

[0059] The standard rail 3D model sheet (such as Figure 2 ) The center of the rail top is laid out along the rail top equation, and the generated rail position is the position of the turnout rail.

[0060] According to the line survey, the three-dimensional model of the rail switch and the standard three-dimensional model of the rail (such as Figure 2 ) is stored in the storage, and the center of the prefabricated frog rail top (three-dimensional) is placed at the specified position according to the rail top equation, and the other parts are made of the rail three-dimensional model sheet (such as Figure 2 ) The center of the rail top is laid out along the rail top equation, and the position of the rail can be output by traversing along the X direction on the plane with Z direction as H.

[0061] The technical solution provided in this embodiment collects the three-dimensional point cloud data corresponding to the turnout in real time through the laser radar set at the front end of the data collection vehicle; determines the point cloud data corresponding to the rail area according to the rail height and the three-dimensional point cloud data; determines the position of the rail according to the point cloud data corresponding to the rail area, and detects and identifies the turnout through the laser radar, first determines the rail area, and only calculates the point cloud data of the rail area. The amount of calculated data is small and the calculation is simple, which effectively solves the problems of high cost, large amount of calculation, and inability to output the position of the turnout components in a short time in the point cloud detection method.

[0062] Moreover, for the complex turnout structure, the point cloud compression scheme provided in this embodiment does not need to distinguish between components such as wing rails, guard rails, long heart rails, short heart rails, frog heart rails, etc. (collectively referred to as rails in the present invention), that is, there is no need to perform separate inspections on the above-mentioned accessories. The inspection method is simple, easy to understand and implement, and can improve the inspection efficiency.

[0063] Based on the above technical solution, Figure 5 As shown, it also includes identifying the position of the sleeper. First, the three-dimensional point cloud data is mapped into a two-dimensional depth map, and the pixel value represents the height of the point cloud; then the position of the sleeper fastener is determined according to the two-dimensional depth map; and then the position of the sleeper is determined according to the position of the sleeper fastener.

[0064] A specific way is Figure 7 :

[0065] First, the 3D point cloud data is mapped into a 2D depth map, where the pixel value represents the height of the rail. A specific implementation is as follows 3.1.

[0066] 3.1. Map the preprocessed point cloud data into a two-dimensional depth map. Specifically: map the three-dimensional point cloud to the XOY plane, map the Z-axis height coordinate size to the pixel size, and map a point (x+b, y+d, z) in the point cloud to the position (x, y) on the image. The pixel value is ,in, Represents the pixel value of the depth map row x and column y. is the mapping hyperparameter; b, d are the offsets in the X and Y directions when the point cloud is mapped to the depth map.

[0067] The positions of the sleeper fasteners are then determined based on the two-dimensional depth map, as in steps 3.2 and 3.3.

[0068] 3.2 Use image processing technology to train the fastener recognition model. First, create a data set, collect multiple turnout point clouds, intercept the point clouds according to fixed length and width, map them into multiple two-dimensional depth maps, and use image recognition technology to identify fasteners. For example, use LabelImg to mark the fasteners in the depth map, use the YOLO algorithm to train the model, and after adjusting the parameters, obtain a better decoding model as the trained fastener recognition model to determine the fastener position.

[0069] 3.3 Use the above trained model to predict the fastener position; segment and map the point cloud collected in real time to obtain N depth maps with a length of K and a width of W, and send a depth map to the pre-trained fastener recognition model for prediction. The output predicted fastener target position is { },in , It is represented by the center position of the identified z-th fastener on the depth map, Represented as the width and length on the depth map of the zth fastener position.

[0070] The sleeper position is determined according to the sleeper fastener position, as in steps 3.4 and 3.5. Further, the fastener target positions are classified according to the corresponding positions of the fasteners and the sleepers, so as to determine the sleeper position according to the positions of the same type of fasteners.

[0071] 3.4 Classify the predicted fasteners and classify the fasteners on the same sleeper into one category. Figure 4 As shown, the sleeper 2 extends along the Y direction and is located at the bottom of the rail 1. The fasteners 3 are arranged on the sleeper 2, and there are fasteners 3 on both sides of the rail 1, which are used to connect the rail 1 and the sleeper 2. Six fasteners 3 are arranged on one sleeper 2, and these six fasteners 3 are regarded as one type.

[0072] The minimum curve radius and sleeper spacing in the railway are designed in accordance with the specifications. Therefore, when the data collection vehicle moves along the rails and the rails in the turnout area are curved, the offset angle of each sleeper relative to the plane of the vehicle head is less than a fixed value. Therefore, the offset value of the fasteners on the same sleeper in the X direction is controllable. Specifically, traverse from the first predicted fastener position: , when dis is less than the threshold value Dis, it is considered that { } is a category.

[0073] 3.5 Center the fasteners in each category Take them out separately and convert them back to point cloud coordinates , and fit the straight line equation to get the equation: y = ax + w; where a is the slope and w is the intercept. That is, the center position of each sleeper. According to the specified model of the sleeper, a fixed width sleeper is generated. The length of the sleeper can be obtained by the intersection of the sleeper center straight line equation and the above-mentioned leftmost and rightmost rail bottom positions, and the position of the sleeper is output.

