Pose estimation of human body based on silhouette images

Abstract

A method for articulated-pose identification from the silhouette of a human body is presented in this paper. First the method employs an energy function to extract a dummy skeleton of a human body from the silhouette, and then estimates the position of the joints by the knowledge of anthropotomy and three rules given by the standard human skeleton model. The three rules can help us to judge whether a point on the skeleton is a joint or not. The experimental results show that, the method is effective. It requires no any additional restrictions on the motion and color of a human body. Meantime the method is robust against the noise in the silhouette. Especially it has good performance in the case of complicated background.