3D Hand pose modeling from uncalibrate monocular images

Abstract

Hand pose modeling is one of the most challenging research areas in Computer Animation due to the difficulty of hand model, which consist of more than 20 degrees of freedom and require the construction of a large high-dimensional database of poses. In this paper, we propose a method of 3D hand pose modeling based on a new set of constraints and optimization framework to reduce the ambignity between 2D joint location and 3D hand poses. Our method does not need prior knowledge of the bone length and camera parameters. Furthermore, a large high-dimension database used for searching is not needed to be constructed. An experiment was performed on the real data set captured from a webcam. the results shows that the proposed set of constraints are sufficient to reconstruct the hand poses with desirable accuracy.