Particle filter tracking with Mean Shift and joint probability data association

The particle filtering technique with multiple cues is a powerful technique for tracking objects in image sequences. In this paper, we proposed a novel particle filter which embeds the Mean Shift optimization and a data association technique based on the joint probabilistic data association (JPDA). We use the adaptive mixture of color and texture cues to represent the distributions of the target. The Mean Shift algorithm is used to help particles move to better positions which are near their original positions. The data association algorithm handles the uncertainty of the measurement origin. Experiment results evaluate the performance the proposed particle filter.