Using head poses to control a virtual robot walking in a virtualmaze

This paper develops a game system based on head pose identification techniques to control a virtual robot walking in a virtual maze environment built with Microsoft Robotics Developer Studio. Adaboost face classifiers are used to detect a face in the image and give a rough head pose classification. Two approaches are proposed to further determine more specific head poses. The first approach adopts PCA with nose template matching method, and the second one firstly uses a nose classifier to detect the nose in a detected face region and then directly utilizes the nose's relative position information to achieve head pose classification. A method to select the best nose area is proposed to solve the problem that multi-nose areas may be detected in one face region. Experiment results show that the two approaches have comparable time consumed, but the second approach provides more correct identification results.