A method for detecting a train obstacle

Abstract

The application provides a train obstacle detection method. The method comprises: collecting full-line point cloud data by a laser radar; calculating full-line limit boundary points offline based on the full-line point cloud data; collecting real-time point cloud data online by the laser radar; and monitoring the obstacle online based on the full-line limit boundary points and point cloud cluster data obtained from the real-time point cloud data. The method can independently detect obstacles in road sections where cameras and millimeter wave radars are prone to monitoring blind spots, such as small-radius curves, and can reduce the difficulty of obstacle judgment and improve the efficiency of online obstacle detection.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of train track transportation, and more particularly, to a train obstacle detection method. BACKGROUND

[0002] With the rapid development of urban rail transit, in China's large, medium and small cities, including subways, light rails, trams and other different types of rail transit systems are widely used. The safety problem of rail transit operation has also been highlighted, especially in the operation area of rail transit lines, the invasion of foreign objects (such as falling rocks, power supply columns, pantographs, pedestrians, etc.) caused by natural or human factors often occurs, which brings great safety hazards to normal operation.

[0003] Because the whole train of rail transit has a large weight and load, it is necessary to detect the existence of obstacles when the distance is far apart to timely execute the corresponding emergency strategy (such as emergency braking). At present, most of the urban rail transit lines still adopt the manned driving mode, and the detection of obstacles mainly relies on the visual judgment of the driver on the road conditions in the front driving area. If the driver himself is not concentrated or in the case of night or bad weather driving, it is easy to miss or misjudge the obstacles, which will damage the safety and quality of the operation.

[0004] In the unmanned driving mode adopted by part of the rail transit, the detection of obstacles mainly relies on cameras and radars, which are the commonly used detection technologies in the field of intelligent transportation. The machine vision detection represented by the camera can obtain the specific category of the detected object, but it is easily affected by the light factor in the environment; the radar detection technology represented by millimeter wave radar and laser radar has small environmental factor interference, but can only obtain the size and distance information of the detected object, so the camera and the radar are often used together to make up for each other. However, the existing obstacle detection method relying on the camera and / or millimeter wave radar is prone to monitoring blind area, and it is difficult to judge the obstacles, which greatly reduces the detection efficiency of the obstacles.

[0005] Therefore, there is a need for a train obstacle detection method that can independently detect obstacles in road sections such as small radius curves where cameras and millimeter wave radars are prone to monitoring blind areas, while reducing the difficulty of obstacle judgment and improving the efficiency of online obstacle detection. SUMMARY

[0006] To solve at least one of the above problems, according to an aspect of the present application, a method for detecting train obstacles is provided, comprising: collecting full-line point cloud data by a laser radar along a full line; calculating full-line limit boundary points offline based on the full-line point cloud data; collecting real-time point cloud data online by the laser radar; and monitoring the obstacles online based on the full-line limit boundary points and point cloud cluster data obtained from the real-time point cloud data.

[0007] In one embodiment, wherein monitoring the obstacles online comprises judging whether boundary points of the point cloud cluster data are within a rectangular surrounding frame of the full-line limit boundary points.

[0008] In one embodiment, wherein when the proportion of the boundary points of the point cloud cluster data falling within the rectangular surrounding frame of the full-line limit boundary points exceeds a threshold value, it is determined that the object represented by the point cloud cluster data is an obstacle.

[0009] In one embodiment, wherein obtaining the point cloud cluster data comprises setting different clustering radii with the laser radar as the origin, and dividing within the clustering radii using the same clustering threshold distance.

[0010] In one embodiment, the method further comprises extracting track point cloud data based on the real-time point cloud data.

[0011] In one embodiment, wherein the track point cloud data is extracted by a principal component analysis algorithm.

[0012] In one embodiment, the method further comprises screening the real-time point cloud data based on a region of interest.

[0013] In one embodiment, wherein the region of interest is selected by a pass-through filtering method.

[0014] In one embodiment, wherein collecting the full-line point cloud data comprises obtaining real-time pose of the train by an inertial measurement unit, thereby converting track lines in the full-line point cloud data from a laser radar coordinate system to a global coordinate system.

[0015] In one embodiment, wherein calculating the full-line limit boundary points offline comprises surrounding a car body limit of the train with a rectangular surrounding frame based on the car body limit, leaving a margin between the rectangular surrounding frame and the car body limit, and nodes of the rectangular surrounding frame are the full-line limit boundary points.

