A method, device and storage medium for lidar point clustering

Abstract

The present application discloses a method, device, and storage medium for lidar point clustering, belonging to the technical field of lidar in autonomous driving. The method includes: projecting non-ground point clouds scanned by the lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial division to obtain projection points corresponding to the non-ground point clouds in the cylindrical polar coordinate system, where the three-dimensional cylindrical polar coordinate system grid is obtained by dividing the cylindrical polar coordinate system; obtaining a projection point grid in the three-dimensional cylindrical polar coordinate system grid that contains the projection points according to the coordinate information of the projection points in the cylindrical polar coordinate system; and clustering the non-ground point clouds according to the positions of the projection point grids in the three-dimensional cylindrical polar coordinate system grid. By constructing the cylindrical polar coordinate system grid, lidar point cloud clustering is realized, the accuracy of target point cloud clustering is improved, and the problems of incorrect clustering of nearby target point clouds and easy filtering of distant targets are solved.

Description

Technical Field

[0001] This application relates to the technical field of lidar in autonomous driving, and particularly to a method, device, and storage medium for lidar point clustering. Background Art

[0002] When extracting targets with lidar, clustering is a very important step because lidar can only obtain a point cloud data set, and the ultimate goal is to process these point cloud data sets to obtain targets. Therefore, a clustering method must be adopted to process the point cloud into targets. If the point cloud is not clustered into targets, then in the next step of target tracking, the tracking cannot be carried out due to the excessive amount of data, making the results of the entire lidar inapplicable to autonomous driving.

[0003] In existing clustering methods, there are the iterative closest point method for direct clustering and the ordinary grid method. The iterative closest point method has a very high time complexity. The ordinary grid method projects the point cloud using a grid, replaces the point cloud with the grid, reduces the number of traversals, and can solve the problem of large computational volume. However, in places far away, it is easy to miss distant targets; in places close by, the targets cannot be correctly clustered. Summary of the Invention

[0004] Aiming at the problems of incorrect clustering of near target point clouds and easy filtering of distant targets currently, this application provides a method, device, and storage medium for lidar point clustering to solve the problems of incorrect clustering of near target point clouds and easy filtering of distant targets.

[0005] To achieve the above object, a technical solution adopted by this application is: to provide a method for lidar point clustering, which includes: projecting the non-ground point cloud scanned by the lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution to obtain the projection points corresponding to the non-ground point cloud in the cylindrical polar coordinate system, where the three-dimensional cylindrical polar coordinate system grid is obtained by dividing the cylindrical polar coordinate system; obtaining the projection point grid containing the projection points in the three-dimensional cylindrical polar coordinate system grid according to the coordinate information of the projection points in the cylindrical polar coordinate system; and clustering the non-ground point cloud according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid.

[0006] Another technical solution adopted by this application is: to provide a device for lidar point clustering, which includes: a module for projecting non-ground point clouds scanned by a lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution to obtain projection points corresponding to the non-ground point clouds in the cylindrical polar coordinate system, where the three-dimensional cylindrical polar coordinate system grid is obtained by dividing the cylindrical polar coordinate system; a module for obtaining a projection point grid in the three-dimensional cylindrical polar coordinate system that contains the projection points according to the coordinate information of the projection points in the cylindrical polar coordinate system; and a module for clustering the non-ground point clouds according to the positions of the projection point grids in the three-dimensional cylindrical polar coordinate system grid.

[0007] Another technical solution adopted by this application is: to provide a computer-readable storage medium that stores computer instructions, and the computer instructions are operated to execute the method for lidar point clustering in Solution 1.

[0008] The beneficial effects of this application are: to provide a method, device, and storage medium for lidar point clustering, which project the point clouds scanned by the lidar by using the method of a three-dimensional cylindrical polar coordinate grid. In the vicinity, small grids are used to solve the problem of dense point clouds nearby, and the point clouds of two targets will not be projected together; in the distance, large grids are used to increase the number of point clouds in each grid, and it is not easy to miss distant targets. The correct clustering of the target point clouds is achieved, and the corresponding targets are obtained accordingly. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1 It is a schematic flowchart of a specific implementation of the method for lidar point clustering in this application;

[0011] Figure 2 It is a planar distribution diagram of a three-dimensional cylindrical polar coordinate grid of a specific example of the method for lidar point clustering in this application;

[0012] Figure 3 It is a schematic diagram of occluded clustering of a specific example of the method for lidar point clustering in this application;

[0013] Figure 4 It is a schematic diagram of clustering split due to occlusion of a specific example of the method for lidar point clustering in this application;

[0014] Figure 5It is a schematic diagram composed of a specific embodiment of the lidar point clustering device of the present application.

