The Improved Particle Filter for Object Tracking

We address the problem of object tracking encountered in video processing. The proposed approach is mainly composed of object modeling, the improved particle filter and mixture filtering. First, each of visual objects can be modeled by multi-part color likelihood model. To tackle self-occlusion of the tracked objects, the color distribution representing the tracked object can be updated over time. We use the improved particle filter to ameliorate the performance of the classical particle filter. To track multiple objects simultaneously, we use multi-component mixture model whose components are modeled by the improved particle filter to form tracking algorithm. Experimental results show the proposed method performs object tracking effectively