Tracking Of Moving Objects Using Multiple Feature Fusion

Tracking of moving objects from different nature such as vehicles, human faces or body has become a major topic in video applications and widely applied in many domains like video surveillance, security systems and human behavior analysis.This thesis presents a multi-feature fusion model based on a particle filter for moving object tracking. The particle filter combines color and edge orientation information by a stochastic fusion scheme. The scheme randomly selects single observation model to evaluate the likelihood of some particles. The stochastic selection probability is adjusted adaptively by the uncertainty associated with a feature model. The experiment shows that the proposed method has strong racking robustness and can effectively solve the occlusion problem.