Partial Observation vs. Blind Tracking through Occlusion

This paper presents a framework for multi-object t racking from a single fixed camera. The region-based representations of each object are tracked and p redicted using a Kalman filter. A scene model is created to help predict and interpret the occluded or exiting objects. Unlike the traditional blind tracking during occlusion, the object states are updated using partial observations whenever available. The observability of each object depends on the predictive measurement of the object, the foreground region measurement, and p erhaps the scene model. This makes the a lgorithm m ore robust i n terms of both qualitative a nd quantitative criteria.