Detection And Tracking Of Obstacles For Intelligent Vehicle Based On 3D LIDAR

Intelligent Vehicle is a complex combination,including environmental perception,planning,decision making and behavior control modules.It integrates computer technology,sensor technology,communication technology,pattern recognition technology and automatic control technology and other high technologies.There are a lot of obstacles hindering vehicles,and accurate perception surrounding vehicles is a very difficult but urgent problem.Due to its long distance measurement,high measurement accuracy and low environmental impact,LIDAR is widely used in the intelligent vehicle sensors and has important significance of theoretical research and engineering application.Based on demand for environment perception,it takes intelligent vehicle developed in Jiangsu University as the research platform.We use 3D LIDAR to detect and track obstacles around vehicles to provide reliable target information for safe driving of intelligent vehicle.The main research content is as follows:(1)The environmental perception platform for intelligent vehicle based on 3D LIDAR is built.Point cloud data of 3D LIDAR and the transformation relationship between the LIDAR coordinate system and the vehicle coordinate system are analyzed.The algorithm and the application software are developed.(2)In order to solve the problem that hanging objects are difficult to detect in point cloud data segmentation and the algorithm robustness is insufficient for the single threshold value,a novel multi feature and multi layer grid map is presented.The point cloud in the grid is divided vertically to detect hanging objects,and the ground area is determined by analyzing of the height and intensity features.Through experimental analysis,this algorithm can effectively separate the ground area,obstacles and hanging obstacles.(3)In order to solve the over-segmentation phenomenon in the process of clustering,the depth value of obstacle grid and the distance of clustering are associated.In addition,the clustering matching is carried out with the characteristics of the adjacent frame obstacles to complete the clustering in different grid blocks of the same target.In the process of object classification,we uses the relative observation angle to characterize the scanning range of the target object contour and uses support vector machine to classify obstacle objects by the position and profile features of targets.(4)In view of the variable number of obstacles and the changing characteristics of the motion state at any time,a combination of multiple hypothesis tracking algorithm and the interactive multiple model algorithm is used.Multiple hypothesis tracking algorithm is applied to multiple target data association,while interactive multiple model algorithm is applied to state estimation of the target.Through the comparison experiment of Kalman filter,it is proved that the algorithm can get position and speed of obstacles more accurately.A large number of experiments have been carried out to verify the real vehicle data and KITTI data collected by the intelligent vehicle testing platform.The results show that the proposed 3D LIDAR data processing algorithm can detect and track obstacles around the vehicle,and it has good feasibility and robustness.