Online Multi-camera Tracking-by-detection Approach with Particle Filter

Multi-camera tracking is the foundation in many applications such as visual surveillance and event recognition. In this paper, we address the problem of automatically detecting and tracking unknown number of people in multiple synchronized cameras. We present a novel approach for multi-camera tracking-by-detection with particle filter. To achieve the necessary robustness, the target-individual appearance model and the location likelihood is integrated to associate detections to trackers. Furthermore, particle filters are generated to cope with fails in detecting. Contrary to recent approaches, we focus on designing an online algorithm which is more suitable for real world applications and only makes use of information from the past. Our experiments on public data set yield good tracking performance and we demonstrate the validity of our approach with quantitative evaluation.