Methods and devices for identifying road boundaries, storage media, and terminals

Abstract

This invention discloses a method, device, storage medium, and terminal for road boundary identification, relating to the field of autonomous driving technology. Its main objective is to address the problem of low accuracy in identifying backward road boundaries. The method primarily includes acquiring path planning data of the target vehicle and point cloud data of the target area, and performing rasterization processing on the point cloud data to obtain the extreme height values ​​of the point cloud within each grid cell. Based on the extreme height values ​​of the point cloud within each grid cell, candidate retaining wall grids are extracted sequentially, and retaining wall category detection is performed. Based on the candidate retaining wall grids and the retaining wall point extraction strategy corresponding to the retaining wall category, a set of retaining wall points is extracted. Based on the clustering of retaining wall points in the retaining wall point set, road boundary point data in at least one direction is determined, and the road boundary point data is sent to the control terminal of the target vehicle, enabling the control terminal to perform backward road boundary identification based on the road boundary point data. This method is mainly used for identifying road boundaries in mining areas.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to a method and apparatus for identifying road boundaries, a storage medium, and a terminal. Background Technology

[0002] During the operation of autonomous vehicles, it is necessary to identify obstacles around the vehicle to delineate the driving area. For passenger cars driving in cities, the characteristics of obstacles are relatively obvious. However, for wide-body vehicles in mining areas, the driving environment is more complex. When loading, unloading, and parking, it is necessary to accurately identify the retaining walls in a large area behind the vehicle to avoid the dumping of excavated soil into the inside of the retaining wall or the collision between the vehicle and the retaining wall. Therefore, wide-body vehicles in mining areas have higher requirements for the accuracy of road boundary recognition.

[0003] Existing road boundary recognition methods use grids as the smallest unit for extracting road boundary point clouds. This limits the extraction of road boundaries to the size of the grid and the distribution of point clouds within the grid, resulting in low recognition accuracy. Furthermore, when performing gradient calculations, voxel filtering is used for downsampling, which means that the final calculation result only reflects the average value. When the gradient of the retaining wall is not obvious or there are few point cloud lines in the grid, the retaining wall is easily missed, further reducing the recognition accuracy of road boundaries. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for identifying road boundaries, a storage medium, and a terminal, the main purpose of which is to solve the problem of low accuracy in identifying backward road boundaries.

[0005] According to one aspect of the present invention, a method for identifying road boundaries is provided, comprising:

[0006] The path planning data of the target vehicle and the point cloud data of the target area collected by the rearward monitoring radar of the target vehicle are obtained, and the point cloud data is rasterized to obtain the extreme height value of the point cloud in each grid in the planar point cloud grid.

[0007] Based on the path planning data and the extreme height values ​​of the point cloud within each grid, retaining wall candidate grids are extracted and retaining wall categories are detected sequentially to determine the retaining wall candidate grids and retaining wall categories within the target area. Based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories, a set of retaining wall points is extracted.

[0008] Based on the clustering of retaining wall points in the set of retaining wall points, road boundary point data in at least one direction is determined, and the road boundary point data is sent to the control terminal of the target vehicle so that the control terminal can perform backward road boundary identification based on the road boundary point data.

[0009] Furthermore, the extraction of the retaining wall point set based on the candidate grid and the retaining wall point extraction strategy corresponding to the retaining wall category includes:

[0010] The first reference point cloud plane is obtained by performing planar fitting on the first reference point cloud of the candidate grid, and the first angle between the normal vector of the first reference point cloud plane and the bottom plane of the target vehicle is calculated.

[0011] Based on the comparison result between the first included angle and the preset included angle threshold, the retaining wall grid is determined from the candidate grid, and the preset included angle threshold is determined based on the retaining wall category;

[0012] The second reference point cloud plane is obtained by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, and the second angle between the normal vector of the second reference point cloud plane and the bottom plane of the target vehicle is calculated.

[0013] Based on the comparison result between the second included angle and the preset included angle threshold, the retaining wall points in the retaining wall grid are determined, and a retaining wall point set is constructed based on all the retaining wall points in the retaining wall grid.

[0014] Furthermore, before obtaining the first reference point cloud plane by performing planar fitting on the first reference point cloud of the candidate raster, the method further includes:

[0015] The centroid of the point cloud in the candidate grid is calculated, and the associated point cloud is divided with the centroid as the center and the first associated point cloud division parameter as the radius to obtain the first associated point cloud;

[0016] Calculate the first height difference between each scan point in the first associated point cloud and the centroid point, and extract the first reference point cloud from the first associated point cloud based on the comparison result of the first height difference and the first preset height difference threshold.

[0017] Before obtaining the second reference point cloud plane by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, the method further includes:

[0018] Using each target scanning point in the retaining wall grid as the center and the second associated point cloud division parameter as the radius, the associated point cloud is divided to obtain the second associated point cloud;

[0019] Calculate the second height difference between each scan point in the second associated point cloud and the target scan point, and extract the second reference point cloud from the second associated point cloud based on the comparison result of the second height difference and the second preset height difference threshold.

[0020] Furthermore, the retaining wall category includes mountain retaining walls and non-mountain retaining walls, and the process of determining the retaining wall grid from the candidate grid based on the comparison result of the first included angle and the preset included angle threshold includes:

[0021] If the retaining wall type is a mountain retaining wall, the retaining wall grid is determined based on the comparison result between the first included angle and the first preset included angle threshold.

[0022] If the retaining wall category is a non-mountain retaining wall, the retaining wall grid is determined based on the comparison results of the first included angle with the second preset included angle threshold and the third preset included angle threshold.

[0023] Furthermore, the height extrema include a maximum height and a minimum height, and the process of rasterizing the point cloud data to obtain the height extrema of the point cloud within each grid cell includes:

[0024] Based on the grid size parameters of a preset planar grid, the grid coordinates of each scan point in the point cloud data are calculated in the preset planar grid, and each grid in the preset planar grid is configured with an atomic lock;

[0025] Assign corresponding threads to different point clouds in the point cloud data, and write the scan points in the different point clouds in parallel into a grid that matches the grid coordinates of the scan points based on the threads, to obtain the point cloud grid of the point cloud data.

[0026] Based on the height values ​​of all scanned points in each grid of the point cloud grid, the maximum height and minimum height value in each grid are determined, where the height value is the height of the scanned point in the normal direction of the preset plane grid.

[0027] Further, the step of extracting candidate retaining wall grids and detecting retaining wall categories based on the path planning data and the extreme height values ​​of the point cloud within each grid, to determine the candidate grids and retaining wall categories for retaining walls within the target area, includes:

[0028] The first height difference of each grid is calculated based on the extreme height of the point cloud within each grid. If the first height difference is less than the third preset height threshold, the grid is determined as a candidate grid for the retaining wall in the target area. The third preset height threshold is configured based on the height of the retaining wall in the target mining area.

[0029] The candidate raster is used as the head raster, and the tail raster is determined according to the preset traversal parameters, which include the distance parameter and the traversal direction parameter in the point cloud raster.

[0030] Based on the height extreme values ​​of the head grid and the tail grid, a second height difference is calculated. If the second height difference is greater than the fourth preset height threshold, the retaining wall category of the retaining wall in the target area is determined to be a mountain retaining wall. The fourth preset height threshold is configured based on the height of the mountain in the target mining area.

[0031] Furthermore, determining road boundary point data in at least one direction based on the clustering of retaining wall points in the retaining wall point set includes:

[0032] Based on the projection values ​​of the clusters on the horizontal and vertical coordinate axes in the target vehicle coordinate system, and the horizontal coordinate values ​​of the scan points in the clusters, the retaining wall points in the retaining wall point set are divided into retaining wall points in at least one direction. The target vehicle coordinate system is a planar coordinate system constructed with the rear axis center point of the target vehicle as the center, the right side of the vehicle as the positive direction of the horizontal coordinate axis, and the front direction of the vehicle as the positive direction of the vertical coordinate axis.

