Research And Implementation Of Obstacles Detection And Tracking Algorithm Based On Lidar

With the advent of the era of artificial intelligence,the development of intelligent driving is becoming more and more important.Lidar,a novel sensor,will be a crucial “eye” in an intelligent vehicle.Aiming at the application of lidar in the direction of intelligent perception,this thesis analyzes the working principle of lidar,studies and implements a series of solutions from lidar calibration and obstacle detection to obstacle tracking.The main contents are as follows:1.Firstly,the obstacle detection and tracking algorithm platform based on lidar is selected and built to filter the original lidar point cloud data.Complete the calibration of lidar and vehicle body coordinate system and realize the coordinate transformation of the position of the obstacle in the lidar coordinate system and vehicle body coordinate system.The motion compensation algorithm is designed to eliminate the point cloud distortion of the lidar produced during the movement of the intelligent vehicle.2.Secondly,the obstacle detection algorithm based on lidar is designed.The method of ground point removal based on multi-sector plane fitting is proposed.Based on the flood water filling algorithm,overwater seed growth clustering algorithm is designed based on the laser radar.In this algorithm,the raster projection on the twodimensional driving plane is combined with the regional growth method to cluster the 3D point clouds rapidly.Then,according to the minimum convex hull,the main direction of the obstacle is determined,and the clustering rectangular frame is extracted.Finally,the obstacle is simply classified according to the point cloud information.3.Then the obstacle tracking algorithm based on lidar is designed,including obstacle correlation matching and dynamic tracking.The correlation between the newly detected obstacle and the tracked obstacle is calculated by using similarity of characteristics(including position and number of point clouds),and then the Hungarian algorithm is used to correlate the newly detected obstacle with the tracked obstacle one by one.Kalman filter is used to update the state of the matched obstacle,and then the obstacle is tracked iteratively and dynamically.4.Finally,in order to verify the effectiveness and real-time performance of the algorithm designed in this thesis,a series of real vehicle experiments,including obstacle clustering experiment,obstacle classification experiment,obstacle association experiment and obstacle status updating experiment,were conducted with the intelligent vehicle of Hunan university as the experimental platform.Experimental results show that the obstacle detection and tracking algorithm designed and developed in this thesis can accurately detect the surrounding obstacles,dynamically track the obstacles,and predict and judge the movement direction of the obstacles,providing important information for the environmental perception of intelligent vehicles.