Vehicle Detection And Tracking Based On LiDAR In Highway Scene

Environmental perception is an important prerequisite for achieving autonomous driving technology,and vehicle detection and tracking is its core task.As a typical working condition for vehicles,highway has characteristics such as high vehicle speed and a more variable environment.There are many technical difficulties in vehicle detection and tracking in this scenario.In response to these problems,this article proposes a solution based on laser perception technology.The main content includes:(1)On the basis of reviewing the research results on ground segmentation,clustering,target detection and multi-target tracking at home and abroad,combined with the characteristics of highway and the problems in current technologies,this article proposes the research content and technical route of this article.(2)For the problem of point cloud ground segmentation and obstacle clustering,this article first uses the neighborhood elevation information and depth image of the point cloud to extract the ground point cloud from the original point cloud with a combination of coarse and fine granularity,and obtains the obstacle point cloud after segmenting the ground.Then,in the spherical coordinate system,the obstacle point cloud is divided into curved voxels,and each point’s neighbor points in the voxel are marked as the same point cloud cluster.Finally,all point cloud clusters are merged for the second time to obtain the clustering result.Through experimental verification,the point cloud ground segmentation and obstacle clustering method proposed in this article has good accuracy and real-time performance.(3)For the problem of point cloud target detection,this article uses the method of fitting L-shaped bounding boxes.First,all two-dimensional bounding boxes within a specified angle range are searched for each point cloud cluster.Then the optimal fitting result is selected from them.Finally,combined with the elevation information of the point cloud cluster,the three-dimensional bounding box of the vehicle target is obtained.Through experimental verification,this method achieves the expected accuracy while meeting the real-time requirements.(4)For the problem of multi-target tracking,this article proposes a solution based on Multi-Anchor.Using the Hungarian algorithm and the Kalman filter algorithm,the input detection targets are matched,and the motion state is predicted and updated.In response to the problems of mismatching,slow convergence of the Kalman filter,and high dependence on the accuracy of target detection results in the current algorithm,this article proposes methods for calculating multi-dimensional correlation features,calculating target initial states,and filtering target detection results.While meeting the real-time requirements,it improves the accuracy and robustness of the multi-target matching system.In summary,this article proposes a vehicle detection and tracking scheme based on laser perception technology in the scenario of highways.Through experimental verification,it is proved that this scheme can improve the accuracy and robustness of vehicle detection and tracking while meeting the real-time requirements.