[0033] For each retaining wall point in each direction, calculate the concave hull data of the retaining wall point, and configure the concave hull data as the road boundary point data of the corresponding direction.

[0034] According to another aspect of the present invention, a road boundary identification device is provided, comprising:

[0035] The acquisition module is used to acquire the path planning data of the target vehicle and the point cloud data of the target area collected by the rear monitoring radar of the target vehicle, and to perform rasterization processing on the point cloud data to obtain the extreme height value of the point cloud in each grid in the planar point cloud grid.

[0036] The extraction module is used to extract retaining wall candidate grids and detect retaining wall categories in sequence based on the path planning data and the height extreme values ​​of the point cloud in each grid, to determine the retaining wall candidate grids and retaining wall categories in the target area, and to extract a set of retaining wall points based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories.

[0037] The determination module is used to determine road boundary point data in at least one direction based on the cluster of retaining wall points in the set of retaining wall points, and send the road boundary point data to the control terminal of the target vehicle so that the control terminal can perform backward road boundary identification based on the road boundary point data.

[0038] Furthermore, the extraction module includes:

[0039] The first fitting unit is used to obtain a first reference point cloud plane by performing planar fitting on the first reference point cloud of the candidate grid, and to calculate the first angle between the normal vector of the first reference point cloud plane and the bottom plane of the target vehicle.

[0040] The first determining unit is used to determine the retaining wall grid from the candidate grid based on the comparison result between the first included angle and the preset included angle threshold, wherein the preset included angle threshold is determined based on the retaining wall category;

[0041] The second fitting unit is used to obtain a second reference point cloud plane by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, and to calculate the second angle between the normal vector of the second reference point cloud plane and the bottom plane of the target vehicle.

[0042] The second determining unit is used to determine the retaining wall points in the retaining wall grid based on the comparison result between the second included angle and the preset included angle threshold, and to construct a retaining wall point set based on all the retaining wall points in the retaining wall grid.

[0043] Furthermore, the device also includes:

[0044] The first partitioning module is used to calculate the centroid of the point cloud in the candidate grid, and to partition the associated point cloud with the centroid as the center and the first associated point cloud partitioning parameter as the radius to obtain the first associated point cloud;

[0045] The first calculation module is used to calculate the first height difference between each scan point in the first associated point cloud and the centroid point, and to extract the first reference point cloud from the first associated point cloud based on the comparison result of the first height difference and the first preset height difference threshold.

[0046] The second partitioning module is used to partition the associated point cloud with each target scanning point in the retaining wall grid as the center and the second associated point cloud partitioning parameter as the radius, so as to obtain the second associated point cloud.

[0047] The second calculation module is used to calculate the second height difference between each scan point in the second associated point cloud and the target scan point, and to extract the second reference point cloud from the second associated point cloud based on the comparison result of the second height difference and the second preset height difference threshold.

[0048] Furthermore, the determining module includes:

[0049] The third determining unit is used to determine the retaining wall grid based on the comparison result between the first included angle and the first preset included angle threshold if the retaining wall type is a mountain retaining wall.

[0050] The fourth determining unit is used to determine the retaining wall grid based on the comparison results of the first included angle with the second preset included angle threshold and the third preset included angle threshold if the retaining wall type is a non-mountain retaining wall.

[0051] Furthermore, the acquisition module includes:

[0052] The first calculation unit is used to calculate the grid coordinates of each scan point in the point cloud data in the preset planar grid based on the grid size parameters of the preset planar grid, wherein each grid in the preset planar grid is configured with an atomic lock;

[0053] The allocation unit is used to allocate corresponding threads to different point clouds in the point cloud data, and write the scan points in the different point clouds in parallel into a grid that matches the grid coordinates of the scan points based on the threads, so as to obtain the point cloud grid of the point cloud data.

[0054] The fifth determining unit is used to determine the maximum height and minimum height of each grid cell based on the height values ​​of all scanned points in each grid cell of the point cloud grid, wherein the height value is the height of the scanned point in the normal direction of the preset plane grid.

[0055] Furthermore, the determining module includes:

[0056] The sixth determining unit is used to calculate the first height difference of each grid cell based on the extreme height of the point cloud in each grid cell. If the first height difference is less than the third preset height threshold, the grid cell is determined as a candidate grid cell for the retaining wall in the target area. The third preset height threshold is configured based on the height of the retaining wall in the target mining area.

[0057] The traversal unit is used to take the candidate grid as the head grid for traversal and determine the tail grid for traversal according to preset traversal parameters, including the distance parameter and the direction parameter for traversal in the point cloud grid.

[0058] The second calculation unit is used to calculate a second height difference based on the height extreme value of the head grid and the height extreme value of the tail grid. If the second height difference is greater than a fourth preset height threshold, the retaining wall category of the retaining wall in the target area is determined to be a mountain retaining wall. The fourth preset height threshold is configured based on the height of the mountain in the target mining area.

[0059] Furthermore, the determining module includes:

[0060] The partitioning unit is used to divide the retaining wall points in the retaining wall point set into retaining wall points in at least one orientation based on the projection values ​​of the clusters on the horizontal and vertical coordinate axes in the target vehicle coordinate system and the horizontal coordinate values ​​of the scan points in the clusters. The target vehicle coordinate system is a planar coordinate system constructed with the center point of the rear axis of the target vehicle as the center, the right side of the vehicle as the positive direction of the horizontal coordinate axis, and the front direction of the vehicle as the positive direction of the vertical coordinate axis.

[0061] The third calculation unit is used to calculate the concave data of the retaining wall point for each direction, and configure the concave data as the road boundary point data of the corresponding direction.

[0062] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform an operation corresponding to the road boundary recognition method described above.

[0063] According to another aspect of the present invention, a terminal is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0064] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the road boundary recognition method described above.

[0065] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages:

[0066] This invention provides a method, apparatus, storage medium, and terminal for identifying road boundaries. In this embodiment, the method acquires path planning data of a target vehicle and point cloud data of the target area collected by the vehicle's rear-view radar. The point cloud data is then rasterized to obtain the extreme height values ​​of the point cloud within each grid cell. Based on the path planning data and the extreme height values ​​of the point cloud within each grid cell, retaining wall candidate grids are extracted and retaining wall category detection is performed sequentially to determine the retaining wall candidate grids and retaining wall categories within the target area. Finally, based on the retaining wall candidate grids and their relationship with the retaining wall category, the method further refines the method. A retaining wall point extraction strategy corresponding to the wall type is used to extract a set of retaining wall points. Based on the clusters of retaining wall points in the set, road boundary point data in at least one direction is determined, and the road boundary point data is sent to the control terminal of the target vehicle. This enables the control terminal to perform backward road boundary recognition based on the road boundary point data, greatly reducing the limitations of the grid on retaining wall point extraction and significantly improving the accuracy of retaining wall point extraction. At the same time, different retaining wall point extraction strategies are used for different retaining wall types, thereby effectively improving the accuracy of retaining wall point recognition and thus improving the accuracy of road boundary recognition.

[0067] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0068] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0069] Figure 1 A flowchart of a road boundary identification method provided by an embodiment of the present invention is shown;

[0070] Figure 2 A flowchart of another road boundary identification method provided by an embodiment of the present invention is shown;

[0071] Figure 3 This invention provides a flowchart for extracting retaining wall points according to an embodiment of the invention.

[0072] Figure 4 This diagram illustrates a block diagram of a road boundary identification device provided in an embodiment of the present invention.