[0015] Through the above-mentioned drawings, specific embodiments of the present application have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. Specific Embodiment

[0016] The following elaborates on the preferred embodiments of the present application in conjunction with the drawings, so that the advantages and features of the present application can be more easily understood by those skilled in the art, thereby making the scope of protection of the present application more clearly defined.

[0017] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, elements defined by the statement "including..." do not exclude the presence of additional identical elements in the process, method, article or device including the said elements.

[0018] In this solution, through analysis, it is found that in the existing clustering methods: the iterative closest point method has a very high time complexity. This is because the data volume of the lidar point cloud is very large. The single-frame point cloud of a 16-line lidar can reach 70,000 points. Even after removing the ground, there are still more than 10,000 points in a single frame. The iterative closest point method needs to traverse all the points in the point cloud every time, with a large amount of calculation and a slow calculation speed, which cannot meet the real-time requirement. While the ordinary grid method can solve the problem of large calculation amount. However, in the far distance, due to the sparse point cloud, after replacing the point cloud with a grid, because the number of points in the point cloud is small, it is relatively easy to filter out the grids in the distance, and it is very easy to miss the targets in the distance. In the near distance, due to the dense point cloud, it is relatively easy to project the point clouds of two targets onto the same grid, making the targets unable to be correctly clustered. Therefore, in combination with the defects in the above-mentioned existing technologies, the solution of this solution is as Figure 1 shown, providing a specific embodiment of the lidar point clustering method.

[0019] In Figure 1 the specific embodiment shown, the lidar point clustering method of the present application includes process S101, process S102 and process S103.

[0020] Figure 1 The process S101 shown includes: projecting the non-ground point cloud scanned by the lidar onto a pre-established three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution to obtain the corresponding projection points of the non-ground point cloud in the cylindrical polar coordinates, where the three-dimensional cylindrical polar coordinate grid is obtained by dividing the cylindrical polar coordinate system.

[0021] In this specific embodiment, preferably, a cylindrical polar coordinate grid is pre-established with the position of the lidar as the origin, and the cylindrical polar coordinate system is subjected to non-uniform grid division to obtain a three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution. The non-ground point cloud scanned by the lidar is projected onto the pre-established three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution to obtain the corresponding projection points of the non-ground point cloud in the cylindrical polar coordinates. Due to the characteristics that the point cloud scanned by the lidar is dense when close to the lidar and sparse when far from the lidar, by using a three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution, a small three-dimensional cylindrical polar coordinate grid is used when close to the lidar to avoid the coincidence of the corresponding projection points of the non-ground point cloud, and a large three-dimensional cylindrical polar coordinate grid is used when far from the lidar to increase the number of corresponding projection points of the non-ground point cloud in each three-dimensional cylindrical polar coordinate grid and avoid missing targets far from the lidar.

[0022] In a specific embodiment of the present application, the pre-establishment process of the pre-established three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution includes: taking the position of the lidar as the origin and establishing a cylindrical polar coordinate system with a radius of R on the horizontal plane, where the radius R is greater than or equal to the scanning range of the lidar.

[0023] In a specific example of the present application, taking the position of the lidar as the origin, a cylindrical polar coordinate system is established on the horizontal plane; wherein, the horizontal coordinates of the cylindrical polar coordinate system use the angle and distance as dimensions respectively. Since the point cloud of the lidar is sparse at a long distance, using the cylindrical polar coordinate system can combine with this characteristic, and the effect is better than the traditional method using the rectangular coordinate system.

[0024] In a specific embodiment of the present application, the pre-establishment process of the pre-established three-dimensional cylindrical polar coordinate grid with non-uniform radial distribution further includes: according to the angular dimension of the cylindrical polar coordinate system, dividing the cylindrical polar coordinate system into M three-dimensional solids with a sector-shaped bottom and a height, where M is an integer not less than 2.

[0025] In a specific example of the present application, taking the position of the lidar as the origin, a cylindrical polar coordinate system with a radius of R is established. According to the angular dimension of the cylindrical polar coordinate system, the cylindrical polar coordinate system is divided into M three-dimensional solids with a fan-shaped base. For example, on the angular dimension of the cylindrical polar coordinate system, a uniform division is performed to obtain M three-dimensional solids with a fan-shaped base and a height, where the central angle of each fan is Δα = 360° / M.

[0026] In a specific embodiment of the present application, the process of pre-establishing the pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution further includes: in each solid, according to the distance dimension of the cylindrical polar coordinate system, the area from the origin of the cylindrical polar coordinate system to a region R meters away from the origin of the cylindrical polar coordinate system is divided to obtain a three-dimensional cylindrical polar coordinate system grid, where the farther away from the origin of the cylindrical polar coordinate system, the larger the width of the three-dimensional cylindrical polar coordinate system grid in the distance dimension.