[0016] In one embodiment, wherein calculating the full-line limit boundary points offline comprises obtaining real-time pose of the train by an inertial measurement unit, thereby obtaining the full-line limit boundary points in a global coordinate system.

[0017] In one embodiment, wherein the online collection of the real-time point cloud data comprises using an algorithm of point cloud registration, the real-time point cloud data obtained online is compared with the full-line point cloud data to determine the current position of the train.

[0018] In one embodiment, wherein the laser radar is a 3D multi-line laser radar.

[0019] In one embodiment, the full-line point cloud data comprises three-dimensional coordinates, reflection intensity, and a trajectory line of the train when driving.

[0020] In one embodiment, the laser radar is installed at a front central position of the head of the train.

[0021] According to another aspect of the present application, a non-transitory computer readable medium having computer program instructions stored thereon is provided, wherein the computer program instructions, when executed by one or more processors, implement the above method.

[0022] The method provided by the present application can independently detect obstacles in road sections where cameras and millimeter wave radars are prone to have monitoring blind areas such as small-radius curves, and can reduce the difficulty of obstacle judgment and improve the efficiency of online obstacle detection. BRIEF DESCRIPTION OF DRAWINGS

[0023] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout the figures. The drawings provided in connection with the present application are intended to further explain, by way of non-limiting example, the principles of the present application, and are not intended to limit the scope of the application to matters specifically disclosed herein. In the drawings, like reference numerals designate like elements throughout the several views.

[0024] Figure 1 FIG. 1 shows a comparison of the detection ranges of a camera, a millimeter wave radar, and a laser radar in obstacle detection in a small-radius curve road section;

[0025] Figure 2 FIG. 2 is a flowchart showing a train obstacle detection method according to an embodiment of the present application;

[0026] Figure 3 FIG. 3 is a schematic diagram showing the installation position of a laser radar according to an embodiment of the present application;

[0027] Figure 4 FIG. 4 is a schematic diagram showing a full-line limit boundary point according to an embodiment of the present application;

[0028] Figure 5 FIG. 5 is a schematic diagram showing a point cloud segmentation clustering radius and accuracy according to an embodiment of the present application; and

[0029] Figure 6 This is a schematic diagram illustrating orbital point cloud extraction according to an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention more apparent, exemplary embodiments according to the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein. Based on the embodiments of the present invention described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of the present invention.

[0031] Figure 1 The comparison of the detection range of cameras, millimeter-wave radar, and lidar is shown in obstacle detection on small-radius curved road sections.

[0032] like Figure 1 As shown, in rail transit scenarios, the types of foreign objects intruding into the operating area are relatively random and difficult to predict in advance, so camera detection may miss some. In obstacle detection on small-radius curved sections, the limited horizontal monitoring area of ​​cameras and millimeter-wave radar may also lead to missed obstacles. LiDAR, however, can collect data horizontally in a circular pattern and can play a major detection role in these situations.

[0033] This invention provides a method for obstacle detection in rail transit based on lidar, capable of detecting obstacles in the road ahead during rail transit operation in real time. It can independently detect obstacles on sections where cameras and millimeter-wave radar have blind spots, such as small-radius curves. The calculations utilize the fixed structure of rail transit lines, employing a combination of offline and online computation to reduce the difficulty of obstacle identification and improve the efficiency of online obstacle detection.

[0034] Below, refer to Figure 2 The method for detecting train obstacles according to embodiments of the present invention will be described in detail below.

[0035] like Figure 2 As shown, a method for detecting obstacles on a train includes: collecting point cloud data along the entire line using a lidar; calculating limit points along the entire line offline based on the point cloud data; collecting real-time point cloud data online using the lidar; and monitoring the obstacles online based on the limit points along the entire line and point cloud cluster data obtained from the real-time point cloud data.

[0036] In one embodiment, the lidar is a 3D multi-line lidar. A multi-line lidar refers to a laser rotating rangefinder that simultaneously emits and receives multiple laser beams. Currently, the main types on the market are 4-line, 8-line, 16-line, 32-line, 64-line, and 128-line lidars. Multi-line lidar can identify the height information of objects and acquire 3D scan images of the surrounding environment.

[0037] The train obstacle detection method based on lidar in this invention combines offline calculation of the entire line point cloud map and the entire line limit boundary points with online detection of track obstacles.