[0073] Figure 5 A schematic diagram of the structure of a terminal provided in an embodiment of the present invention is shown. Detailed Implementation

[0074] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0075] Existing road boundary recognition methods use grids as the smallest unit for extracting road boundary point clouds. This limits the extraction accuracy to the grid size and the distribution of point clouds within the grid, resulting in low recognition accuracy. Furthermore, voxel filtering for downsampling during gradient calculation means the final result only reflects the average value. When the gradient of a retaining wall is not significant or the point cloud line bundles within the grid are few, retaining walls are easily missed, further reducing the accuracy of road boundary recognition. This invention provides a road boundary recognition method, such as... Figure 1 As shown, the method includes:

[0076] 101. Obtain the path planning data of the target vehicle and the point cloud data of the target area collected by the rearward monitoring radar of the target vehicle, and perform rasterization processing on the effective point cloud data to obtain the extreme value of the height of the point cloud in each grid in the planar point cloud grid.

[0077] In this embodiment of the invention, the target vehicle is a wide-body mining vehicle whose rear road boundary needs to be identified. The path planning data is the data of the target vehicle's upcoming rearward driving path, i.e., the reversing driving path data, which can be obtained from the target vehicle's vehicle control system. The rearward monitoring radar is a blind spot radar installed at the rear of the target vehicle; the installation location and specific type of radar are not specifically limited in this embodiment. The point cloud consists of a large number of scan points obtained by the radar equipment scanning the surrounding environment; these scan points are distributed with varying degrees of sparsity depending on the volume of the scanned object. The reference coordinate system for the point cloud data and path planning data is a target vehicle coordinate system constructed with the rear axle center point of the target vehicle as the origin, the direction parallel to the vehicle body and pointing towards the front of the vehicle as the positive Y-axis, and the direction perpendicular to the Y-axis and pointing towards the right side of the target vehicle as the positive X-axis. After acquiring the current frame point cloud data collected by the radar, the point cloud data is rasterized based on its coordinates in the target vehicle coordinate system and the size of the grid in the XY plane. This involves calculating the grid coordinates of the scanned points in the point cloud, and then calculating the maximum and minimum Z-axis coordinates (i.e., height extrema) of each grid cell based on the Z-axis coordinates of the point cloud in each cell. The Z-axis is a coordinate axis perpendicular to the XY plane and pointing upwards.

[0078] It should be noted that, due to the wide scanning range of radar, it can identify obstacles more than 20 meters away from the target vehicle. However, for the target vehicle to travel, only point cloud data within a few meters is needed to identify the retaining wall, while point cloud data at greater distances is invalid. Therefore, before rasterizing the point cloud data, it can be filtered based on a preset coordinate threshold to obtain the point cloud data within the area to be identified (the target area), and then the raster coordinates corresponding to the point cloud data within the target area can be calculated. The preset coordinate threshold can include an X-axis threshold and / or a Y-axis threshold, and the threshold value can be 2m, 3m, or customized according to the actual scenario requirements; this embodiment of the invention does not impose specific limitations. By filtering the point cloud data currently acquired by the radar, distant and invalid point clouds can be removed, thereby effectively reducing the amount of data computation in subsequent data processing.

[0079] 102. Based on the path planning data and the extreme height values ​​of the point cloud in each grid, retaining wall candidate grids are extracted and retaining wall categories are detected in sequence to determine the retaining wall candidate grids and retaining wall categories in the target area. Based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories, a set of retaining wall points is extracted.

[0080] In this embodiment of the invention, the traversal starting point is determined based on path planning data. Specifically, the coordinates of the planned path point r_i are set as (road_x_i, road_y_i, road_z_i), then the corresponding starting coordinates are (x_start, y_start) = (((road_x_i - point_x_min) / mesh_x_size, road_y_i / mesh_y_size). Based on the starting grid (x_start, y_start), traversal proceeds left and right... The last three directions are traversed. The point cloud within the target area includes ground point clouds, retaining wall point clouds, and other obstacle point clouds, such as vehicle point clouds and pedestrian point clouds. The point cloud to be identified is the retaining wall point cloud, which serves as the road boundary. In mining environments, retaining walls are generally low walls close to the height of vehicles, typically around 1 meter high, while wide-body vehicles are typically around 3 meters high. Therefore, the height of the corresponding object in the point cloud can be determined based on the maximum and minimum height values ​​of the point cloud in the grid, and then the grid can be determined as either a retaining wall point cloud grid or a vehicle point cloud grid based on the height. For example, if the difference between the maximum and minimum height values ​​in a grid cell is greater than 1.5 meters, it indicates that the object corresponding to the point cloud in that grid cell may be a vehicle. Conversely, if the difference between the maximum and minimum height values ​​in a grid cell is less than 1.5 meters, it indicates that the object corresponding to the point cloud in that grid cell may be a retaining wall, and that grid cell can be considered a candidate retaining wall grid. Furthermore, in some environments, mountains and man-made low walls may both serve as retaining walls. To more accurately identify retaining walls, after identifying each grid cell as a candidate retaining wall grid, further refinement of the candidate retaining wall grid is performed. To determine if a mountain exists next to the current grid, since mountains are much taller than retaining walls for people and wide vehicles, and mountains have a certain width, the height difference between the current grid and grids within a certain range can be compared to determine whether the current retaining wall includes a mountain, classifying it as a mountain-type retaining wall. After determining the candidate grids and retaining wall categories, to accurately identify retaining wall points from these candidate grids, retaining wall points need to be extracted from the candidate grids based on retaining wall point extraction strategies corresponding to different retaining wall categories, thus obtaining the points that completely correspond to the retaining wall, i.e., the current point set.

[0081] It should be noted that the retaining wall point extraction strategy can construct a planar normal vector based on the point cloud of each candidate retaining wall grid and the surrounding point cloud. The angle between this planar normal vector and the XY plane is used to determine whether a candidate retaining wall grid is indeed a retaining wall grid. Further, planar normal vectors are constructed point-by-point for the point cloud within the retaining wall grid. Based on the angle between this point-by-point planar normal vector and the XY plane, each scan point within the retaining wall grid is further refined. However, since the criteria for determining the angle differ for retaining walls that are either mountain retaining walls or non-mountain retaining walls (i.e., the corresponding angle thresholds differ), it is necessary to extract candidate retaining wall grids and their scan points based on the angle threshold corresponding to the retaining wall category. By using a multi-step extraction method based on planar normal vectors, the limitation of grid size on retaining wall point extraction can be overcome, allowing for point-by-point identification of the point cloud within the grid, thereby effectively improving the extraction accuracy of retaining wall points. Furthermore, by identifying the type of retaining wall and using the retaining wall point extraction strategy corresponding to that type, the retaining wall identification can be made to include retaining walls that contain mountains, thus making the identification of retaining walls more comprehensive.

[0082] 103. Based on the clustering of retaining wall points in the set of retaining wall points, determine road boundary point data in at least one direction, and send the road boundary point data to the control terminal of the target vehicle so that the control terminal can perform backward road boundary identification based on the road boundary point data.

[0083] In this embodiment of the invention, since the point cloud collected by the rearward radar of the target vehicle includes the rearward direction, the left direction, and the right direction, and the road boundary that the target vehicle needs to identify is also a road boundary that surrounds the sides and rear of the vehicle, the location of the retaining wall point is determined, and then...

[0084] Road boundary data is determined by identifying retaining wall points at different locations. Specifically, retaining wall points in the direct rear direction can be determined based on the 5-degree projection length of the cluster, and left and right retaining wall points can be determined based on the positive or negative coordinates of the retaining wall points. Specifically, Euclidean clustering can be used to cluster the retaining wall points; however, this embodiment of the invention does not impose specific limitations. The road boundary point data can specifically be concave hull point data; this embodiment of the invention does not impose specific limitations. After obtaining the road boundary point data, in order to enable target vehicles to adjust...

