An appearance-based approach to gesture-recognition

Abstract

This paper describes techniques for the design of a system able to interact with the user by visual recognition of hand gestures. The system is composed of three modules including tracking, posture classification and gesture recognition. A description of each module is given. In order to increase the robustness and the precision of the tracking, several complementary tracking processes are coupled. A classification process is presented for recognizing hand posture using distance in an eigenspace. The classification of hand posture leads to the gesture recognition by a set of finite state machines.