Research On Road Scene Point Cloud Classification Based On Vehicle-mounted LiDAR

With the rise of concepts such as autonomous driving,smart cities,and high-definition maps and the development of environmental perception-related technologies,the vehiclemounted scanning system composed of IMU,camera,millimeter-wave radar and Li DAR with vehicles as the carrying platform is an important way to obtain 3D information in road scenes.It has important theoretical value and practical significance.This thesis conducts classification research on the road scene point cloud of vehicle-mounted Li DAR,and the main work includes the following aspects:(1)The working principle of vehicle-mounted Li DAR and the characteristics of Li DAR point cloud data are analyzed,and the current situation of ground segmentation,aboveground object segmentation and classification of point clouds is analyzed and studied,which provides strong support for subsequent point cloud data processing.(2)Research on the ground feature segmentation algorithm based on depth map,according to the working principle of Li DAR collection point cloud,the 3D point cloud is projected as a depth map,and by analyzing the characteristics of the ground point cloud,the ground figure is segmented on the depth map according to the angle threshold,and then restored to the 3D point cloud to realize the ground point cloud segmentation,which is convenient for the subsequent point cloud segmentation of above-ground objects.(3)In this thesis,a point cloud segmentation optimization algorithm based on depth map is proposed to realize the rapid segmentation of objects in road scenes.On the basis of ground point cloud segmentation,in the process of collecting point clouds from adjacent laser beams of Li DAR,the angle threshold is designed to realize the segmentation of aboveground objects,and the interpolation optimization of the segmentation results is carried out by using the principle of K-nearest neighbor algorithm,which further improves the segmentation accuracy,reduces the occurrence of undersegmentation and oversegmentation,and meets the requirements of real-time segmentation.(4)Based on the identification and classification research of point cloud targets in vehicle Li DAR road scenes,this thesis classifies semantic targets according to the objects in the selected dataset,determines the categories of objects in the road scene according to different dimensions,uses the CSF algorithm to segment the ground objects to extract the ground point cloud,uses the DBSCAN algorithm to segment the point cloud of the aboveground objects,classifies the divided object point clouds according to the characteristics of the above-ground objects,and merges some misclassified point cloud clusters.Improved the accuracy and recall of point cloud classification in road scenes.