[0085] The entire vehicle's driving trajectory is used to avoid road boundaries, and the road boundary point data is sent to the control terminal of the target vehicle. This control terminal can be a control system or control module in the target vehicle used to control autonomous driving. This embodiment of the invention does not make specific limitations.

[0086] In one embodiment of the present invention, for further illustration and limitation, such as Figure 2As shown, the step of extracting the retaining wall point set based on the candidate grid and the retaining wall point extraction strategy corresponding to the retaining wall category includes: 5 201, obtaining the first reference point cloud plane by performing plane fitting on the first reference point cloud of the candidate grid, and calculating the first angle between the normal vector of the first reference point cloud plane and the bottom plane of the target vehicle.

[0087] 202. Based on the comparison result between the first included angle and the preset included angle threshold, the retaining wall grid is determined from the candidate grid.

[0088] 0 203. A second reference point cloud plane is obtained by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, and the second angle between the normal vector of the second reference point cloud plane and the bottom plane of the target vehicle is calculated.

[0089] 204. Based on the comparison result between the second included angle and the preset included angle threshold, the retaining wall points in the retaining wall grid are determined, and a set of 5 retaining wall points is constructed based on all the retaining wall points in the retaining wall grid.

[0090] In this embodiment of the invention, in addition to retaining wall grids, there may be ground grids and / or vehicle grids that need to be filtered out in the candidate grids. In order to filter out retaining wall grids from the candidate grids, the filtering of retaining wall grids is achieved based on the difference in the directional characteristics of the normal vectors of the fitting planes corresponding to the ground grid reference point cloud, vehicle grid reference point cloud, and retaining wall grid reference point cloud. Since the normal vector of the fitting plane corresponding to the ground grid reference point cloud is perpendicular or approximately perpendicular to the vehicle's bottom surface, and the angle between the normal vector of the fitting plane corresponding to the vehicle grid reference point cloud and the vehicle's bottom surface is approximately zero, while the normal vector of the fitting plane corresponding to the retaining wall grid reference point cloud forms a certain angle with the vehicle's bottom surface, the retaining wall grid can be extracted using the angle between the normal vector of the first reference point cloud plane and the target vehicle's bottom plane, and a preset angle threshold. Similarly, retaining wall points in the target scan points can be extracted based on the angle between the normal vector of the fitting plane (second reference point cloud plane) corresponding to the second reference point cloud (the target scan point in the retaining wall grid) and the vehicle's bottom surface, thus obtaining a set of retaining wall points. The preset angle threshold is determined based on the retaining wall category; that is, the same preset angle threshold applies to both the retaining wall grid extraction and the current point grid extraction process, while different retaining wall categories correspond to different preset angle thresholds.

[0091] It should be noted that the first reference point cloud plane is obtained by fitting the first reference point cloud using Principal Component Analysis (PCA). Similarly, the second reference point cloud plane is obtained by fitting the second reference point cloud using PCA. Since the PCA algorithm only considers the positions of most points in the reference point cloud for plane fitting, it can effectively filter sparse scan points in the reference point cloud, greatly improving the accuracy of subsequent normal vector angle calculation, thereby effectively improving the accuracy of retaining wall point extraction.

[0092] Furthermore, this embodiment of the invention extracts the retaining wall grid by referring to the angle between the point cloud plane normal vector and the vehicle bottom plane, and performs point-by-point calculation on each scan point in the retaining wall grid based on the same method to complete the extraction of retaining wall points, achieving precise extraction of retaining wall points. In this process, the extraction of retaining wall points is obtained by point-by-point calculation on each scan point in the retaining wall grid, breaking the limitation of grid size on the accuracy of retaining wall point extraction, thereby effectively improving the accuracy and precision of retaining wall point extraction. Moreover, since point-by-point calculation involves a large amount of computation, CPU-based processing cannot meet the real-time requirements of boundary recognition. Therefore, this embodiment of the invention utilizes the already configured Compute Unified Device Architecture (CUDA) in the Graphics Processing Unit (GPU) to execute the calculation process for retaining wall grid and retaining wall point extraction, greatly improving the computational speed and reducing the computational time, thereby improving the computational efficiency of retaining wall point extraction and meeting the real-time requirements of boundary recognition.

[0093] In one embodiment of the present invention, for further explanation and limitation, before the step of obtaining the first reference point cloud plane by performing planar fitting on the first reference point cloud of the candidate raster, the method further includes:

[0094] The centroid of the point cloud in the candidate grid is calculated, and the associated point cloud is divided with the centroid as the center and the first associated point cloud division parameter as the radius to obtain the first associated point cloud;

[0095] Calculate the first height difference between each scan point in the first associated point cloud and the centroid point, and extract the first reference point cloud from the first associated point cloud based on the comparison result of the first height difference and the first preset height difference threshold.

[0096] Before obtaining the second reference point cloud plane by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, the method further includes:

[0097] Using each target scanning point in the retaining wall grid as the center and the second associated point cloud division parameter as the radius, the associated point cloud is divided to obtain the second associated point cloud;

[0098] Calculate the second height difference between each scan point in the second associated point cloud and the target scan point, and extract the second reference point cloud from the second associated point cloud based on the comparison result of the second height difference and the second preset height difference threshold.

[0099] In this embodiment of the invention, to determine the spatial plane corresponding to the current candidate grid, a first circular region is defined with the centroid of the point cloud in the current candidate grid as the center and the first associated point cloud partitioning parameter as the radius. The point clouds of all grids within this first circular region are considered as the first associated point cloud of the current candidate grid. The first height difference between each scan point and the centroid in this first associated point cloud is calculated. If the first height difference is greater than a first preset height difference threshold, the scan point is added to the set of the first reference point cloud. If the first height difference is less than the first preset height difference threshold, the scan point is deleted. After traversing and filtering each scan point in the surrounding point cloud based on the first preset height difference threshold, the first reference point cloud is obtained. The centroid of the point cloud of the candidate grid is the geometric center of all scan points in the grid. The purpose of filtering each scan point in the first associated point cloud based on the first preset height difference threshold is to extract the scan points that have a more significant height change relative to the centroid and are more representative as reference points for subsequent normal vector calculation. The first preset height difference threshold can be 0.5 or can be customized according to the actual application scenario. This embodiment of the invention does not impose specific limitations.

[0100] In this embodiment of the invention, the retaining wall grid includes multiple target scanning points, and these target scanning points may not all be retaining wall points. For example, there are 10 scanning points in the grid, of which 7 are retaining wall scanning points, and the remaining 3 are scanning points outside the retaining wall. In order to accurately extract the retaining wall points and filter out the target scanning points in the retaining wall grid that do not belong to the retaining wall points, after determining the retaining wall grid, each target scanning point in the retaining wall grid is used as the center, and a second circular area is delineated with the second reference point division parameter as the radius. The point cloud of all grids within the second circular area is used as the second associated point cloud of the current target scanning point. Based on the second height difference between each scanning point in this second associated point cloud and the current target scanning point and the second preset height difference threshold, a second reference point cloud is determined. Then, based on the second angle formed by the normal vector of the plane where the second reference point cloud is located and the XY plane, and the preset angle threshold, it is determined whether the current scanning point in the retaining wall grid is a retaining wall point. The processes for filtering the second reference point cloud, determining the second included angle, identifying the retaining wall points, and determining the retaining wall grid are the same, and will not be repeated here. The first preset height difference threshold and the second preset height difference threshold can be configured with the same value, or different values ​​can be configured according to the actual application scenario; the first associated point cloud partitioning parameter and the second reference point partitioning parameter can be configured with the same value, or different values ​​can be configured according to the actual application scenario, and are not specifically limited in this embodiment.

