Mixed-State Particle Filtering for Simultaneous Tracking and Re-identification in Non-overlapping Camera Networks

This article presents a novel approach to person tracking within large-scale indoor environments monitored by non-overlapping field-of-view camera networks. We address the image-based tracking problem with distributed particle filters using a hierarchical color model. The novelty of our approach resides in the embedding of an already-seen-people database in the particle filter framework. Doing so, the filter performs not only position estimation but also does establish identity probabilities for the current targets in the network. Thus we use online person re-identification as a way to introduce continuity to track people in disjoint camera networks. No calibration stage is required. We demonstrate the performances of our approach on a 5 camera-disjoint network and a 16-person database.