[0038] The full-line point cloud map is collected after the train, equipped with a 3D multi-line LiDAR, has traveled the entire line a certain number of times. This full-line point cloud data can be used for positioning during the train's official operation. The full-line limit points are rectangular bounding boxes simplified from the train's body clearance. During online detection, if the real-time point cloud of an object encroaches on the full-line limit point range corresponding to the object's location, it is identified as an obstacle. The full-line limit points are calculated offline from the train's trajectory collected during the full-line point cloud map acquisition process.

[0039] The online detection of track obstacles includes real-time point cloud data acquisition, real-time point cloud data filtering, track point cloud extraction, object point cloud segmentation and clustering, and obstacle judgment, all of which are obtained through online calculations by the 3D multi-line lidar equipped on the train during its operation along the service line.

[0040] Next, combined Figure 3 To illustrate the installation location of the lidar according to an embodiment of the present invention.

[0041] like Figure 3 As shown, in one embodiment, the lidar is installed at the center of the front of the train's head. In a specific implementation, considering simplifying the calculation and processing of the point cloud data obtained from the detection and ensuring the detection range of the lidar, the lidar is preferably installed at the center of the front of the train's head, and at a horizontal angle. The lidar can be placed on a pan-tilt unit for angle adjustment. Considering the possibility of turnarounds in urban rail transit, one lidar can be installed at each of the train's head and tail.

[0042] Below, in conjunction with Figures 4 to 6 The principle of the train obstacle detection method according to an embodiment of the present invention will be explained.

[0043] S1: The laser radar collects the whole line map information along the whole line. The whole line point cloud map is the whole line point cloud data collected after the train equipped with a 3D multi-line laser radar travels a certain number of times along the line served by the train. In an embodiment, the whole line point cloud data includes three-dimensional coordinates, reflection intensity, and the trajectory line of the train when traveling. Specifically, the whole line point cloud data includes the three-dimensional coordinates and reflection intensity of the radar scanning points, mainly to provide positioning information for the train running on the line. In addition, the point cloud map also contains the trajectory line of the train when traveling on the line. The trajectory line is a curve in space, represented as a series of consecutive three-dimensional discrete points, and the point set will be used to calculate the limit boundary point and extract the track point cloud.

[0044] In an embodiment, collecting the whole line point cloud data includes obtaining the real-time pose of the train by an inertial measurement unit, so as to convert the trajectory line in the whole line point cloud data from the laser radar coordinate system to the global coordinate system.

[0045] Specifically, in order to facilitate the conversion of the trajectory line data from the laser radar local coordinate system to the global coordinate system, the train needs to be equipped with an IMU (Inertial Measurement Unit) to obtain the real-time pose of the train, so as to complete the conversion of the trajectory line from the laser radar coordinate system→train coordinate system→global coordinate system. The conversion formula is as follows:

[0046]

[0047] wherein is the coordinate of the trajectory line in the global coordinate system, is the coordinate of the origin of the train coordinate system in the global coordinate system, is the coordinate of the trajectory line in the radar coordinate system, is the coordinate of the origin of the radar coordinate system in the train coordinate system, R B is the rotation matrix of the radar coordinate system to the train coordinate system, determined by the installation pose of the radar on the train, R P is the rotation matrix of the train coordinate system to the global coordinate system, provided by the IMU.

[0048] S2: offline computing the full line limit boundary points. The limit refers to the profile size line that is not allowed to be exceeded, which is prescribed for the train and the buildings and equipment close to the line, in order to ensure the safety of the train running on the railway line and prevent the train from colliding with the buildings and equipment adjacent to the line. The full line limit boundary points are the demarcation criteria for judging whether the object point cloud is an obstacle. In an embodiment, offline computing the full line limit boundary points includes surrounding the car body limit with a rectangular surrounding frame on the basis of the car body limit, a margin is left between the rectangular surrounding frame and the car body limit, and the nodes of the rectangular surrounding frame are the full line limit boundary points.

[0049] Specifically, considering the convenience of calculation and storage, as shown in Figure 4 , a rectangular frame can be used to surround the limit on the basis of the car body limit, and a certain margin can be left between the rectangular surrounding frame and the limit. The larger the margin, the higher the safety, but it will also increase the risk of false positives. The four nodes of the rectangular surrounding frame are the limit boundary points.

[0050] In an embodiment, offline computing the full line limit boundary points includes obtaining the real-time pose of the train by the inertial measurement unit, so as to obtain the full line limit boundary points in the global coordinate system.