[0027] In a specific example of the present application, in each solid, the area within the range from the origin of the cylindrical polar coordinate system to R meters is divided into a plurality of grids, where the farther away from the origin of the cylindrical polar coordinate system, the larger the width of the grid; that is, after division on the angular dimension of the cylindrical polar coordinate system, non-uniform division is performed on the distance dimension of the cylindrical polar coordinate system, so that the divided three-dimensional cylindrical polar coordinate system grid is a non-uniformly distributed grid on the plane. Among them, according to the order from near to far from the origin, in any two grids, the width of the grid closer to the origin is not greater than the width of the grid farther from the origin. Since when the distance from the origin is relatively close, the point cloud of the lidar is relatively dense, smaller grids can be used to mark the point cloud, and when the distance from the origin is far, the lidar point cloud is relatively sparse, larger grids can be used to mark the point cloud. Therefore, when dividing the grid of the cylindrical polar coordinates, the non-uniform division method can be used according to the characteristics of the lidar point cloud to obtain a three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution, which is convenient for subsequent correct clustering of the lidar point cloud.

[0028] In a specific example of the present application, preferably, as Figure 2As shown, within the range scanned by the lidar, a uniform division is performed in the angular dimension of the cylindrical polar coordinate system. The angular dimension of the cylindrical polar coordinate system is divided into 12 sectors with equal circumferential angles, where each sector has an angle of 30 degrees. The first angle is within the range of 0 - 30 degrees in the distance - angle coordinate information, and the angle number 1 is used for the first angle; that is, the angle numbers of the projection points within the angular dimension coordinate information of 0 - 30 degrees are all 1. The second angle is within the range of 30 - 60 degrees in the distance - angle coordinate information, and the angle number 2 is used for the second angle; that is, the angle numbers of the projection points within the angular dimension coordinate information of 30 - 60 degrees are all 2. The third angle is within the range of 60 - 90 degrees in the distance - angle coordinate information, and the angle number 3 is used for the third angle; that is, the angle numbers of the projection points within the angular dimension coordinate information of 60 - 90 degrees are all 3. The fourth angle is within the range of 90 - 120 degrees in the distance - angle coordinate information, and the angle number 4 is used for the fourth angle; that is, the angle numbers of the projection points within the angular dimension coordinate information of 90 - 120 degrees are all 4. The fifth angle is within the range of 120 - 150 degrees in the distance - angle coordinate information, and the angle number 5 is used for the fifth angle; that is, the angle numbers of the projection points within the angular dimension coordinate information of 120 - 150 degrees are all 5. And so on, the angle numbers of 12 angles are obtained respectively.

[0029] After the division in the angular dimension is completed, in the distance dimension, according to the distance from the origin of the lidar from near to far, the division is carried out in the distance dimension of the cylindrical polar coordinate system by the method that the width of the next distance grid = the width of the previous distance grid + a fixed difference (it can also be a non-fixed variable difference), so that the grid farther from the origin of the lidar has a larger width in the distance dimension. Taking the fixed difference as an example, when 5 meters is used as the fixed difference, the width of the first distance grid in the distance dimension is 5 meters, and the distance number 1 is used for the first distance grid; that is, the three-dimensional cylindrical polar coordinate system grid where the projection points with distance dimension coordinate information of 0 - 5 meters are located is the first distance grid, and the angle numbers of the projections in the first distance grid are all 1. The width of the second distance grid in the distance dimension is 10 meters, and the distance number 2 is used for the second distance grid; that is, the three-dimensional cylindrical polar coordinate system grid where the projection points with distance dimension coordinate information of 5 - 15 meters are located is the second distance grid, and the angle numbers of the projections in the second distance grid are all 2. The width of the third distance grid in the distance dimension is 15 meters, and the distance number 3 is used for the third distance grid; that is, the three-dimensional cylindrical polar coordinate system grid where the projection points with distance dimension coordinate information of 15 - 30 meters are located is the third distance grid, and the angle numbers of the projections in the third distance grid are all 3. The width of the fourth distance grid in the distance dimension is 20 meters, and the distance number 4 is used for the fourth distance grid; that is, the three-dimensional cylindrical polar coordinate system grid where the projection points with distance dimension coordinate information of 30 - 50 meters are located is the fourth distance grid, and the angle numbers of the projections in the fourth distance grid are all 4. The width of the fifth distance grid in the distance dimension is 25 meters, and the distance number 5 is used for the fifth distance grid; that is, the three-dimensional cylindrical polar coordinate system grid where the projection points with distance dimension coordinate information of 50 - 75 meters are located is the fifth distance grid, and the angle numbers of the projections in the fifth distance grid are all 5. And so on, the division of the three-dimensional cylindrical polar coordinate system grid in the distance dimension is completed. In the above process of dividing the distance dimension, the angle of each distance grid is 30 degrees. The above grid division method, combined with the characteristic that the point cloud of the lidar is sparser when it is farther from the origin of the lidar, can correctly and truly record the point cloud and facilitate the subsequent clustering process.

