Rao-blackwellized particle filter for multiple object tracking in video analysis

Object tracking is one of the most important tasks in video analysis systems. Starting with a precise object tracker it is possible to perform video analysis tasks such as people counting, object classification or determine abnormal behaviors to name a few. This paper reports a Rao-Blackwellized Particle Filter model for multiple object tracking. The reported model shows good results handling with single, multiple and unknown number of targets. It was also tested considering various occlusion conditions, which are not frequently reported in literature. The model works on a binary image generated with a moving object segmentation algorithm, differentiating object and background classes. This characteristic provides the opportunity of integrating this particle filter model to other segmentation algorithms and moving object detectors in video sequences. The paper reports both qualitative results and quantitative metrics to show the performance of the systems under diverse conditions.