Long Term Multi-Target Tracking based on Detection and Data Association
Ai Min Li, Pil Seong Park · International Journal of Electronics and Electrical Engineering · 2013
Multi-target tracking is widely studied, but it is still an attractive but difficult research area because of existence of occlusion and interaction between target images. We propose a novel detection-based multi-target tracking method using data association. The main contribution of our method is providing a strategy to quickly correct the wrong tracker aroused by occlusion. We use somewhat unreliable detection confidence to assist a particle filter tracker. The resulting algorithm is tested using movies having much occlusion. The result shows good performance, fast enough to run real time for a relatively long period of time.