Recovering three dimenensional hand motions of sign language from monocular image sequence

Abstract

This paper proposes a method to recover three dimensional hand motions of Japanese sign language. This method uses skin-color information and edge and difference images. In this method, joint points are determined by considering a human body consisting of hierarchical structure. Also, we attempt to spot each word from a sentence by using estimated hand motion transitions. The experimental results show the validity of the proposed method.