Face Tracking Based on Particle Filter with Multi- feature Fusion

Traditional particle filter cannot accommodate to the environment of background interferences, illumination variations and occlusions. This paper presents a face tracking method with fusion of color histogram, contour features and grey model based on particle filter. First, it brought in contour features as the main cue of multiple features when tracking the face without stable color histogram. Then, as prior information was neglected in traditional particle filter, this paper employed GM(1,1) model to yield proposal distribution, such that the proposal distribution would bear a higher approximation to posterior probability. Finally, in the importance sampling step, sampling was corresponded to the particle weight in case of the particle degradation. The experiments show that our method outperformed the previous with more accuracy and flexibility, particularly under the condition of color background interferences, drastic illumination variations and complete occlusions.