Researchon Road Information Extraction And Target Tracking Algorithm Based On Lidar

As an important aspect of artificial intelligence,driverless has gradually become a hot research topic at home and abroad.Intelligent vehicle has not only eased the pressure on transportation,but also reduces the resource consumption of the scenic spot.As a complete knowledge system,intelligent vehicle mainly includes four aspects: perception,decision,positioning and control.The environmental perception part is especially important.Multi-layer lidar plays an important role in the field of environment perception because of its large amount of data,high measurement accuracy and strong robustness.The research direction of 3D lidar in the environment perceptionis mainly divided into two parts,one is the extraction of roadside information,which is mainly applied to structured roads,and the other is the detection and tracking of multi-targets.It is mainly used to prevent vehicle collisions on urban roads.The main research contents of the paper based on 3D lidar are as follows:(1)In order to extract the roadside information of structured roads,a road information extraction algorithm is proposed.Firstly,the path edge point setsare extracted by the interval collinear point algorithm according to the characteristics of the straight line edge.Secondly,the left and right path point sets are classified according to the improved density peak based clustering algorithm.Finally,the roadside path is adopted by least squares method.The classification algorithm based on ? is used to improve the clustering algorithm based on density peak,which solves the problem of manually selecting the cluster center point and improves the accuracy of cluster.Road information extraction facilitates target detection in the travelable area,which canachieve the purpose of safe driving.(2)The main steps of the target detection algorithm are as follows: Firstly,due to the excessive number of point clouds in the lidar,in order to improve the real-time performance of the algorithm,the Voxel Grid filter is used for downsampling to reduce the density of the point cloud while maintaining the basic characteristics of the point cloud.Secondly,in order to improve the accuracy of the algorithm,the ground pointsare filtered by the ground segmentation algorithm,and the noise pointsare removed by using a filtering algorithm.Finally,the obtained point cloudsare segmented every target by the European clustering algorithm.And we uses the RANSAC algorithm to construct the Bounding Box to mark every target.We can obtain the feature attributes from the Bounding Boxsuch as the length,width and center position coordinates of the target.(3)After getting the target's feature attributes,you need to track the target.Target tracking is mainly divided into two parts: target match and target motion state estimation.The existing Hungarian association algorithm can not accurately correlate uncertain targets.So,I proposd a improvedHungarian algorithm.The algorithm is mainly improved from the following two aspects:1)adding the appropriate tracker management strategy to the original Hungarian algorithm,updating the correlation matrix in real time,tracking the uncertain number of targets;2)in terms ofdefining the costcalculation equations of the associated value,other characteristics of the target and tracker are considered on the basis of the nearest neighbor algorithm,which improves the accuracy of the association.