3D Point Cloud Object Detection Algorithm Research Based On LiDAR

With the rapid Development of science and technology,the automotive industry has made great progress.With the revolution of computer technology and the rise of artificial intelligence technology,unmanned vehicle technology has gradually developed.Another reason for the rise of unmanned vehicle technology is the increase of car ownership,and too many vehicles will lead to increased risks in some complex scenes.At present,unmanned driving technology has a system,which mainly includes environment perception,path planning and intelligent decision making.Among them,environment perception is the basic module of all technologies and plays a huge role in the whole system.The existing endto-end 3D object detection network cannot solve the problem of false detection and missed detection in short distance,which reduces the safety of unmanned driving system.In this paper,object detection is realized by using the spatial geometric information of 3D point cloud,and the safety of environment perception is improved by increasing the redundancy.The main work and contribution of this paper are as follows:1.Aiming at the ground point cloud would be under-segmented,this paper proposes to map the 3D point cloud into the grid map,and then use RANSAC and least squares to detect and filter the ground point cloud.2.With the noise in non-ground point cloud,which affects the object clustering,the density-based clustering method is used,so that the noise point cloud can be filtered directly and the opbject clustering results of nonground point cloud can be obtained at the same time.3.Aiming at the object detection in the region of interest,after a series of preprocessing of the clustered point cloud,the classifier based on the point cloud is obtained by training Point Net++ classification network.