[0101] In one embodiment of the present invention, for further explanation and limitation, the step of determining the retaining wall grid from the candidate grid based on the comparison result of the first included angle and the preset included angle threshold includes:

[0102] If the retaining wall type is a mountain retaining wall, the retaining wall grid is determined based on the comparison result between the first included angle and the first preset included angle threshold.

[0103] If the retaining wall category is a non-mountain retaining wall, the retaining wall grid is determined based on the comparison results of the first included angle with the second preset included angle threshold and the third preset included angle threshold.

[0104] In this embodiment of the invention, the retaining wall is a slope with a certain inclination. The angle between the normal vector of the first reference point cloud plane corresponding to the retaining wall grid and the bottom surface of the target vehicle is significantly less than 90°, while the angle between the normal vector of the first reference point cloud plane corresponding to the ground grid and the bottom surface of the target vehicle is close to 90°. Simultaneously, the side of the vehicle is nearly perpendicular to the bottom surface of the target vehicle, and the angle between the normal vector of the first reference point cloud plane corresponding to the vehicle grid and the bottom surface of the target vehicle is close to 0°. The scanning surface of the mountain is similar to the side of the vehicle; therefore, the angle between the normal vector of the first reference point cloud plane corresponding to the mountain grid and the bottom surface of the target vehicle is close to 0°. To ensure the accuracy of retaining wall grid and point extraction in the presence of a retaining wall, different preset angle thresholds are determined based on the current category. The retaining wall category includes mountain retaining walls and non-mountain retaining walls. For non-mountain retaining walls, i.e., retaining walls that do not contain mountains, when the first included angle approaches 0°, the current retaining wall grid can be determined to be a vehicle grid. Therefore, if the first included angle is between the second preset included angle threshold and the third preset included angle threshold, the current candidate grid is determined to be a retaining wall grid. If the first included angle is equal to the second preset included angle threshold or the third preset included angle threshold, the current candidate grid is determined to be a non-retaining wall grid, thereby realizing the extraction of retaining wall grids. The second and third preset included angle thresholds can be configured as 0° and 90°, respectively, or can be configured as angle values ​​slightly greater than 0° and less than 90° according to the actual application scenario. For example, the second preset included angle threshold is 5° and the third preset included angle threshold is 80°. This embodiment of the invention does not impose specific limitations. For retaining walls that incorporate a portion of a mountain as a retaining wall, the likelihood of vehicles appearing near the mountain is relatively low. When the first included angle approaches 0°, the current retaining wall grid can be identified as a mountain grid. Therefore, filtering of the ground retaining wall grid is only required based on the first preset included angle threshold. The first preset included angle threshold can be equal to the larger of the second preset included angle threshold and the third preset included angle threshold, or it can be a value that is not equal to the second preset included angle threshold and the third preset included angle threshold, but is less than or equal to 90°. This embodiment of the invention does not impose specific limitations.

[0105] In specific application scenarios, the flowchart for wall point extraction is as follows: Figure 3As shown, the grid is traversed from the starting point in three directions: left, right, and back. If the traversal is not completed, the centroid of the current grid and the point cloud within a certain range (determined by the associated point cloud partitioning parameters) are obtained. Then, it is determined whether there is a mountain within the distance range (determined by the preset traversal parameters), and then the preset included angle threshold for judgment is determined. Based on the normal vector algorithm, it is determined whether the current grid belongs to the retaining wall. If it does not belong, the next grid is traversed. If it belongs, all points in the grid are further extracted and determined based on the normal vector algorithm whether they belong to the retaining wall. If they belong, the point is added to the retaining wall point set and the process jumps to the next row. If it does not belong, the next grid is traversed until the traversal ends, and the retaining wall point set is obtained.

[0106] It should be noted that the extraction process for retaining wall points is the same as the extraction process for retaining wall grids. It also requires identifying retaining wall points based on different retaining wall categories and corresponding preset angle thresholds. This will not be elaborated further in this embodiment. Since the characteristics of mountains are quite similar to those of vehicles, if retaining walls containing mountains and those not containing mountains are judged based on the same preset angle threshold, retaining walls in the mountain area will be identified as interfering vehicles, resulting in inaccurate identification. However, by determining the corresponding preset angle threshold based on the retaining wall category, and filtering for cases where mountains are used as retaining walls using the corresponding preset angle threshold, the accuracy of retaining wall identification can still be ensured even when mountains are used as retaining walls, thereby greatly improving the applicability of road boundary recognition.

[0107] In one embodiment of the present invention, for further explanation and limitation, the step of performing rasterization processing on the point cloud data to obtain the extreme height values ​​of the point cloud within each raster includes:

[0108] Based on the grid size parameters of the preset planar grid, calculate the grid coordinates of each scan point in the point cloud data within the preset planar grid;

[0109] Assign corresponding threads to different point clouds in the point cloud data, and write the scan points in the different point clouds in parallel into a grid that matches the grid coordinates of the scan points based on the threads, to obtain the point cloud grid of the point cloud data.

[0110] Based on the height values ​​of all scanned points in each grid of the point cloud grid, the maximum and minimum height values ​​in each grid are determined.

[0111] In this embodiment of the invention, the preset planar grid is a two-dimensional planar grid constructed based on the target vehicle coordinate system, that is, the result of dividing the plane containing the target vehicle coordinate system into grids according to the grid size parameter. The grid size parameter is the size of a unit grid, which can be customized according to actual scenario requirements; this embodiment of the invention does not impose specific limitations. The calculation method for grid coordinates has been explained in step 101 and will not be repeated here. To rasterize point cloud data based on the GPU, a CUDA core is allocated as a thread for each point cloud, and an atomic lock is configured for each grid. The CUDA core corresponding to each point cloud allocates the scan points in the point cloud to the corresponding grid in parallel. For the CUDA core, after determining the grid coordinates of each scan point in the point cloud data, the atomic lock state of the grid corresponding to the current scan point is obtained iteratively until the atomic lock of the corresponding grid is obtained, and the current scan point is stored in the grid. For each scan point in the grid, there are height extreme values, including the maximum height and the minimum height. The height value is the height of the scan point in the normal direction of the preset planar grid, and is also the Z-axis coordinate value of the scan point. To determine the extreme height values, while storing the current scan point in the grid, the Z-axis coordinate of the current scan point is compared with the maximum and minimum Z-axis coordinates of other scan points in the grid. If the Z-axis coordinate of the current scan point is greater than the maximum Z-axis coordinate, the maximum Z-axis coordinate is updated; if the Z-axis coordinate of the current scan point is less than the minimum Z-axis coordinate, the minimum Z-axis coordinate is updated. After the operation is completed, the atomic lock is released, thus obtaining the extreme height values ​​of each grid after all CUDA cores have completed the rasterization processing of the corresponding point cloud.

[0112] It should be noted that by using GPUs to rasterize point cloud data, the processing speed of point cloud data rasterization can be effectively improved. At the same time, configuring atomic locks for each grid can prevent other CUDA cores from operating on the current grid and causing errors when one CUDA core is writing to the current grid. This greatly improves the efficiency of rasterization processing while ensuring the accuracy and orderliness of parallel processing.

[0113] In one embodiment of the present invention, for further explanation and limitation, the step of extracting candidate retaining wall grids and detecting retaining wall categories based on the 5 path planning data and the extreme height values ​​of the point cloud within each grid, and determining the candidate grids and retaining wall categories of the retaining wall within the target area includes:

[0114] The first height difference of each grid is calculated based on the extreme height of the point cloud within each grid. If the first height difference is less than a third preset height difference threshold, the grid is determined as a candidate grid for the retaining wall within the target area.

