Robust Tracking of Walking Persons by Elite-Type Particle Filters and RGB-D Images

In this paper, we propose a robust real-time tracking system using RGB-D image sequence which are obtained through stereo camera. We apply ‘Elite-type’ particle filter, which is novel structure of particle filter, for tracking multiple persons. In Elite-type particle filter, to be robust to change of appearance and partial occlusion, likelihood is designed based on histogram and each particle possess their own model histogram. The system assign this particle filter to each person, and estimate state of the target person which vary from frame to frame. Furthermore, the system is able to measure the height of person’s head, which is effective for analysis human behavior. Real-time tracking performance of multiple persons was confirmed by experiments which simulating a real shop.