A clustering approach to the vision-based interface for interactive computer games

Abstract

In interactive computer games, vision can be a powerful interface between humans and computers. In this paper, we propose a vision-based interface for 3D action games. We make dynamic gestures to input of the interface and represent a user's gesture as an ordered sequence of a user's poses. To estimate a human poses, we classify whole frames using K-Means clustering. For recognizing a gesture, each symbols from input sequence is matched with templates composed of ordered pose symbol sequences that indicate the specific gestures. Our interface recognizes ten gesture commands with a single commercial camera and no markers. Experimental results with 50 humans show an average recognition rate of 93.72 % per a gesture command.