[0115] 0. The candidate grid is used as the head grid for traversal, and the tail grid for traversal is determined according to the preset traversal parameters.

[0116] Based on the height extreme values ​​of the head grid and the height extreme values ​​of the tail grid, a second height difference is calculated. If the second height difference is greater than a fourth preset height difference threshold, the retaining wall category of the retaining wall in the target area is determined to be a mountain retaining wall.

[0117] 5. In this embodiment of the invention, since the height of the retaining wall in the mining area is lower than the height of the wide-body vehicle in the mining area,

[0118] Therefore, a third preset height threshold is configured based on the height of the retaining wall in the target mining area. If the difference between the maximum and minimum heights within the current grid, i.e., the first height difference, is greater than the third preset height threshold, it indicates that the scan point within the current grid is not a scan point on the retaining wall, thus determining that the current grid is non-retaining.

[0119] Candidate grid. If the difference between the maximum and minimum heights within the current grid, i.e., the first height difference 0, is less than or equal to a third preset height threshold, it indicates that the scan point within the current grid is a scan point on the retaining wall, thus determining the current grid as a candidate grid. The third preset height threshold can be equal to or less than the height of the retaining wall in the target mining area; this embodiment of the invention does not impose specific limitations.

[0120] In this embodiment of the invention, since the mountains in the mining area are natural mountains or excavated sections, their height is much greater than the height of the mine carts. Therefore, a fourth preset height difference threshold of 5 degrees is configured based on the height of the mountains in the target mining area. Furthermore, since mountains generally have a certain length, preset traversal parameters are configured. The currently confirmed candidate grid is used as the traversal starting point, i.e., the head grid, and the last grid to be traversed, i.e., the tail grid, is determined according to the preset traversal parameters. The preset traversal parameters include distance parameters and traversal direction parameters within the point cloud grid. For example, if the distance parameter is 10 grids and the direction parameter is the row, then if the current grid is the first grid, the tail grid to be traversed is the tenth grid. The distance and direction parameters can be customized according to the actual application scenario, and this embodiment of the invention does not impose specific limitations. After determining the head and tail grids, the difference between the maximum height of the head grid and the minimum height of the tail grid is taken to obtain a second height difference value. If the second height difference value is greater than the fourth preset height difference threshold, a second height difference value is obtained.

[0121] If a height difference threshold is set, it indicates that a mountain grid exists within the current traversal range. The retaining wall corresponding to the current candidate grid cell is then classified as a mountain retaining wall. If the second height difference is less than or equal to the fourth preset height difference threshold, it indicates that a mountain grid exists within the current traversal range, and the retaining wall corresponding to the current candidate grid cell is classified as a non-mountain retaining wall. The fourth preset height difference threshold can be equal to or slightly less than the average height of mountains within the target mining area; this embodiment of the invention does not impose specific limitations.

[0122] In one embodiment of the present invention, for further explanation and limitation, the step of determining road boundary point data in at least one orientation based on the clustering of retaining wall points in the set of retaining wall points includes:

[0123] Based on the projection values ​​of the clusters on the horizontal and vertical coordinate axes in the target vehicle coordinate system, and the horizontal coordinate values ​​of the scan points in the clusters, the retaining wall points in the retaining wall point set are divided into retaining wall points in at least one orientation.

[0124] For each retaining wall point in each direction, calculate the concave hull data of the retaining wall point, and configure the five concave hull data as the road boundary point data of the corresponding direction.

[0125] In this embodiment of the invention, Euclidean clustering is performed on the retaining wall points in the retaining wall point set based on a pre-configured clustering radius to obtain multiple clusters. The clustering radius can be customized according to the actual application scenario, and this embodiment of the invention does not impose a specific limitation. For the target vehicle, the rearward radar can...

[0126] The collected scanning points include three directions: left, right, and rear. The rear retaining wall extends for a certain length of 0 degrees. Therefore, the rear retaining wall points are determined based on the projection of the clusters onto the X and Y axes in the target vehicle coordinate system.

[0127] Specifically, the first projection value of the current cluster along the X-axis and the second projection value along the Y-axis are calculated, and the ratio of the second projection value to the first projection value is calculated to obtain the projection ratio. If both the first and second projection values ​​are greater than a first preset projection threshold, then further determination is made.

[0128] The ratio of the first projection value to the second projection value, i.e., the comparison result of the projection ratio and the second preset projection threshold, is used. If either the first projection value or the second projection value is less than the first preset projection threshold, the current cluster is considered an invalid cluster. If the projection ratio is less than the second preset projection threshold, the scan point in the current cluster is determined as the rear retaining wall point. If the projection ratio is greater than or equal to the second preset projection threshold, the scan point with a positive X-coordinate in the current cluster is determined as the right retaining wall point, and the scan point with a negative X-coordinate in the current cluster is determined as the left retaining wall point. The target vehicle coordinate system is a planar coordinate system constructed with the rear axis center point of the target vehicle as the center, the right side of the vehicle as the positive horizontal coordinate axis, and the front direction of the vehicle as the positive vertical coordinate axis. The first preset projection threshold can be 0, and the second preset projection threshold can be 1, or can be customized according to the actual application scenario requirements; this embodiment of the invention does not impose specific limitations. After determining the location of the retaining wall points, concave hull data is calculated for each location. Taking the left retaining wall as an example, we will explain the process. We will build a KD (k-dimensional) tree for the left retaining wall points and calculate the geometric center of all retaining wall points. We will take this geometric center point as the initial point p0. (1) Find the point with the smallest y-axis direction. If the y-direction is the same, find the point with the smallest x-direction and represent this point as p1. (2) With p0 as the center and R as the radius, we will search for the point set A. We will traverse each point in A and set it as p2_i. We will calculate the angle p0->p1->p2_i. If the angle is less than the fourth preset angle threshold, we will continue to traverse the next p2_i. If there is a point with the smallest angle, we will select this point as p2. Otherwise, we will expand the radius R and repeat the current step. (3) We will set the current point p1 and p2 as p0 and p1 respectively and repeat step (2) until we return to the initial point or traverse a circle to obtain the concave hull data.

[0129] This invention provides a method for road boundary identification. In this embodiment, the method acquires path planning data of a target vehicle and point cloud data of the target area collected by the vehicle's rearward monitoring radar. The point cloud data is then rasterized to obtain the extreme height values ​​of the point cloud within each grid cell. Based on the path planning data and the extreme height values ​​of the point cloud within each grid cell, retaining wall candidate grids are extracted and retaining wall categories are detected sequentially to determine the retaining wall candidate grids and retaining wall categories within the target area. Based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories, a set of retaining wall points is extracted. Based on the clusters of retaining wall points in the retaining wall point set, road boundary point data in at least one direction is determined and sent to the control terminal of the target vehicle. This allows the control terminal to perform rearward road boundary identification based on the road boundary point data. This significantly reduces the limitations of grids on retaining wall point extraction, greatly improving the accuracy of retaining wall point extraction. Furthermore, different retaining wall point extraction strategies are applied to different retaining wall categories, effectively improving the accuracy of retaining wall point identification and thus improving the accuracy of road boundary identification.

[0130] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides a road boundary identification device, such as... Figure 4 As shown, the device includes:

[0131] The acquisition module 31 is used to acquire the path planning data of the target vehicle and the point cloud data of the target area collected by the rear monitoring radar of the target vehicle, and to perform rasterization processing on the point cloud data to obtain the extreme value of the height of the point cloud in each grid in the planar point cloud grid.

