Method for realizing laser point cloud ground segmentation based on angle differentiation

Abstract

The invention discloses a method for realizing laser point cloud ground segmentation based on angle differentiation, which comprises the following steps of: firstly, carrying out downsampling processing on a point cloud by utilizing a voxel filter; and then cutting out the point cloud of the ground area according to the height. Ground point clouds are regarded as wire harnesses emitted with the laser radar as the center, whether laser points are ground points or not is judged according to the range of a local height threshold value and a global height threshold value through angle differentiation, and therefore effective segmentation of ground point clouds and non-ground point clouds is achieved. The method has the characteristics of low complexity, small calculation amount and high real-time performance, and has obvious advantages in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud data processing in laser radar perception, and specifically relates to a method for implementing laser point cloud segmentation of the ground based on angle differential. Background Art

[0002] In the field of unmanned driving, laser radar perception, segmenting the laser radar ground point cloud is a key step in data processing, mainly to improve the impact of ground point cloud on the clustering of target laser point cloud above the ground, so as to improve the accuracy of target point cloud recognition and segmentation. And in the field of robot autonomous navigation, when converting a three-dimensional point cloud map into a two-dimensional grid map, the impact on the quality of the converted two-dimensional grid map. The existing algorithms for segmenting ground point clouds, such as those based on ground model fitting or deep learning, have the disadvantages of high algorithm complexity and excessive consumption of computer resources, and usually require a lot of computing power and memory. If it is difficult to meet, the real-time performance of the point cloud algorithm will be reduced, which will ultimately affect the accuracy of navigation and decision-making.

[0003] The segmentation method based on angle differential uses the spatial angle feature information of the laser point cloud to segment the ground and non-ground. This method avoids the complex feature extraction and ground model fitting process, thereby simplifying the data processing flow and algorithm complexity. Therefore, this method has low requirements for computing resources and can effectively improve the real-time performance of ground point cloud segmentation, making the method of implementing laser point cloud segmentation based on angle differential have obvious advantages in the existing technology. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention proposes a method for realizing ground segmentation by laser point cloud based on angle differential. The algorithm design is simple and reasonable, and the ground point cloud and non-ground point cloud can be segmented quickly and in real time.

[0005] To achieve the above object, the present invention adopts the following technical solution: a method for implementing laser point cloud segmentation of the ground based on angle differential, comprising the following steps:

[0006] Step 1: Use voxel filter to downsample the input raw laser point cloud;

[0007] Step 2: Crop the downsampled point cloud according to the preset Z-axis coordinate threshold to filter out the valid area point cloud;

[0008] Step 3: To avoid the influence of reflection from objects around the radar, perform close-range laser point cloud cropping on the cropped point cloud;

[0009] Step 4: Project the point cloud after close-range cropping in the horizontal direction, and calculate the horizontal distance between each point in the point cloud and the laser radar, that is, the radius;

[0010] Step 5: Perform angle differentiation on the point cloud, classify the point cloud according to the angle between the laser point line and the positive direction of the laser radar, and sort the points on each line beam according to the radius;

[0011] Step 6: Traverse the point cloud on each line beam, initialize the point with the smallest radius on each line beam, that is, the first point, set the current point to a non-ground point, set the radius of the previous point to 0, and the height to the negative value of the laser radar installation height;

[0012] Step 7: Calculate the distance between the current point and the previous point, and calculate the local height threshold and the global height threshold between the two points according to the preset local maximum slope and the global maximum slope, and set the minimum local height threshold as the minimum local height threshold;

[0013] Step 8: Determine whether the height of the current point is within the range of the negative value of the laser radar installation height plus or minus the minimum local height threshold. If it is within the range, continue to determine whether the previous point is a ground point. If it is a ground point, the current point is a ground point. If the previous point is a non-ground point, continue to determine whether the height of the current point is within the range of the negative value of the laser radar installation height plus or minus the global height threshold. If it is within the range, the current point is a ground point, otherwise it is a non-ground point.

