Color appearance-based approach to robust tracking and recognition of multiple people

In this paper, we propose an automated tracking system which can robustly recognize multiple people in the video sequence. By checking the foreground coverage measure, the system first segments and localizes isolated people from multiple foreground regions. We then model each tracked person by his/her color histogram so that the system is capable of performing optimal recognition after occlusion from a statistical viewpoint. Since shadow could severely degrade color histograms, we propose a novel shadow-removal scheme to suppress shadow effects and thus help improve the reliability of people recognition. The people recognition technique is also extended to identify people that re-enter the scene of a closed-environment.