Real-time Vehicle Detection And Tracking Using 3D LiDAR

With the deepening of the revolution of science and technology,human society has entered the new era of intelligent Internet.With the rapid development of the automotive industry and artificial intelligence,autonomous vehicles have gradually become the focus of the industry and guide the future development direction of automobiles.The autonomous vehicle is the product of the deep integration of automotive industry with the Internet of things,artificial intelligence and high-performance computing chips,it is a comprehensive system which integrates various technologies such as environmental perception,navigation and positioning,road planning,behavior control and so on.In the real road environment,vehicles are important interactive objects for the autonomous vehicle,so it is very important to study the vehicle detection and tracking,which can avoid the collision of the autonomous vehicle and protect the safety of the occupant.3D LiDAR is an important sensor for environmental perception module of the autonomous vehicle,the environmental information obtained from 3D LiDAR-based vehicle detection and tracking system also helps in further path planning and SLAM.In order to accurately detect and track other vehicles around the autonomous vehicle,a point cloud clustering algorithm is designed in this thesis,which can effectively extract a series of independent objects from the original point cloud data.Some features are selected and a classifier trained by SVM is used to detect vehicles.Unscented Kalman Filter and GNN are used to track vehicles,then the vehicle detection performance is further improved by the aid of tracking results.The vehicle detection and tracking methods designed in this thesis have been verified on the autonomous vehicle platform.