[0132] The extraction module 32 is used to extract retaining wall candidate grids and detect retaining wall categories in sequence based on the path planning data and the height extreme values ​​of the point cloud in each grid, to determine the retaining wall candidate grids and retaining wall categories in the target area, and to extract a retaining wall point set based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall category.

[0133] The determination module 33 is used to determine road boundary point data in at least one direction based on the cluster of retaining wall points in the set of retaining wall points, and send the road boundary point data to the control terminal of the target vehicle so that the control terminal can perform backward road boundary identification based on the road boundary point data.

[0134] Furthermore, the extraction module 32 includes:

[0135] The first fitting unit is used to obtain a first reference point cloud plane by performing planar fitting on the first reference point cloud of the candidate grid, and to calculate the first angle between the normal vector of the first reference point cloud plane and the bottom plane of the target vehicle.

[0136] The first determining unit is used to determine the retaining wall grid from the candidate grid based on the comparison result between the first included angle and the preset included angle threshold, wherein the preset included angle threshold is determined based on the retaining wall category;

[0137] The second fitting unit is used to obtain a second reference point cloud plane by performing planar fitting on the second reference point cloud of each scan point in the retaining wall grid, and to calculate the second angle between the normal vector of the second reference point cloud plane and the bottom plane of the target vehicle.

[0138] The second determining unit is used to determine the retaining wall points in the retaining wall grid based on the comparison result between the second included angle and the preset included angle threshold, and to construct a retaining wall point set based on all the retaining wall points in the retaining wall grid.

[0139] Furthermore, the device also includes:

[0140] The first partitioning module is used to calculate the centroid of the point cloud in the candidate grid, and to partition the associated point cloud with the centroid as the center and the first associated point cloud partitioning parameter as the radius to obtain the first associated point cloud;

[0141] The first calculation module is used to calculate the first height difference between each scan point in the first associated point cloud and the centroid point, and to extract the first reference point cloud from the first associated point cloud based on the comparison result of the first height difference and the first preset height difference threshold.

[0142] The second partitioning module is used to partition the associated point cloud with each target scanning point in the retaining wall grid as the center and the second associated point cloud partitioning parameter as the radius, so as to obtain the second associated point cloud.

[0143] The second calculation module is used to calculate the second height difference between each scan point in the second associated point cloud and the target scan point, and to extract the second reference point cloud from the second associated point cloud based on the comparison result of the second height difference and the second preset height difference threshold.

[0144] Furthermore, the extraction module 32 includes:

[0145] The third determining unit is used to determine the retaining wall grid based on the comparison result between the first included angle and the first preset included angle threshold if the retaining wall type is a mountain retaining wall.

[0146] The fourth determining unit is used to determine the retaining wall grid based on the comparison results of the first included angle with the second preset included angle threshold and the third preset included angle threshold if the retaining wall type is a non-mountain retaining wall.

[0147] Furthermore, the acquisition module 31 includes:

[0148] The first calculation unit is used to calculate the grid coordinates of each scan point in the point cloud data in the preset planar grid based on the grid size parameters of the preset planar grid, wherein each grid in the preset planar grid is configured with an atomic lock;

[0149] The allocation unit is used to allocate corresponding threads to different point clouds in the point cloud data, and write the scan points in the different point clouds in parallel into a grid that matches the grid coordinates of the scan points based on the threads, so as to obtain the point cloud grid of the point cloud data.

[0150] The fifth determining unit is used to determine the maximum height and minimum height of each grid cell based on the height values ​​of all scanned points in each grid cell of the point cloud grid, wherein the height value is the height of the scanned point in the normal direction of the preset plane grid.

[0151] Furthermore, the determining module 33 includes:

[0152] The sixth determining unit is used to calculate the first height difference of each grid cell based on the extreme height of the point cloud in each grid cell. If the first height difference is less than the third preset height threshold, the grid cell is determined as a candidate grid cell for the retaining wall in the target area. The third preset height threshold is configured based on the height of the retaining wall in the target mining area.

[0153] The traversal unit is used to take the candidate grid as the head grid for traversal and determine the tail grid for traversal according to preset traversal parameters, including the distance parameter and the direction parameter for traversal in the point cloud grid.

[0154] The second calculation unit is used to calculate a second height difference based on the height extreme value of the head grid and the height extreme value of the tail grid. If the second height difference is greater than a fourth preset height threshold, the retaining wall category of the retaining wall in the target area is determined to be a mountain retaining wall. The fourth preset height threshold is configured based on the height of the mountain in the target mining area.

[0155] Furthermore, the determining module 33 includes:

[0156] The partitioning unit is used to divide the retaining wall points in the retaining wall point set into retaining wall points in at least one orientation based on the projection values ​​of the clusters on the horizontal and vertical coordinate axes in the target vehicle coordinate system and the horizontal coordinate values ​​of the scan points in the clusters. The target vehicle coordinate system is a planar coordinate system constructed with the center point of the rear axis of the target vehicle as the center, the right side of the vehicle as the positive direction of the horizontal coordinate axis, and the front direction of the vehicle as the positive direction of the vertical coordinate axis.

[0157] The third calculation unit is used to calculate the concave data of the retaining wall point for each direction, and configure the concave data as the road boundary point data of the corresponding direction.

[0158] This invention provides a road boundary identification device. In its embodiments, the device acquires path planning data of a target vehicle and point cloud data of the target area collected by the vehicle's rearward monitoring radar. The point cloud data is then rasterized to obtain the extreme height values ​​of the point cloud within each grid cell. Based on the path planning data and the extreme height values ​​of the point cloud within each grid cell, retaining wall candidate grids are extracted and retaining wall categories are detected sequentially to determine the retaining wall candidate grids and retaining wall categories within the target area. Based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories, a set of retaining wall points is extracted. Based on the clusters of retaining wall points in the retaining wall point set, road boundary point data in at least one direction is determined, and this road boundary point data is sent to the control terminal of the target vehicle. This allows the control terminal to perform rearward road boundary identification based on the road boundary point data. This significantly reduces the limitations of grids on retaining wall point extraction, greatly improving the accuracy of retaining wall point extraction. Furthermore, different retaining wall point extraction strategies are applied to different retaining wall categories, effectively improving the accuracy of retaining wall point identification and thus enhancing the accuracy of road boundary identification.

[0159] According to one embodiment of the present invention, a storage medium is provided, the storage medium storing at least one executable instruction, the computer-executable instruction being capable of executing the road boundary identification method in any of the above method embodiments.

[0160] Figure 5 The diagram shows a structural schematic of a terminal according to an embodiment of the present invention. The specific implementation of the terminal is not limited by the specific embodiments of the present invention.

[0161] like Figure 5 As shown, the terminal may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0162] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.

[0163] Communication interface 404 is used to communicate with other network elements such as clients or other servers.

[0164] The processor 402 is used to execute program 410, specifically to execute the relevant steps in the above-described road boundary recognition method embodiment.

[0165] Specifically, program 410 may include program code that includes computer operation instructions.

[0166] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The terminal may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0167] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0168] Specifically, program 410 can be used to cause processor 402 to perform the following operations:

[0169] The path planning data of the target vehicle and the point cloud data of the target area collected by the rearward monitoring radar of the target vehicle are obtained, and the point cloud data is rasterized to obtain the extreme height value of the point cloud in each grid in the planar point cloud grid.

[0170] Based on the path planning data and the extreme height values ​​of the point cloud within each grid, retaining wall candidate grids are extracted and retaining wall categories are detected sequentially to determine the retaining wall candidate grids and retaining wall categories within the target area. Based on the retaining wall candidate grids and the retaining wall point extraction strategy corresponding to the retaining wall categories, a set of retaining wall points is extracted.

