Visual Tracking Algorithm by Particle Filter and EM Based on Multi-cues

To overcome the real-time and robust problem of visual tracking, a visual tracking algorithm based on a fusion of multi cue and particle filter was proposed. Firstly, the integrating formula was transmuted on the framework of particle filter base on multi-cues which integrated multi-cues based on cues rather than feature points so that it could be used for parallel computing. Secondly, the EM algorithm was used in multi-cues integration without the weight of each cue which improved the visual tracking algorithm robust by acclimatizing the appearance model to the object change. Experiment results show that the algorithm can track targets in complex environments and under pose and local occlusion.