2P2-E16 Voxel based hand pose estimation using multi-viewpoint camera system

Abstract

The development of accurate 3D hand model is a key to improve a model matching based hand pose estimation. This paper proposes a method to create an individualized 3D surface model of a human hand using voxel expression from images observed by multiple cameras. The link lengths of individual fingers of the user's hand are the most important features needed to create the individualized model. To create a hand model that corresponds the closest to the user hand, the lengths of the finger links are estimated by searching a set of models generated by the motion of the finger links. Using these models, the accuracy of the hand pose estimation was evaluated.