Research on LiDAR Multi-Target Tracking Method Based on BEV Perspective

Multi-target tracking is one of the core technologies of perception in autonomous driving, but the current multi-target tracking related work has been carried out in the three-dimensional coordinate system, and the research direction of BEV perception is becoming more and more popular, and multi-target tracking from the perspective of BEV has gradually been paid attention to, so we use an improved JPDA algorithm from the perspective of BEV, use this algorithm to correlate the detection results of lidar and multi-target tracking, and add the multi-target tracking framework of Deepsort, and finally use the KITTI evaluation tool to evaluate the performance, and compare it with the three-dimensional point cloud tracking algorithms AB3DMOT, The results show that compared with the algorithm, the performance indicators are improved, and the tracking performance can still be good in the case of occlusion false detection, and the calculation speed is increased by more than 15% compared with the traditional JPDA algorithm under the same conditions.