Obstacle Target Detection And Tracking Based On 3D LiDAR

Lidar is widely used in unmanned driving environment perception system due to the advantages of abundant environmental information and being unaffected by illumination factors.It has important theoretical research and practical application value.This article focuses on the environmental perception needs of unmanned driving systems for surrounding obstacle information,using self-developed smart cars as a research and testing platform to carry out research on obstacle detection and tracking methods based on 3D laser radar.For the safety of unmanned vehicles driving,these methods provides the status information of obstacle targets.The main research contents are as follows:(1)Build a smart car test platform based on 3D lidar,analyze the characteristics of point cloud data and establish the lidar coordinate system,vehicle coordinate system and coordinate conversion relationship between the two,and develop application software to provide subsequent point cloud data processing basis.(2)To solve the problem of insufficient accuracy of the segmentation algorithm based on a single height threshold in the segmentation of ground point cloud,a multiiteration plane fitting method is proposed.Plane fitting is carried out by least squares method to restore the environmental pavement,and the ground point cloud is determined by combining the height characteristics;on this basis,the ground point cloud data is removed to obtain the obstacle point cloud data.(3)According to the working scanning principle of lidar,a clustering algorithm based on scan lines is proposed;the range resolution of lidar is analyzed,and the dynamic threshold of distance threshold in the clustering algorithm is established,which can be effectively reduced The calculation cost of the clustering algorithm and improve the clustering effect of the algorithm.Use the moment body model to model the clustering results and extract the location information of obstacles.(4)In the process of obstacle data association,combining the characteristics of the similar obstacle point cloud shape between adjacent frames,the obstacle feature parameters are introduced on the basis of the nearest neighbor data association algorithm to improve the robustness of the association algorithm;Based on the information of obstacle targets in and out of the region of interest,a dynamic obstacle extraction algorithm based on distance difference characteristics is proposed,and a Kalman filter is used to track the dynamic obstacle target.This article collects field data in a smart car test platform and conducts a large number of experimental verifications.The results show that the proposed point cloud data processing algorithm can achieve accurate detection and stable tracking of obstacle targets in the area of interest.