Real Time Face Tracking Using Particle Filtering and Mean Shift

Particle filter is widely used in object tracking. However, it has one notable weaknesses that is sample degeneracy problem. This paper proposes a novel algorithm to overcome this problem by incorporating mean shift into particle filtering. Mean shift reacting on sample herds the samples in the reference mode area, which could make less samples be used while tracking. The propose algorithm is used in face tracking. Results demonstrate that our approach has better performance than that of the mean shift tracker and the conventional particle filter. Moreover, the computation time in each frame is less than that of the mean shift tracker or the conventional particle filter.