A Long-time Multi-object Tracking Method for Football Game Analysis
Yuan Chen, Wenbin Huang, Sailing He, Yaoran Sun · 2019
Multi-object tracking (MOT) is crucial in many applications, such as video analysis, intelligence surveillance, and robot navigation. Occlusion, complicated motion, and similar appearance pose challenges to achieving reliable MOT. Here we propose a long-time multi-object tracking method to overcome these challenges. In our framework, a number recognition method based on tracklet is proposed to provide reliable number labels, which facilitates long-time tracking in football video. Extensive experiments are taken to demonstrate that our method achieves excellent performance and provides reliable trajectories for football game analysis.