Remarks on robust extraction of hand´s silhouette for 3D hand motion capture system

This paper investigates a human hand motion capture based on back projection of human hand´s silhouette images extracted from multi-camera images. The multi-camera system consists of a server-client system to run the motion capture in real time, and small USB cameras are utilized to make the system compact and low cost. To achieve robust extraction of the hand´s silhouette that is used in silhouette volume intersection for 3D reconstruction, the skin colour subtraction algorithm that combines two skin colour models in RGB colour space is proposed. One model is based on brightness and chromaticity, and the other model utilizes the normalized RG plane. To evaluate the proposed hand motion capture system, 3D reconstruction experiments of human hand posture are carried out. The experimental results show both the feasibility and effectiveness of the skin colour subtraction method and confirm that the 3D human hand posture reconstruction runs in real time.