Research On Road Segmentation And Obstacle Detection Of Autonomous Vehicle Based On Lidar

The determination of the area on the road that the vehicle can drive and the detection of obstacles in the relevant range that affect driving are key technologies to ensure the safety and robustness of Intelligent vehicle in the process of automatic driving.It can not only provide accurate and reasonable path planning constraints for intelligent vehicle lane change and overtaking,but also provide safe distance information for intelligent vehicle active braking.This paper takes the laboratory intelligent vehicle as the experimental platform and VLP-16 line laser radar as the vehicle sensor to research the road boundary extraction and obstacle detection.Firstly,the quantitative data of lidar are parsed in this paper,which in the spherical coordinate system of both angle and distance of each point cloud are got.Then,based on the known result,the other result which is in the rectangular coordinate system can be calculated.Finally,The Effective Information of lidar data is reversed by Sampling to make date useful for Vehicle.Secondly,a β sub-height feature difference algorithm is proposed for road segmentation.First,a point cloud classifier is designed to classify each frame of lidar point cloud into obstacle point cloud and non-obstacle point cloud.Taking the non-obstacle point cloud as the candidate point of road point cloud,the height of adjacent point cloud in each ray is β sub-differenced to enlarge the difference between road and non-road point cloud,and the interested position of road boundary is obtained.In this position of interest,the sliding window is established,and then the direction feature and angle feature of the point cloud in the window are solved successively,and the accurate road boundary point cloud is calculated.With this constraint,the road point cloud is segmented by the direct-pass filter to determine the driving area.Then,for the obstacle point cloud obtained by classification,an adaptive clustering algorithm is proposed whose clustering radius is based on the palindromic spatial distance of each point cloud.The first step of the algorithm is downsampling,the voxel grid method is used to downsample each frame of point cloud to reduce the number of point cloud in clustering.Then,according to the distance information returned by point cloud and the distance information of adjacent point cloud after downsampling,different clustering radius thresholds in clustering process are calculated.Considering the problem that the detection result does not match the actual obstacle size due to the over-segmentation of spatial clustering,a method for confirming the target size based on the maximum overlap length of the center pair is proposed.For each obstacle clustering result,the center coordinate distance between the obstacle clustering result and the adjacent ray obstacle clustering result and the overlap length of the two clustering lengths in the coordinate system are calculated.The clustering results are reintegrated to obtain accurate obstacle information.Finally,with the help of the intelligent vehicle built in the laboratory,the vehicle experiment under actual working conditions is carried out.First,the effectiveness of point cloud classification algorithm is verified,Then,the road segmentation experiments are carried out on structured and unstructured roads,and the experimental results are quantitatively analyzed.Then,Obstacle detection experiment based on my algorithm.And it is compared with the planar European clustering algorithm to prove my algorithm is more better.Finally,based on the road boundary constraint and obstacle position constraint solved by lidar perception,two experiments of automatic brake and lane changing to avoide obstacle of intelligent vehicle are carried out respectively,which verifies that the proposed algorithm can provide effective and accurate safety data for the safe driving of vehicles.