Adaptive Fuzzy Particle Filter Tracker for a PTZ Camera in an IP Surveillance System

We propose an adaptive fuzzy particle filter (PF) (AFPF) method adapted to general object tracking with an IP pan-tilt-zoom (PTZ) camera. PF samples are weighted using fuzzy membership functions and are applied to geometric and appearance features. In our PF, targets are modeled and tracked based on sampling around predicted positions obtained by a position predictor and moving regions detected by optical flow. Sample features are scored based on fuzzy rules. In this paper, we apply the AFPF to a human-tracking application in an IP PTZ surveillance system. Results show that our system has good target-detection precision (>; 93.9%), low track fragmentation, and a high processing rate, and the target is almost always located within one-sixth of the image diameter from the image center.