Real time hand gesture recognition using 3D prediction model

This paper proposes a new hand gesture recognition method using a 3D prediction model used for developing human-computer interfaces. It focuses on real-time hand-pose estimation, which is indispensable for virtual object manipulation. In the first stage, simple but effective parallel pipelined algorithms are employed to extract hand features from binocular images. In the second stage, the hand pose is estimated using the hand's features. Even if occlusion occurs, a ghost correspondence eliminating table, used with the 3D prediction model, makes the estimation possible on the assumption that finger tips are occluded by other finger tips. The hand-feature extraction method, the pose estimation method, the constructed human-computer interfaces, and a hand-pose estimation experiment are discussed.<
>