[0030] Figure 1 The process S102 shown includes: obtaining the projection point grid containing the projection point in the three-dimensional cylindrical polar coordinate system grid according to the coordinate information of the projection point in the cylindrical polar coordinate system.

[0031] In a specific embodiment of the present application, the process of obtaining the projection point grid containing the projection point in the three-dimensional cylindrical polar coordinate system grid according to the coordinate information of the projection point in the cylindrical polar coordinate system includes: respectively retrieving the projection point in the cylindrical polar coordinate grid in the angular dimension and the distance dimension, and determining the three-dimensional cylindrical polar coordinate system grid with the projection point as the projection point grid.

[0032] In a specific example of the present application, preferably, the three-dimensional cylindrical polar coordinate system grid can be numbered to obtain the distance number and angle number of each three-dimensional cylindrical polar coordinate system grid; starting from the grid with a distance number of 1 and an angle number of 1, retrieve the three-dimensional cylindrical polar coordinate system grid where the projection point corresponding to the non-ground point cloud is located respectively from the angle dimension and the distance dimension, and determine the three-dimensional cylindrical polar coordinate system grid with the projection point corresponding to the non-ground point cloud as the projection point grid, and mark the projection point grid.

[0033] Figure 1 The process S103 shown includes: clustering the non-ground point cloud according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid.

[0034] In this specific embodiment, preferably, the three-dimensional cylindrical polar coordinate system grid is numbered respectively from the angle dimension and the distance dimension to obtain the angle dimension number and distance dimension number of each three-dimensional cylindrical polar coordinate system grid. According to whether there is a projection point in the three-dimensional cylindrical polar coordinate system grid, determine the three-dimensional cylindrical polar coordinate system grid with the projection point as the projection point grid; among them, the angle dimension number and distance dimension number of the projection point grid do not change after the determination. Use the angle dimension number and distance dimension number of the projection point grid as its position in the three-dimensional cylindrical polar coordinate system grid to cluster the non-ground point cloud corresponding to the projection point in the projection point grid. Since the point cloud of the lidar is denser when it is closer to the origin and sparser when it is farther from the origin, an uneven distribution occurs. Therefore, when dividing the three-dimensional cylindrical polar coordinate system grid unevenly, the characteristics of the lidar point cloud can be utilized to adopt an uneven division method to obtain a three-dimensional cylindrical polar coordinate system grid with a non-uniform radial distribution, which is convenient for the subsequent correct clustering of the lidar point cloud. For example, if the coordinate information of the projection point corresponding to a certain non-ground point in the cylindrical polar coordinate system is: distance 13.8 meters, angle 156.4 degrees, according to Figure 2 the shown three-dimensional cylindrical polar coordinate system grid plane distribution diagram, the number of the three-dimensional cylindrical polar coordinate system grid where the projection point corresponding to this non-ground point is located is: the distance dimension number is 2, and the angle dimension number is 6; that is, the grid number of the projection point where the projection point corresponding to this non-ground point is located is: the distance dimension number is 2, and the angle dimension number is 6. Obtain the grid numbers of the projection point grids where each projection point is located according to the coordinate information of the projection points corresponding to each non-ground point in the cylindrical polar coordinate system, and make corresponding judgments on each adjacent numbered projection point grid according to the numbers of each projection point grid, so as to realize the clustering of the non-ground point cloud corresponding to the projection point. Since the present application uses a three-dimensional cylindrical polar coordinate system grid with a non-uniform radial distribution, it realizes the correct clustering of the non-ground point cloud corresponding to the projection point, avoiding the situation of incorrect clustering due to the coincidence of projection points corresponding to incorrect non-ground point clouds.

[0035] In a specific embodiment of the present application, the process of clustering non-ground point clouds according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid includes: retrieving the projection point grid in the three-dimensional cylindrical polar coordinate system grid respectively according to the angular dimension and the distance dimension to obtain multiple groups of adjacent projection point grids.

[0036] In a specific example of the present application, preferably, when searching for adjacent projection point grids of a certain projection point grid, the breadth-first algorithm is used. First, according to the angular dimension of the cylindrical polar coordinate system, search for grids with adjacent numbers in the angular dimension in the horizontal coordinate where the cylindrical polar coordinate system is located. Then, according to the distance dimension of the cylindrical polar coordinate system, search for grids with adjacent numbers in the distance dimension. If the grid with an adjacent number searched in the angular dimension of the cylindrical polar coordinate system is a non-projection point grid, stop the search in the angular dimension of the cylindrical polar coordinate system and start searching for grids with adjacent numbers of this projection point grid in the distance dimension of the cylindrical polar coordinate system.