[0051] Specifically, after obtaining the track line in the aforementioned global coordinate system through S1, combined with the real-time pose data of the train obtained by the IMU, the full line limit boundary point set in the global coordinate system can be obtained through spatial geometric calculation. Each element in the set corresponds to a point in the aforementioned track line, each element contains four limit boundary points, and each limit boundary point contains a spatial coordinate data. When calculating the limit, the distance H1 from the track center line to the top of the limit, the distance H2 from the track center line to the bottom of the limit, and the distance W from the track center line to the two sides of the limit need to be determined; according to the obtained track line shape, it is divided into multiple segments, each segment can be approximated as a straight segment, the direction vector of the track center line of the segment is estimated, and the node coordinates are calculated one by one from the midpoint of the track center line according to the direction vector and the distance from the point to the straight line.

[0052] Specifically, the track line Track is divided into n segments Track = {track i}(i = 1, 2, …, n), each track i segment can be approximated as a straight segment. track i is a set of space points, that is, for track i containing m points, track i = {P j (x j ,y j ,z j )}(j = 1, 2, …, m). For any track iThe centroid coordinates W(x w ,y w ,z w ) of the track i , the first three principal directions (i.e., the x, y, z direction vectors of the track x p y , p z , and p i , and the dimensions l x , l y , and l z of the track i in these three directions can be calculated by the PCA algorithm. The problem of finding the limit boundary points can be abstracted as finding the coordinates (x2, y2, z2) of point 2 by the coordinates (x1, y1, z1) of point 1, the vector (direction vector) coordinates p d (d x , d y , d z ) of point 2 and point 1, and the distance l 12 between point 1 and point 2. The solution method is to solve a set of simultaneous equations:

[0053]

[0054] In the calculation, p d is taken as p x , p y , and p z in turn, and the corresponding l 12 is taken as l x / 2, W, H1, and H2 in turn, so that the coordinates of the eight limit boundary points can be solved.

[0055] S3: Laser radar online data acquisition. The train carrying a 3D multi-line laser radar scans and obtains real-time point cloud data in real time during driving on a designated line. At the same time, the real-time point cloud data and the positioning information corresponding to the real-time point cloud data are obtained. The positioning information is the position data corresponding to the aforementioned track line where the laser radar is currently located. The positioning method is not unique. In an embodiment, the online acquisition of the real-time point cloud data includes using a point cloud registration algorithm to compare the real-time point cloud data obtained online with the full-line point cloud data to determine the current position of the train. Other positioning methods such as GPS positioning, wheel speed positioning, etc. can also be used.

[0056] S4: Real-time point cloud data filtering. In one embodiment, the method further comprises filtering the real-time point cloud data based on a region of interest. Specifically, for each frame of real-time point cloud data obtained, in order to avoid unnecessary computational overhead, a region of interest (ROI) can be selected based on the train positioning information corresponding to the frame of real-time point cloud data.

[0057] The method of selecting the ROI is not unique. In one embodiment, the region of interest is selected by a pass-through filtering method. Specifically, after obtaining the positioning information of the train, the planar and longitudinal profile information of the track ahead of the train, such as the slope and the curvature, can be obtained based on the track line data. Thus, a section in the x (longitudinal), y (lateral), and z (vertical) directions can be drawn around the lidar, and points outside the drawn section can be discarded. If there is a slope in the track ahead, the section in the z (vertical) direction needs to be drawn longer to cover the track with the slope; if there is a curvature in the track ahead, the section in the y (lateral) direction needs to be drawn longer to cover the track with the curvature; and the section in the x (longitudinal) direction can be drawn according to the effective detection distance of the lidar used.

[0058] In addition, the ground point cloud data is more, its characteristics are obvious and there is no obstacle information, it is easy to remove and it is convenient to calculate after removal, so the ground point cloud data should also be removed. When the track is a certain distance from the ground, the aforementioned ROI selection method can already remove the ground point cloud. If the track is close to the ground, a plane segmentation algorithm is also needed to remove the ground plane. The plane segmentation algorithm is not unique, and as a preferred method, the RanSAC (Random Sample Consensus) algorithm can be used. The basic steps of this algorithm are to randomly select a certain number of seed points from the point cloud, fit a model equation using the seed points, calculate whether the model can adapt to more points under the error threshold, if yes, record this model as the result, if not, select new seed points for cyclic iteration. This method can quickly filter the main plane in the real-time point cloud data.

