Abstract
In this paper, we propose a shape reconstruction method for multiple human bodies with occlusion by human body parts matching and temporal pose data of human body parts. We define occlusions as three types: First type is a mutual-occlusion occurred between multiple human. Second type is a self-occlusion occurred on a single human body. Third type is a mutual-self-occlusion occurred in a case when the self-occlusion occurred in the mutual-occlusion. Firstly, the proposed method reconstructs human shape in the mutual-occlusion and the self-occlusion scenes by matching human body parts obtained in advance to human surface data out of these occlusion regions. Secondly, in order to reconstruct shape of human bodies in the mutual-self-occlusion, the proposed method estimates poses of human parts in the mutual-self-occlusion region by using temporal position-pose data of these parts and reconstructs shape of human bodies by applying these parts with estimated poses. In this paper, we quantitatively and qualitatively evaluated effectiveness of the proposed method through simulation and experiments with actual human bodies.