[0037] For example, if the distance number of the three-dimensional cylindrical polar coordinate system grid where the projection point corresponding to a certain non-ground point is located is 24 and the angle number is 35; when searching for adjacent numbered grids, first search for the grid with a distance number of 24 and an angle number of 36 in the three-dimensional cylindrical polar coordinate system grid, and so on, until there is no projection point corresponding to the non-ground point in the grids with adjacent angle numbers in the three-dimensional cylindrical polar coordinate system grid. Then start searching for the grid with a distance number of 25 and an angle number of 35 in the three-dimensional cylindrical polar coordinate system grid, and so on, until there is no projection point corresponding to the non-ground point in the grids with adjacent distance numbers in the three-dimensional cylindrical polar coordinate system grid. All the projection point grids with adjacent numbers searched to the grid with a distance number of 24 and an angle number of 35 are used as the adjacent projection point grids of this projection point grid.

[0038] In a specific embodiment of the present application, the process of clustering non-ground point clouds according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid further includes: in each group of adjacent projection point grids, calculating the height difference between other projection point grids and any projection point grid according to the height of any projection point grid.

[0039] In a specific embodiment of the present application, the process of calculating the height difference between other projection point grids and any projection point grid according to the height of any projection point grid in each group of adjacent projection point grids includes: determining the height of the projection point grid according to the height of the non-ground point cloud points corresponding to the projection points in the projection point grid, where the highest height of the non-ground point cloud points corresponding to the projection points in the projection point grid is the highest height of this projection point grid, and the lowest height of the non-ground point cloud points corresponding to the projection points in the projection point grid is the lowest height of this projection point grid.

[0040] In a specific example of the present application, preferably, in a set of adjacent projection point grids, any one of the projection point grids is selected, and the height differences between the selected projection point grid and other projection point grids in the group where the selected projection point grid is located are calculated respectively. During the calculation of the height differences, the height of the non-ground point cloud point with the highest height among the non-ground point cloud points corresponding to the projection points of each projection point grid is the highest height of the projection point grid where the non-ground point cloud point corresponding to the projection point is located; the height of the non-ground point cloud point with the lowest height among the non-ground point cloud points corresponding to the projection points of each projection point grid in the two projection point grids is the lowest height of the projection point grid where the non-ground point cloud point corresponding to the projection point is located. The highest height and the lowest height of the projection point grid are respectively used to calculate the height differences between the selected projection point grid and other projection point grids in the group where the selected projection point grid is located.

[0041] In a specific embodiment of the present application, in each set of adjacent projection point grids, the process of calculating the height differences between any projection point grid and other projection point grids according to the height of any projection point grid further includes: in each set of adjacent projection point grids, comparing the highest height of any projection point grid with the highest heights of other projection point grids to obtain the higher grids and lower grids of other projection point grids relative to any projection point grid; calculating the first height difference between the lowest height of the higher grid and the highest height of any projection point grid and the second height difference between the highest height of the lower grid and the lowest height of any projection point grid respectively, where the height differences include the first height difference and the second height difference.

[0042] In a specific example of the present application, preferably, in a set of adjacent projection point grids, the highest height of the selected projection point grid is respectively compared with the highest height of each projection point grid in other projection point grids. When the highest height of the first projection point grid in other projection point grids is greater than the highest height of the selected projection point grid, the projection point grid is determined as the higher grid, and at this time the selected projection point grid is used as the lower grid; when the highest height of the second projection point grid in other projection point grids is less than the highest height of the selected projection point grid, the projection point grid is determined as the lower grid, and at this time the selected projection point grid is used as the higher grid. In the above two cases, when the selected projection point grid is used as the lower grid, calculate the height difference between the lowest height of the higher grid and the highest height of the selected projection point grid, and use it as the first height difference. When the selected projection point grid is used as the higher grid, calculate the height difference between the highest height of the lower grid and the lowest height of the selected projection point grid, and use it as the second height difference, and use the obtained first height difference and second height difference as subsequent clustering conditions.

[0043] In a specific embodiment of the present application, the process of clustering non-ground point clouds according to the positions of the projection point grids in the three-dimensional columnar polar coordinate system grids further includes: in each group of adjacent projection point grids, clustering the non-ground point clouds corresponding to the projection points in other projection point grids with a height difference less than or equal to a height threshold and the non-ground point clouds corresponding to the projection points in any one projection point grid into one category.

