Research On Multi-target Tracking Method Based On Multi Line Lidar

Tracking multiple moving targets around in a real traffic environment is often prone to the situation that the moving target in front is blocked or the sensor is temporarily blocked.At this time,the shape feature information and centroid coordinate information of the moving target detected by the sensor will change,which will lead to instability or even failure in the tracking of the moving target.To solve this problem,this paper studies the related technologies of multi-target tracking.(1)The occlusion level and classification criteria of detection targets based on multi line lidar are designed.Taking automobile as the main research object,the detection targets based on multi line lidar are divided into three levels: no occlusion,light occlusion and heavy occlusion.The classification basis is determined through experiments: when the ratio of the number of laser point clouds of the detection target before and after occlusion is more than 70%,the detection target is in the state of no occlusion;When the ratio of the number of laser point clouds of the detection target before and after occlusion is between50%～70%,the detection target is slightly occluded;When the ratio of the number of laser point clouds of the detection target before and after occlusion is less than 50%,the detection target is in the state of severe occlusion.(2)Improve the data association algorithm to improve the accuracy of multi-target tracking in occluded environment.Firstly,the tracking double gate filtering algorithm is used to screen the detection targets within a certain range of the centroid coordinates of the tracking target,so as to reduce the calculation of the correlation degree.Then,different correlation degree calculation methods are used for the detection targets with different occlusion degrees.For the detection target without occlusion,refer to the correlation degree calculation method provided by Apollo;For the detection target with slight occlusion,the correlation degree calculation method based on endpoint coordinates is used;For heavily occluded detection targets,the correlation degree is not calculated.By setting detection density threshold and frame loss density threshold,it is ensured that sudden obstacles will not become new tracking tracks immediately,and the tracking tracks with short-term loss of detection targets or severe occlusion will not disappear immediately.The experimental results show that the accuracy of the improved data association algorithm is5% to 10% higher than that of the common methods,and the tracking trajectory is more stable.(3)The location information of the vehicle and the detection target information are fused to obtain the coordinate information of the moving target in the global coordinate system.The global coordinates are used for state estimation to avoid the influence of autonomous vehicles on the state estimation of moving targets.It is verified by experiments that the motion state of moving targets can be estimated more accurately by using global coordinates than radar coordinates.Secondly,using the improved interactive multi model algorithm,the autonomous vehicle can track different models at the same time.Through simulation,the state of the moving target calculated by the algorithm is compared with the original data.The results show that the algorithm can accurately obtain the state of the moving target when the moving target moves in a straight line and curve.