[0059] Only the obstacles within the ROI need to be concerned during calculation, since the lidar scans 360°, the data for positions behind the train or far above the train do not need to be concerned, for example, taking x (longitudinal) = [0:150], y (lateral) = [-50:50], z (vertical) = [-5:5] as the center of the lidar, which means only the data within the range of 150m in front of the lidar, 50m to the left, 50m to the right, 5m below, and 5m above are calculated.

[0060] S5: Point cloud segmentation and clustering. The segmentation and clustering method is not unique; a preferred method is a segmentation and clustering algorithm based on Euclidean distance. This algorithm can be combined with the characteristic of uneven distribution of lidar beams. For example... Figure 5 As shown, in one embodiment, obtaining the point cloud cluster data includes setting different clustering radii with the lidar as the origin, and using the same clustering threshold distance to divide the data within each clustering radius. Specifically, different clustering radii are set with the lidar as the origin, and the same clustering threshold distance is used within each clustering radius. The larger the clustering radius, the larger the clustering threshold distance. The clustering threshold distance indicates that two points are considered to belong to the same cluster when the Euclidean distance between them is less than or equal to this value.

[0061] S6: Track point cloud extraction. In one embodiment, track point cloud data is extracted based on the real-time point cloud data. Track point cloud extraction can refer to existing road train lane line point cloud extraction algorithms. The most direct method is to use the reflection intensity values ​​from the LiDAR returned data for pass-through filtering to select the track. In one embodiment, the track point cloud data is extracted using a principal component analysis algorithm.

[0062] Specifically, Principal Component Analysis (PCA) algorithm can accurately extract orbital point clouds. PCA can analyze k-dimensional orthogonal features (i.e., k principal components) from n-dimensional data. The calculation process is as follows:

[0063] ① Calculate the mean of each dimension of data, and then subtract the mean from each dimension of data.

[0064] ② Solve for the eigencovariance matrix:

[0065]

[0066] ③ Find the eigenvalues ​​and eigenvectors of the covariance matrix, arrange the eigenvalues ​​from largest to smallest, and the first k eigenvalues ​​are the k-dimensional principal features, and their corresponding eigenvectors are the principal directions.

[0067] For reference Figure 6 Let the main direction corresponding to the largest eigenvalue be the x-direction. Since the orbital structure is long and continuous, its longitudinal extension direction should be the x-direction. Therefore, according to the order of clustering radius from small to large as described in S5, we can determine whether the current point cloud cluster is an orbit by calculating whether the deviation value of the x-direction of the point cloud cluster and the x-direction of the orbital point cloud cluster in the previous radius is less than the threshold (when searching in the first radius, we change to calculating the deviation value of the x-direction with the radar coordinate system).

[0068] S7: Obstacle point cloud judgment. After obtaining the track point cloud in S6, all the track point clouds in the result of S5 are removed to obtain the remaining point cloud clusters. In an embodiment, the online monitoring of the obstacle includes judging whether the boundary points of the point cloud cluster data are in the rectangular surrounding box of the full-line limit boundary point. Specifically, the remaining point cloud clusters are projected onto the xOy plane, and then the boundaries of each point cloud cluster in the xOy plane are extracted and down-sampled. Then, the points in the down-sampled object boundary are traversed, and the corresponding previously calculated track body limit boundary point is queried according to the x coordinate of the point, and it is judged whether the point is in the rectangular surrounding box of the limit boundary point.

[0069] In an embodiment, when the proportion of the boundary points of the point cloud cluster data falling into the rectangular surrounding box of the full-line limit boundary point exceeds a threshold value, it is determined that the object represented by the point cloud cluster data is an obstacle. Specifically, if the proportion of the boundary points of the point cloud cluster data falling into the rectangular surrounding box of the full-line limit boundary point exceeds a certain threshold value, it is determined that the object represented by the point cloud cluster is an obstacle, which invades the normal driving area of the train and affects the safe driving.

[0070] The application also provides a non-transitory computer readable medium having computer program instructions stored thereon. When the computer program instructions are executed by one or more processors, the above method is implemented.

[0071] In summary, the application takes advantage of the fixed characteristics of the line structure of rail transit, and combines offline calculation of the safe driving area with online calculation of the object position, thereby reducing the difficulty of the obstacle judgment algorithm and improving the efficiency of online obstacle detection.

