Human motion estimation from monocular image sequence based on cross-entropy regularization

Human motion estimation is crucial for many important applications. In this paper, a novel approach to human motion estimation from monocular image sequence is proposed. First, a non-rigid motion model called relative deformation model is developed. This model is based on the notion of relative deformation that introduces a new way for anthropomorphic body locomotion analysis including clinical gait analysis and robots motion analysis. Then, in order to deal with the ill-posed estimation problem, a regularization method based on Kullback's cross-entropy is proposed. By imposing the motion smoothness constraint, the entropy regularization converts the ill-posed problem into a well-posed one and guarantees the unique solution. Experimental results on image sequences of different walking men with different motion pattern demonstrate the feasibility of the proposed approach.