[0014] Step 9: If the height of the current point is not within the range of the negative value of the laser radar installation height plus or minus the minimum local height threshold, continue to judge. If the distance between the two points is greater than the preset reclassification distance threshold and the height of the current point is within the range of the negative value of the laser radar installation height plus or minus the local height threshold, then the point is a ground point, otherwise it is a non-ground point;

[0015] Step 10: Update the previous point status, set the previous point radius to the current point radius, the height to the current point height, and whether it is a ground point status, and then loop to determine the next point status.

[0016] The present invention has the following beneficial effects:

[0017] First, downsampling is performed using a voxel filter, and then the effective area point cloud is cut out according to the Z-axis size height, and then the close-range laser point cloud is cut out, and finally, by angle differentiation, the local height threshold and the global height threshold range are used to determine whether the laser point is a ground point. The present invention has the characteristics of simplicity, high efficiency, strong adaptability, etc., and can handle various terrains and adapt to different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of a method for implementing laser point cloud segmentation of the ground based on angle differential according to the present invention;

[0019] Figure 2It is a schematic diagram of the cross-section of the laser radar point cloud in the present invention. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0021] like Figure 1 As shown, a method for implementing laser point cloud ground segmentation based on angle differential includes the following steps:

[0022] Step 1: Input the original laser point cloud X = {P1,P2,...P n}, use voxel filter to implement downsampling processing, after downsampling, the point cloud is Q = {P1,P2,...P n}. Where X represents the input original laser point cloud, n is the number of laser point clouds, and each laser point is represented by P i ={x i ,y i ,z i}, x, y, z represent the coordinates of the laser point respectively, and i is the index of the laser point;

[0023] Step 2: After traversing the downsampling laser point cloud Q = {P1, P2, ... P n}, filter the laser point cloud whose Z-axis coordinate is not within the preset Z threshold, that is, cut out the effective area point cloud according to the height as P = {P1, P2, ... P n};

[0024] Step 3: After traversing the height clipping point cloud P = {P1, P2, ... P n}, perform close-range laser point cloud cropping to avoid the influence of reflections from objects around the radar, shield and delete close-range point clouds, and after cropping, the point cloud is W = {P1, P2, ... P n};

[0025] Step 4: After close cropping, point cloud W = {P1, P2, ... P n} Project in the horizontal direction, and the point cloud after projection is T i =(x i ,y i ), and for the projected point T i =(x i ,y i ) Calculate the distance between each point and the laser radar 1, that is, the radius

[0026] Step 5: Calculate the angle between the laser point line and the positive direction of LiDAR 1 in the same direction:

[0027] By classifying all point clouds according to theta size, 360° / θ line beams can be obtained, where θ is the angular resolution of the LiDAR, and the points on each line beam are sorted according to the radius distance.

[0028] Step 6: Traverse the point cloud on each line bundle, initialize the point with the smallest radius on each line bundle, that is, the first point, set the current point to a non-ground point, and initially set the radius of the previous point to 3R prev is 0, and the height is the negative value of the laser radar installation height 2, that is, -H sensor , the state is a non-ground point;

[0029] Step 7: Figure 2 As shown, the difference R between the radius 4 of the current point 6 and the radius 3 of the previous point 5 is calculated. current -R prev , and according to the preset local maximum slope and K local Global maximum slope K global The local height threshold 7 and the global height threshold 8 between the two points are calculated respectively, that is, the calculation formula of the local height threshold 7 is:

[0030] H local_threshold =(R current -R prev )×K local

[0031] The calculation formula of global height threshold 8 is:

[0032] H global_threshold =(R current -R prev )×K global

[0033] And set the minimum local height threshold 7 as the minimum local height threshold H min_local_threshold ;

[0034] Step 8: If the current point 6 height is within the range of the negative value of the laser radar installation height 2 plus or minus the minimum local height threshold, it satisfies:

[0035] H current ∈(-H sensor -H min_local_threshold ,-H sensor +H min_local_threshold )

[0036] Continue to determine whether the previous point 5 is on the ground 9. If it is a ground point, the current point 6 is a ground point. If the previous point 5 is not a ground point, continue to determine whether the height of the current point 6 is within the range of the negative value of the laser radar installation height 2 plus or minus the global height threshold 8, that is, it satisfies

