A Hierarchical Pose Estimation Method Based on Graph Model

Posture estimation plays an important role in human-computer interaction, many algorithms have been presented, but their precisions are not good enough, moreover, some approaches can only be applied under certain initial gestures. We propose a hierarchical estimation of body poses by adopting geodesic information in initiation and combining curvature constraints into graph model. Unlike other approaches, we regard the most remarkable joints rather than torso as the initial joints, and employ constraints of neighbor joints to locate the less distinguishable joints. Experiments show that constraints passing though the body make results to be robust and accurate.