Online Multi-target Tracking for Pedestrian by Fusion of Millimeter Wave Radar and Vision
Fucheng Cui, Yuying Song, Jingxuan Wu, Zhouzhen Xie, Chunyi Song, Zhiwei Xu, Kai Ding
Abstract
Pedestrian trajectory tracking is crucial to ensure pedestrian safety in autonomous driving. Recently developed multi-sensor based multi-target tracking algorithms either overrely on the detection performance of some certain sensors or underutilize sensors' inherent information. Aiming at improving reliability and robustness of tracking under complex autonomous driving scenes, a new multi-sensor based tracking algorithm for pedestrians is proposed in this paper, which realizes sensor-fusion tracking by employing a newly proposed back-projection mechanism and a novel multi-hypothesis association approach. For the performance evaluation, massive experiments are performed to produce the dataset with 20 sequences. Qualitative experiment results demonstrate the superiority of the proposed algorithm over the single sensor based conventional algorithm. Quantitative experiment results further verify that the proposed algorithm reduces false negatives and improves tracking accuracy as compared to conventional multi-sensor based tracking algorithms.