[0044] In a specific example of the present application, preferably, in a certain group of adjacent projection point grids, based on the first height difference and the second height difference obtained above, determine between the first height difference and the second height difference and the height threshold respectively. If both the first height difference and the second height difference are greater than the height threshold, it is determined that the non-ground point clouds corresponding to the projection points in the selected projection point grid cannot be clustered with the non-ground point clouds corresponding to the projection points in the first projection point grid and / or the non-ground point clouds corresponding to the projection points in the second projection point grid in other projection point grids. If both the first height difference and the second height difference are less than or equal to the height threshold, it is determined that the non-ground point clouds corresponding to the projection points in the selected projection point grid, the non-ground point clouds corresponding to the projection points in the first projection point grid in other projection point grids, and the non-ground point clouds corresponding to the projection points in the second projection point grid can be clustered into one category. If the first height difference is greater than the height threshold and the second height difference is less than or equal to the height threshold, it is determined that the non-ground point clouds corresponding to the projection points in the selected projection point grid can be clustered with the non-ground point clouds corresponding to the projection points in the second projection point grid in other projection point grids. If the first height difference is less than or equal to the height threshold and the second height difference is greater than the height threshold, it is determined that the non-ground point clouds corresponding to the projection points in the selected projection point grid can be clustered with the non-ground point clouds corresponding to the projection points in the first projection point grid in other projection point grids. And assign a clustering classification number to each cluster. For example, set the height threshold to 0.2 meters, cluster the non-ground point clouds corresponding to the projection points in other projection point grids with a height difference less than or equal to 0.2 meters from the selected projection point grid and the non-ground point clouds corresponding to the projection points in the selected projection point grid into one category, and assign a clustering classification number to each cluster. Among them, the process of clustering the non-ground point clouds corresponding to the projection points in other projection point grids with a height difference less than or equal to 0.2 meters from the selected projection point grid and the non-ground point clouds corresponding to the projection points in the selected projection point grid takes into account the three-dimensional space, and can separate two non-connected targets up and down; effectively distinguish targets that are close in distance but not at the same height. For example, for a traffic sign and a car parked under the sign, they need to be divided into two clusters to generate correct targets.

[0045] In a specific embodiment of the present application, the method for clustering lidar points further includes: in the non-ground point clouds of the same cluster, if the number of non-ground points in the non-ground point clouds of the cluster is less than a threshold value, discard the cluster.

[0046] In a specific example of the present application, a threshold value is set for the number of non-ground point clouds with the same clustering classification number. If the number of non-ground point clouds in a certain cluster is less than the threshold value, the clustering classification number is discarded; the threshold value is usually set to 10. This process improves the accuracy of the target cluster and reduces the interference of cluttered targets.

[0047] In a specific embodiment of the present application, the method for clustering lidar point clouds further includes: if the non-ground point cloud contour of a cluster is completely blocked by the non-ground point cloud contour of another cluster, the cluster corresponding to the non-ground point cloud contour in the blocked cluster is discarded.

[0048] In a specific example of the present application, according to the line-of-sight occlusion method, the target contours that are completely blocked by the line of sight are filtered to obtain the final effective grid. As Figure 3 shown, when cluster 2 is blocked by cluster 1, according to the line-of-sight occlusion rule, cluster 2 is filtered and only cluster 1 is retained; as Figure 4 shown, cluster 2 and cluster 3 are split into two clusters of the same target cluster due to being blocked by cluster 1. According to the line-of-sight occlusion method, due to the occlusion of cluster 1, both cluster 2 and cluster 3 are in the blind area of the line of sight, so cluster 2 and cluster 3 can be discarded. The line-of-sight occlusion method is used to filter out the completely blocked target contours to solve the problems of target splitting and trailing of distant vehicles.

[0049] According to the above steps, the label of the final effective grid is obtained. The corresponding point cloud is reverse-indexed according to the final effective grid, and the non-ground point clouds in each cluster are used to generate the target contour corresponding to the cluster. Among them, the accuracy of generating the corresponding target contour from the non-ground point clouds in the same cluster is higher than that of generating the corresponding target contour using the grid of the same cluster. Due to the characteristic of the three-dimensional cylindrical polar coordinate system grid being divided from small to large, if the method of generating the corresponding target contour using the grid of the same cluster is adopted, the corresponding target contour generated by the distant grid will be very large. However, if the method of generating the corresponding target contour using non-ground point clouds is adopted, whether it is a near target or a distant target, the true contour of the target will not be affected by the grid. The grid is only used to assist in calculating the connectivity of the point cloud. When finally generating the target contour, the cluster point cloud is found by reverse-indexing according to the connected grid, which is efficient and accurate.