[0074] Based on the above technical solutions, this embodiment further provides a turnout identification device, comprising: a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the turnout identification method provided in any of the above contents.

[0075] On the basis of the above technical solutions, this embodiment further provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor to implement the turnout identification method provided by any of the above contents.

[0076] The technical solution provided in this embodiment can realize automatic detection without human intervention, which can greatly improve the work efficiency of railway line inspection, maintenance, etc., and is of great significance for intelligent and unmanned detection in the railway industry.

[0077] Compared with the solution of the track inspection vehicle equipped with a high-precision laser radar (or in conjunction with an inertial navigation system), this application has slightly lower requirements for point cloud quality, lower sensor accuracy requirements, and less data to process. The cost of sensors is greatly reduced. Compared with the solution of directly using point cloud matching and point cloud deep learning, this application has a small amount of calculation and can greatly shorten the operation time.

[0078] Compared with other sensors such as electromagnetic induction and cameras, the technical solution provided in this application uses lidar as the front-end data acquisition device, which has strong adaptability to the environment, is not easily affected by external interference, and has rich point cloud data information, and can capture the surface characteristics of the turnout in multiple dimensions.

[0079] The technical solution provided in this application is easy to migrate and has strong flexibility. The laser radar can be installed on any device that runs in contact with the rails, which can be an operating engineering vehicle, a track inspection vehicle, or any other device that needs to identify turnout rails or sleepers. The output detection results can be flexibly combined and can be applied to a variety of working conditions.

[0080] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program codes. The scheme in the embodiments of the present application may be implemented in various computer languages, for example, C language, VHDL language, Verilog language, object-oriented programming language Java, and literal scripting language JavaScript, etc.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0084] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0085] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0086] In this application, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or can communicate with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0087] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0088] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A turnout identification method, characterized in that: include: The laser radar installed at the front end of the data collection vehicle collects the three-dimensional point cloud data corresponding to the turnout in real time; At least two laser radars are set at the front end of the data collection vehicle; Unify the three-dimensional point cloud data collected by at least two laser radars into the same coordinate system; Determine the point cloud data corresponding to the rail area according to the rail height and the three-dimensional point cloud data; Determine the position of the rails according to the point cloud data corresponding to the rail area; Map the 3D point cloud data into a 2D depth map, where the pixel value represents the height of the point cloud; Determine the position of the sleeper fasteners based on the two-dimensional depth map; Determine the sleeper position based on the sleeper fastener position; Determine the position of the rails based on the point cloud data corresponding to the rail area, including: The point cloud data corresponding to the rail top surface is extracted in the rail height direction by using straight-through filtering; The point cloud data corresponding to the rail top surface are clustered using a point cloud density clustering algorithm to distinguish the point cloud data corresponding to multiple rail top surfaces; The point cloud edge contour extraction algorithm is used to calculate the point cloud data corresponding to the top surface of each rail to determine the contour line of the top surface of each rail; Determine the center line of the rail top surface according to the contour line of the rail top surface; The position of the rails is determined according to the center line of the top surface of each rail.

2. The turnout identification method according to claim 1, characterized in that: Determine the position of the rails based on the point cloud data corresponding to the rail area, including: Mapping the point cloud data corresponding to the rail area to multiple two-dimensional planes arranged along the height direction of the rail; The point cloud data corresponding to each two-dimensional plane are reconstructed to determine the position of the rail.

3. The turnout identification method according to claim 2, characterized in that: Map the point cloud data corresponding to the rail area to multiple two-dimensional planes arranged along the height direction of the rail, including: Along the height direction of the rail, the point cloud data of the area (H -ε+(n-1)d, H+nd) are extracted in sequence and compressed into a two-dimensional plane, where n is a positive integer, nd≥the distance from the top to the bottom of the rail, d is the unit two-dimensional plane thickness; ε is the tolerance error, and H is the height difference from the center zero point of the point cloud to the top of the rail.

4. The turnout identification method according to claim 1, characterized in that: Determine the center line of the rail top surface according to the contour line of the rail top surface, including: Generate a binary image according to the contour line of the rail top surface, and fill the area surrounded by the contour line of the rail top surface; The pixel points on the binary image are traversed to determine the center line of the rail top surface.

5. The turnout identification method according to claim 1, characterized in that: Determine the position of sleeper fasteners based on a 2D depth map, including: Train the fastener recognition model based on preset point cloud data and known fasteners; Map the 3D point cloud data into a 2D depth map and feed it into the trained fastener recognition model for prediction, and output the fastener target position; The turnout identification method further comprises: The fastener target positions are classified according to the corresponding positions of the fasteners and the sleepers, so as to determine the sleeper positions according to the positions of the same type of fasteners, wherein each type of fasteners refers to the fasteners belonging to the same sleeper.

6. A turnout identification device, characterized in that: include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and is configured to be executed by the processor to implement the turnout identification method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that: A computer program is stored thereon; the computer program is executed by a processor to implement the turnout identification method as described in any one of claims 1-5.

Images