[0072] Although example embodiments have been described herein with reference to the accompanying drawings, it is to be understood that the example embodiments are only exemplary and are not intended to limit the scope of the application. Those of ordinary skill in the art can make various changes and modifications without departing from the scope and spirit of the application. All such changes and modifications are intended to be included within the scope of the application as claimed in the appended claims.

[0073] In the specification provided herein, a large number of specific details are described. However, it can be understood that the embodiments of the application can be practiced without these specific details. In some examples, well-known methods, structures and techniques are not shown in detail in order not to obscure the understanding of the present specification.

[0074] Similarly, it is to be understood that the embodiments of the present application can be used in any combination, whether such combinations are specifically noted herein or not. For example, a combination of any of the embodiments described herein can be used in any combination. Descriptions of examples of the present application in the detailed description and drawings are illustrative of specific embodiments of the present application and are not intended to be limiting. Other embodiments of the present application will be apparent to those skilled in the art in view of the following detailed description of the embodiments and the drawings.

[0075] Those skilled in the art will appreciate that all features described herein (including all accompanying claims, abstract and drawings) can be taken in combination with any other features described herein, and that the description of a particular feature can be taken as applying to its equivalents whether or not those equivalents are explicitly described elsewhere. Unless specifically stated otherwise, it is appreciated that, wherever used, the description herein of any feature of the application should be taken as a disclosure of a broad and an equivalent narrow feature.

[0076] Furthermore, to the extent that the terms "comprises", "comprising", "including", "includes", "containing", "contains" or variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to that of the term "comprising" as an open transition term without precluding any additional or other elements or steps.

[0077] The above description is only specific examples of the present application or the specific description of the present application, the protection scope of the present application is not limited to this, any skilled person in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. The protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of detecting a train obstacle, characterized by, The method comprises: acquiring, by a laser radar, full-line point cloud data along a full line, the full-line point cloud data including three-dimensional coordinates, reflection intensity, and a trajectory line of the train when the train is running; calculating, offline, full-line limit boundary points based on the full-line point cloud data, wherein the offline calculation of the full-line limit boundary points comprises: surrounding the train body limit with a rectangular surrounding frame based on the train body limit, leaving a margin between the rectangular surrounding frame and the train body limit, and the nodes of the rectangular surrounding frame being the full-line limit boundary points, and obtaining real-time poses of the train by an inertial measurement unit, thereby obtaining the full-line limit boundary points in a global coordinate system; acquiring, by the laser radar, real-time point cloud data online; and monitoring, online, the obstacles based on the full-line limit boundary points and point cloud cluster data obtained from the real-time point cloud data. The online monitoring of the obstacles comprises judging whether the boundary points of the point cloud cluster data are within the rectangular surrounding frame of the full-line limit boundary points.

2. The method of claim 1, wherein, When the proportion of the boundary points of the point cloud cluster data falling within the rectangular surrounding frame of the full-line limit boundary points exceeds a threshold value, the point cloud cluster data is determined to represent an obstacle.

3. The method of claim 2, wherein, The point cloud cluster data is obtained by setting different clustering radii with the laser radar as the origin and dividing within the clustering radii using the same clustering threshold distance.

4. The method of claim 1, wherein, The method further comprises extracting track point cloud data based on the real-time point cloud data.

5. The method of claim 1, wherein, The track point cloud data is extracted by a principal component analysis algorithm.

6. The method of claim 5, wherein, The method further comprises screening the real-time point cloud data based on a region of interest.

7. The method of claim 1, wherein, The region of interest is selected by a straight-through filtering method.

8. The method of claim 7, wherein, The acquisition of the full-line point cloud data comprises obtaining real-time poses of the train by an inertial measurement unit, thereby converting the trajectory line in the full-line point cloud data from a laser radar coordinate system to a global coordinate system.

9. The method of claim 1, wherein, The online acquisition of the real-time point cloud data comprises using a point cloud registration algorithm to compare the real-time point cloud data obtained online with the full-line point cloud data to determine the current position of the train.

10. The method of claim 1, wherein, The laser radar is a 3D multi-line laser radar.

11. The method of claim 1, wherein, The laser radar is installed at a central position of the front face of the head of the train.

12. The method of claim 1, wherein, When the computer program instructions are executed by one or more processors, the steps of the method of any one of claims 1 to 12 are implemented. 13.A non-transitory computer readable medium having stored thereon computer program instructions, wherein, When the computer program instructions are executed by one or more processors, the steps of the method of any one of claims 1 to 12 are implemented.

Images