Efficient visual tracking using particle filter with incremental likelihood calculation

In this paper, we propose a particle filter that determines the weight of each particle employing the incremental likelihood calculation. Since there is usually a large overlap region between the two particles that are sequentially generated, determining the weight of the particle has only a small time cost. Therefore, the real-time performance of the proposed tracker can be dramatically improved. Extensive experimental results for single-object and multiple-object tracking scenarios are presented to demonstrate the efficiency of the proposed approach. Finally, an interesting color-based active vision system is developed for a free-floating space robot testbed to facilitate teleoperation.