Head Pose Classification by using Body-Conducted Sound

Abstract

Vibrations generated by human activity have been used for recognizing human behavior and developing user interfaces; however, it is difficult to estimate static poses that do not generate a vibration. This can be solved using active acoustic sensing; however, this method is not suitable for emitting some vibrations around the head in terms of the influence of audition. Therefore, we propose a method for estimating head poses using body-conducted sound naturally and regularly generated in the human body. The support vector classification recognizes vertical and horizontal directions of the head, and we confirmed the feasibility of the proposed method through experiments.