A methodological approach on real-time gesture recognition using multiple silhouette models

This paper describes a new method of real-time gesture recognition and its applications. The key idea of the method is that we use multiple silhouette models of human body and estimate the pose of arms and legs by comparing similarities of the input image to the models. To calculate the similarity for a certain specific part of the body, such as arms, a simple mathematical operation deducting the inference of different parts is adopted. As an application of gesture recognition, we realize a real-time interactive system, a gesture game, which can animate the game characters by gestures in real-time. Stretched arms like flying aeroplane, for example, can be interpreted into button operations for a flying simulation game with this system.