Probabilistic Visual Tracking using Global and Local Object's Information
Lin Gao, Zhifang Liu, Peng Tang · 2008
This paper presents a robust visual tracking algorithm by exploiting global and local objectpsilas information under a particle filter tracking framework. The proposed algorithm utilize a two-stage cascade scheme to update the particles within the state space in a coarse to fine manner. At the first stage, the object state is estimated by a coarse search using global color statistics. Then a refined estimate is achieved through exploring the local binary pattern feature. To overcome the problem of poor priors in the conventional particle filter which uses system transition as the proposal distribution, we use the state estimation made at the first stage to construct the proposal distribution that seamlessly integrates the current observation. Experimental results demonstrate the efficiency and accuracy of the proposed algorithm.