Human Pose Estimation Using Articulated ICP

This paper proposes a method of human pose reconstruction using the depth data inputs in multi depth sensors. Utilizing the articulated ICP technique for pose estimation, the proposed method firstly maps a model from parent nodes, and then searches the points adjacent to joints. The proposed method addresses the challenges of the existing method of using a single sensor such as occlusion, and enables real-time 3D pose reconstruction involving movements and 360°omni-directional rotations while tracking 1~5 user IDs. Precluding the necessity of pre-training data or motion capture data, the method uses a simple model to compute the ICP and applies the point search, which allows the fast and accurate pose reconstruction.