Gesture recognition for a webcam-controlled first person shooter

Abstract

In this paper we describe a vision-based interface for a video game that relies on gestures that are made up of one pose that the player needs to hold as long as the corresponding action is desired. These gestures are easy to learn, perform and detect. After a background subtraction and skin detection stage, the method tracks the face and hands of the player by approximating them with ellipses that are updated from frame to frame. Gestures are recognized using a grid that is anchored to the player's face. The system is very robust, has a low tracking error, and a 97% recognition rate. © 2008 Springer Berlin Heidelberg.