Hybrid conditional random fields for multi-object tracking with a mobile robot

As the precondition to perform many tasks including person following and dynamic obstacle avoiding, object tracking is very important for mobile robot systems especially in populated dynamic environment. A novel hybrid conditional random field model which has a hierarchical structure and includes hidden states is proposed for multi-object tracking with a mobile platform. Since conditional random field is a kind of discriminative model which makes no assumptions about the dependency structure between observations and allows non-local dependencies between state and observations. The proposed method cannot only integrate moving object detection and tracking perfectly well, but also can fuse multiple cues including shape information and motion information to improve the stability of tracking. Experimental results with the mobile robot developed in our lab show that the proposed method has higher precise and stability than JPDAF.