[0050] Figure 5 is a schematic diagram composed of a specific implementation manner of a device for lidar point clustering of the present application. In Figure 5In the specific embodiment shown, the apparatus for lidar point clustering of the present application includes: Module 501 is a module for projecting non-ground point clouds scanned by a lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution, to obtain projection points corresponding to the non-ground point clouds in the cylindrical polar coordinate system, where the three-dimensional cylindrical polar coordinate system grid is obtained by dividing the cylindrical polar coordinate system; Module 502 is a module for obtaining a projection point grid in the three-dimensional cylindrical polar coordinate system that contains the projection points according to the coordinate information of the projection points in the cylindrical polar coordinate system; and Module 503 is a module for clustering the non-ground point clouds according to the positions of the projection point grids in the three-dimensional cylindrical polar coordinate system grid.

[0051] In a specific embodiment of the present application, each module in the apparatus for lidar point clustering of the present application can be directly in hardware, in a software module executed by a processor, or in a combination of the two.

[0052] The software module can reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. The exemplary storage medium is coupled to the processor such that the processor can read information from and write information to the storage medium.

[0053] The processor can be a central processing unit (English: Central Processing Unit, abbreviated: CPU), or can also be other general-purpose processors, digital signal processors (English: Digital Signal Processor, abbreviated: DSP), application specific integrated circuits (English: Application Specific Integrated Circuit, abbreviated: ASIC), field programmable gate arrays (English: Field Programmable Gate Array, abbreviated: FPGA), or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof. The general-purpose processor can be a microprocessor, but in an alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration. In an alternative, the storage medium can be integrated with the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In an alternative, the processor and the storage medium can reside in the user terminal as discrete components.

[0054] An apparatus for lidar point clustering according to the present application can be used to execute the lidar point clustering method described in any of the above embodiments. The implementation principle and technical effects are similar and will not be elaborated here.

[0055] In another specific embodiment of the present application, a computer-readable storage medium stores computer instructions, and the computer instructions are operated to execute the lidar point clustering method in Solution 1.

[0056] The present application treats all non-ground points scanned by the lidar as having the same status, without distinguishing feature points and domain points. The present application projects the point cloud using the method of a three-dimensional cylindrical polar coordinate system grid. Utilizing the sparse characteristics of non-ground points scanned by the lidar at a long distance, the three-dimensional cylindrical polar coordinate system grid is a radially non-uniform grid; at a short distance, using small grids can solve the problem of dense point clouds at close range, and in the tangent plane, the point clouds of two targets will not be projected together. At a long distance, using large grids can increase the number of point clouds in each grid and it is not easy to miss distant targets. Moreover, the time complexity of the algorithm is low and the efficiency is very high. It can effectively manage target trajectories and reduce false targets during the process of lidar target tracking. In addition, the present application adopts the line-of-sight occlusion method, which can filter out completely occluded targets and solve the problems of target splitting and trailing of distant vehicles.

[0057] In the embodiments provided by the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of the apparatus or unit can be in electrical, mechanical or other forms.

[0058] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0059] The above are only the embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural transformation made using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be included in the patent protection scope of the present application by the same token.

Claims

1. A method for lidar point clustering, characterized in that, Including: Projecting the non-ground point cloud scanned by the lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution to obtain the projection points corresponding to the non-ground point cloud in the cylindrical polar coordinate system. The three-dimensional cylindrical polar coordinate system grid is obtained by dividing the cylindrical polar coordinate system. The process of pre-establishing the pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution includes: taking the position of the lidar as the origin, establishing the cylindrical polar coordinate system with a radius of R on the horizontal plane, where the radius R is greater than or equal to the scanning range of the lidar; according to the angular dimension of the cylindrical polar coordinate system, dividing the cylindrical polar coordinate system into M three-dimensional solids with a sector-shaped bottom and a height, where M is an integer not less than 2; and in each solid, according to the distance dimension of the cylindrical polar coordinate system, dividing the area from the origin of the cylindrical polar coordinate system to a distance of R meters from the origin of the cylindrical polar coordinate system to obtain the three-dimensional cylindrical polar coordinate system grid. The farther away from the origin of the cylindrical polar coordinate system, the larger the width of the three-dimensional cylindrical polar coordinate system grid in the distance dimension. According to the order of the distance from the origin from near to far, the width of the grid closer to the origin is not greater than the width of the grid farther from the origin in any two grids; According to the coordinate information of the projection points in the cylindrical polar coordinate system, obtaining the projection point grid in the three-dimensional cylindrical polar coordinate system that contains the projection points; and According to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid, clustering the non-ground point cloud. Among them, the grid containing the point cloud of the clustering result is obtained as the final valid grid, and the corresponding point cloud is reverse-indexed by the final valid grid. Using the non-ground point cloud in each cluster, the target contour corresponding to the cluster is generated.