[0171] Based on the clustering of retaining wall points in the set of retaining wall points, road boundary point data in at least one direction is determined, and the road boundary point data is sent to the control terminal of the target vehicle so that the control terminal can perform backward road boundary identification based on the road boundary point data.

[0172] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0173] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of recognizing a road boundary, characterized by, The method comprises the following steps: acquire path planning data of a target vehicle and point cloud data of a target area collected by a rear monitoring radar of the target vehicle, and perform rasterization processing on the point cloud data to obtain height extreme values of point clouds in each grid in a planar point cloud grid; based on the path planning data and the height extreme values of the point clouds in each grid, perform barrier candidate grid extraction and barrier category detection in sequence to determine barrier candidate grids and barrier categories in the target area, and based on the barrier candidate grids and barrier point extraction strategies corresponding to the barrier categories, extract a barrier point set, including: performing planar fitting on first reference point clouds of the barrier candidate grids to obtain first reference point cloud planes, and calculating a first included angle between a normal vector of the first reference point cloud planes and a plane of the target vehicle; based on a comparison result of the first included angle and a preset included angle threshold, determine barrier grids from the barrier candidate grids, including: if the barrier category is a mountain barrier, determine barrier grids based on a comparison result of the first included angle and a first preset included angle threshold; if the barrier category is a non-mountain barrier, determine barrier grids based on comparison results of the first included angle and a second preset included angle threshold and a third preset included angle threshold, the preset included angle threshold being determined based on the barrier category; perform planar fitting on second reference point clouds of each scanning point in the barrier grids to obtain second reference point cloud planes, and calculate a second included angle between a normal vector of the second reference point cloud planes and a bottom plane of the target vehicle; based on a comparison result of the second included angle and the preset included angle threshold, determine barrier points in the barrier grids, and based on barrier points in all the barrier grids, construct a barrier point set; based on clustering of barrier points in the barrier point set, determine road boundary point data of at least one orientation, and send the road boundary point data to a control end of the target vehicle to enable the control end to perform rear road boundary identification based on the road boundary point data.

2. The method of claim 1, wherein, Before the planar fitting on the first reference point clouds of the barrier candidate grids, the method further comprises: calculate a centroid point of the point clouds in the barrier candidate grids, and perform associated point cloud division with the centroid point as the center and a first associated point cloud division parameter as the radius to obtain a first associated point cloud; calculate a first height difference value between each scanning point in the first associated point cloud and the centroid point, and based on a comparison result of the first height difference value and a first preset height difference threshold, extract first reference point clouds from the first associated point cloud; Before the planar fitting on the second reference point clouds of each scanning point in the barrier grids, the method further comprises: perform associated point cloud division with each target scanning point in the barrier grids as the center and a second associated point cloud division parameter as the radius to obtain a second associated point cloud; Calculate a second height difference value of each scanning point in the second associated point cloud and the target scanning point, and extract a second reference point cloud from the second associated point cloud based on a comparison result of the second height difference value and a second preset height difference threshold value.

3. The method of claim 1, wherein, The height extreme value includes a height maximum value and a height minimum value, and the point cloud data is rasterized to obtain the height extreme value of the point cloud in each grid. Based on a preset plane grid size parameter, the grid coordinates of each scanning point in the point cloud data in the preset plane grid are calculated, and each grid in the preset plane grid is configured with an atomic lock; Different point clouds in the point cloud data are assigned corresponding threads, and the scanning points in the different point clouds are written into the grid matching the grid coordinates of the scanning points in parallel based on the threads, to obtain the point cloud grid of the point cloud data. Based on the height values of all scanning points in each grid in the point cloud grid, the height maximum value and the height minimum value in each grid are determined, and the height value is the height of the scanning point in the normal direction of the preset plane grid.

4. The method of claim 1, wherein, Based on the path planning data and the height extreme value of the point cloud in each grid, the dam candidate grid extraction and dam type detection are sequentially performed to determine the dam candidate grid and the dam type in the target region. Based on the height extreme value of the point cloud in each grid, a first height difference value of each grid is calculated, and if the first height difference value is less than a third preset height threshold value, the grid is determined as a dam candidate grid of the dam in the target region, and the third preset height threshold value is configured based on the height of the dam in the target mining area. The dam candidate grid is taken as a head grid for traversal, and a tail grid for traversal is determined according to a preset traversal parameter, and the preset traversal parameter includes a distance parameter for traversal in the point cloud grid and a direction parameter for traversal. Based on the height extreme value of the head grid and the height extreme value of the tail grid, a second height difference value is calculated, and if the second height difference value is greater than a fourth preset height threshold value, it is determined that the dam type of the dam in the target region is a mountain dam, and the fourth preset height threshold value is configured based on the height of the mountain in the target mining area.

5. The method of claim 1, wherein, The determination of the road boundary point data of at least one orientation based on the clustering cluster of the dam point in the dam point set includes: Based on the projection value of the clustering cluster in the horizontal coordinate axis and the vertical coordinate axis in the target vehicle coordinate system, and the horizontal coordinate value of the scanning point in the clustering cluster, the dam points in the dam point set are divided into at least one orientation of dam points, and the target vehicle coordinate system is a plane coordinate system with the rear axle center point of the target vehicle as the center, the right side of the vehicle as the positive direction of the horizontal coordinate axis, and the head direction as the positive direction of the vertical coordinate axis. For each orientation of the dam point, the concave hull data of the dam point is calculated respectively, and the concave hull data is configured as the road boundary point data of the corresponding orientation.

6. An apparatus for recognizing a road boundary, characterized by comprising: It includes: The acquisition module is configured to acquire path planning data of a target vehicle and point cloud data of a target region collected by a rear monitoring radar of the target vehicle, and perform rasterization processing on the point cloud data to obtain height extreme values of point clouds in each grid in a planar point cloud grid. The extraction module is configured to perform, based on the path planning data and the height extreme values of the point clouds in each grid, wall candidate grid extraction and wall type detection in sequence to determine wall candidate grids and wall types in the target region, and extract a wall point set based on the wall candidate grids and a wall point extraction strategy corresponding to the wall types, including: obtaining a first reference point cloud plane by performing planar fitting on first reference point clouds of the wall candidate grids, and calculating a first included angle between a normal vector of the first reference point cloud plane and a plane of the target vehicle; determining a wall grid from the wall candidate grids based on a comparison result of the first included angle and a preset included angle threshold, including: if the wall type is a mountain wall, determining the wall grid based on a comparison result of the first included angle and a first preset included angle threshold; if the wall type is a non-mountain wall, determining the wall grid based on comparison results of the first included angle and a second preset included angle threshold and a third preset included angle threshold, the preset included angle threshold being determined based on the wall type; obtaining a second reference point cloud plane by performing planar fitting on second reference point clouds of each scanning point in the wall grid, and calculating a second included angle between a normal vector of the second reference point cloud plane and a bottom plane of the target vehicle; determining wall points in the wall grid based on a comparison result of the second included angle and the preset included angle threshold, and constructing the wall point set based on wall points in all the wall grids. The determination module is configured to determine road boundary point data of at least one orientation based on clustering clusters of wall points in the wall point set, and send the road boundary point data to a control end of the target vehicle to enable the control end to perform rear road boundary identification based on the road boundary point data.

7. A storage medium having at least one executable instruction stored therein, the executable instruction causing a processor to perform operations corresponding to the road boundary identification method of any one of claims 1-5.

8. A terminal comprising: A processor, a memory, a communication interface, and a communication bus, the processor, the memory, and the communication interface performing communication with each other through the communication bus; The memory is configured to store at least one executable instruction, the executable instruction causing the processor to perform operations corresponding to the road boundary identification method of any one of claims 1-5.

Images