Visual Tracking Based on Mixture Motion Model and Incorporate Observation Distribution

Accurate visual object tracking through long sequences is a challenging task since object's appearance changes and complex motion happens. We present mixture motion model and incorporate observation model within the Monte Carlo framework to achieve robust visual tracking. The mixture motion model which employs important history motion information of the target is built according to a motion measurement matrix to model the target's transition state. Meanwhile, the incorporate observation model is established by introducing SVM classification scores into normal tracking observation model. A particles filter's implementation with these mixture models is demonstrated, which leads to robust tracking results, especially in occlusion and complex scene.