Coupled State-Space Models for Tracking People through Camera Networks

Emerging new surveillance technologies that use camera networks offer unprecedented capabilities for law enforcement agencies as well as health-care opportunities. A challenging problem is to track people through very sparse sensor measurements to reduce the cost of expensive sensors and be robust to sensor failure. In this paper we propose a Coupled State Space Model (CSSM) to track people across camera networks using a very sparse set of measurements. CSSM simultaneously models the geometry of the camera network as well as the dynamics of the human being tracked. We apply CSSM to the problem of tracking elderly people in a nursing home. Experiments on synthetic and real data show that CSSM can predict the states of cameras and reconstruct the trajectory of the walking person, using very sparse labeling information.