Joint Sequential Shape Classification and Piecewise Elastic Motion Estimation

This paper proposes a novel motion model for classifying general non-rigid motion into piecewise elastic motion, so as to achieve the non-rigid motion estimation without any priori shape models. Three interrelated sub-problems have to be addressed: classifying the whole motion sequence, estimating motion inside each segment and connecting the piecewise motions. In this paper, a Markov chain is used to classify the motion into a pre-defined number of classes, at the same time the estimation of elastic motion in individual segments of the sequence, and then all piecewise motions are connected with the correspondence between neighbor frames. Finally, the experiment results on human motions show the capability and robustness of proposed algorithm.