Abstract
Multiple Object Tracking (MOT) is an important task in computer vision. MOTis still challenging due to the occlusion problem, especially in dense scenes.Following the tracking-by-detection framework, we propose the Box-PlaneMatching (BPM) method to improve the MOT performacne in dense scenes. First, wedesign the Layer-wise Aggregation Discriminative Model (LADM) to filter thenoisy detections. Then, to associate remaining detections correctly, weintroduce the Global Attention Feature Model (GAFM) to extract appearancefeature and use it to calculate the appearance similarity between historytracklets and current detections. Finally, we propose the Box-Plane Matchingstrategy to achieve data association according to the motion similarity andappearance similarity between tracklets and detections. With the effectivenessof the three modules, our team achieves the 1st place on the Track-1leaderboard in the ACM MM Grand Challenge HiEve 2020.