2. The method for lidar point clustering according to claim 1, wherein, The process of obtaining the projection point grid in the three-dimensional cylindrical polar coordinate system that contains the projection points according to the coordinate information of the projection points in the cylindrical polar coordinate system includes: Respectively retrieving the projection points in the three-dimensional cylindrical polar coordinate system grid according to the angular dimension and the distance dimension, and determining the three-dimensional cylindrical polar coordinate system grid with the projection points as the projection point grid.

3. The method for lidar point clustering according to claim 1 or 2, characterized in that, The process of clustering the non-ground point cloud according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid includes: Respectively retrieving the projection point grid in the three-dimensional cylindrical polar coordinate system grid according to the angular dimension and the distance dimension to obtain multiple groups of adjacent projection point grids; Respectively calculating the height difference between other projection point grids and any one projection point grid according to the height of any one projection point grid in each group of adjacent projection point grids; In each group of adjacent projection point grids, clustering the non-ground point cloud corresponding to the projection points in the other projection point grids with the height difference less than or equal to the height threshold and the non-ground point cloud corresponding to the projection points in any one projection point grid into one category.

4. The method for laser radar point clustering according to claim 3, wherein The process of calculating the height difference between other projection point grids and any one of the projection point grids according to the height of any one of the projection point grids in each group of adjacent projection point grids includes: Determining the height of the projection point grid according to the height of the non-ground point cloud points corresponding to the projection points in the projection point grid, where the highest height of the non-ground point cloud points corresponding to the projection points in the projection point grid is the highest height of the projection point grid, and the lowest height of the non-ground point cloud points corresponding to the projection points in the projection point grid is the lowest height of the projection point grid.

5. The method for lidar point clustering according to claim 4, wherein The process of calculating the height difference between other projection point grids and any one of the projection point grids according to the height of any one of the projection point grids in each group of adjacent projection point grids includes: In each group of adjacent projection point grids, comparing the highest height of any one of the projection point grids with the highest height in the other projection point grids to obtain the higher grid and the lower grid of the other projection point grids relative to any one of the projection point grids; Calculating the first height difference between the lowest height of the higher grid and the highest height of any one of the projection point grids and the second height difference between the highest height of the lower grid and the lowest height of any one of the projection point grids respectively, where the height difference includes the first height difference and the second height difference.

6. The method for lidar point clustering according to claim 1, wherein It also includes: In the non-ground point cloud of the same cluster, if the number of non-ground points in the non-ground point cloud of the cluster is less than the threshold value, then discard the cluster.

7. The method for lidar point clustering according to claim 1, wherein, It also includes: If the non-ground point cloud contour in a cluster is completely blocked by the non-ground point cloud contour in another cluster, then discard the cluster corresponding to the non-ground point cloud contour in the blocked cluster.

8. An apparatus for lidar point clustering, characterized in that, It includes: A module for projecting the non-ground point cloud scanned by the lidar onto a pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution to obtain the projection points corresponding to the non-ground point cloud in the cylindrical polar coordinate system. The pre-establishment process of the pre-established three-dimensional cylindrical polar coordinate system grid with non-uniform radial distribution includes: taking the position of the lidar as the origin, establishing the cylindrical polar coordinate system with a radius of R on the horizontal plane, where the radius R is greater than or equal to the scanning range of the lidar; dividing the cylindrical polar coordinate system into M three-dimensional solids with a sector-shaped bottom surface and a height according to the angular dimension of the cylindrical polar coordinate system, where M is an integer not less than 2; and in each solid, dividing the area from the origin of the cylindrical polar coordinate system to a distance of R meters from the origin of the cylindrical polar coordinate system according to the distance dimension of the cylindrical polar coordinate system to obtain the three-dimensional cylindrical polar coordinate system grid, where the farther away from the origin of the cylindrical polar coordinate system, the larger the width of the three-dimensional cylindrical polar coordinate system grid in the distance dimension, and in any two grids in the order from near to far from the origin, the width of the grid closer to the origin is not greater than the width of the grid farther from the origin; A module for obtaining a projection point grid in the three-dimensional cylindrical polar coordinate system grid that contains the projection point according to the coordinate information of the projection point in the cylindrical polar coordinate system; and A module for clustering the non-ground point cloud according to the position of the projection point grid in the three-dimensional cylindrical polar coordinate system grid, wherein a grid containing the clustered result point cloud is obtained as the final valid grid, the corresponding point cloud is reverse-indexed using the final valid grid, and the non-ground point cloud in each cluster is used to generate a target contour corresponding to the cluster.

9. A computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are operated to execute the method for lidar point clustering according to any one of claims 1-7.

Images