Virtual playing ground interface using upper-body gesture recognition

In this paper, we describe a real-time interface using upper-body gesture recognition for interactive game systems. To recognize human gesture, firstly we analyze the pose of the upper-body. However, because of the complex structure of the human body, analyzing the pose is very difficult. In order to solve the difficulty, our approach uses two images that are from two infrared cameras being set at front and side of user. We analyze the image by deleting the background and transform it to silhouette image, and then apply it to a SVM classifier using non-linear RBF kernel function. By using the RBF kernel, it makes easy to classify the complicated poses in non-linear space. Since the gestures are temporal sequences of the poses, so transforming it with a state transition matrix is very desirable and by using the matrix, gesture recognition straightforwardly can be done. The recognition result is used to control the game as an interface by mapping a user´s action onto the game system as an adequate command.