[0037] H current∈(-H sensor -H global_threshold ,-H sensor +H global_threshold )

[0038] If it satisfies the requirement, the current point 6 is a ground point, otherwise it is a non-ground point;

[0039] Step 9: If the height of the current point 6 is not within the range of the negative value of the laser radar installation height 2 plus or minus the minimum local height threshold, continue to judge. If the distance between the two points is greater than the preset reclassification distance threshold D, and the height of the current point is within the range of the negative value of the laser radar installation height 2 plus or minus the local height threshold 7, then the point is a ground point, otherwise it is a non-ground point;

[0040] Step 10: Update the previous point status, set the previous point radius to the current point radius, the height to the current point height, and whether it is a ground point status, and then loop to determine the next point status.

Claims

1. A method for implementing laser point cloud ground segmentation based on angle differential, characterized in that: The steps include: Step 1: Input the original laser point cloud X = {P1,P2,...P n }, use voxel filter to implement downsampling processing, after downsampling, the point cloud is Q = {P1,P2,...P n }, where X represents the input original laser point cloud, n is the number of laser point clouds, and each laser point is represented by P i ={x i ,y i ,z i }, x, y, z represent the coordinates of the laser point respectively, and i is the index of the laser point; Step 2: After traversing the downsampling laser point cloud Q = {P1, P2, ... P n }, filter the laser point cloud whose Z-axis coordinate is not within the preset Z threshold, that is, cut out the effective area point cloud according to the height as P = {P1, P2, ... P n }; Step 3: After traversing the height clipping point cloud P = {P1, P2, ... P n }, perform close-range laser point cloud cropping to avoid the influence of reflections from objects around the radar, shield and delete close-range point clouds, and after cropping, the point cloud is W = {P1, P2, ... P n }; Step 4: After close cropping, point cloud W = {P1, P2, ... P n } Project in the horizontal direction, and the point cloud after projection is T i =(x i ,y i ), and for the projected point T i =(x i ,y i ) calculate the distance between each point and the laser radar, that is, the radius Step 5: Calculate the angle between the laser point line and the positive direction of the laser radar in the same direction: By classifying all point clouds according to theta size, 360° / θ line beams can be obtained, where θ is the angular resolution of the LiDAR, and the points on each line beam are sorted according to the radius distance. Step 6: Traverse the point cloud on each line bundle, initialize the point with the smallest radius on each line bundle, that is, the first point, set the current point to a non-ground point, and initially set the radius R of the previous point prev is 0, and the height is the negative value of the laser radar installation height, that is, -H sensor , the state is a non-ground point; Step 7: Calculate the difference R between the radius of the current point and the radius of the previous point current -R prev , and according to the preset local maximum slope and K local Global maximum slope K global Calculate the local height threshold and global height threshold between two points respectively, that is, the calculation formula of the local height threshold is: H local_threshold =(R current -R prev )×K local The global height threshold calculation formula is: H global_threshold =(R current -R prev )×K global And set the minimum local height threshold as the minimum local height threshold H min_local_threshold ; Step 8: If the current point height is within the range of the negative value of the laser radar installation height plus or minus the minimum local height threshold, it satisfies: H current ∈(-H sensor -H min_local_threshold ,-H sensor +H min_local_threshold ) Continue to determine whether the previous point is on the ground. If it is a ground point, the current point is a ground point. If the previous point is not a ground point, continue to determine whether the height of the current point is within the range of the negative value of the lidar installation height plus or minus the global height threshold. H current ∈(-H sensor -H global_threshold ,-H sensor +H global_threshold ) If it satisfies the requirement, the current point is a ground point, otherwise it is a non-ground point; Step 9: If the height of the current point is not within the range of the negative value of the laser radar installation height plus or minus the minimum local height threshold, continue to judge. If the distance between the two points is greater than the preset reclassification distance threshold D, and the height of the current point is within the range of the negative value of the laser radar installation height plus or minus the local height threshold, then the point is a ground point, otherwise it is a non-ground point; Step 10: Update the previous point status, set the previous point radius to the current point radius, the height to the current point height, and whether it is a ground point status, and then loop to determine the next point status.

Images