Monocular Tracking of Human Motion with Local Prior Models

A novel approach to modelling the dynamics of human motion is presented.The proposed method utilizes locally learnt prior models to encode the time-varying characteristics of human motion.The local priors consist of the probability density of poses and dynamical process of motions.For each input image,the proposed method automatically learns the parameters of these models from a set of training examples that closely match with the query.The experimental results showed that the proposed method outperforms those with global motion models.