Head Posture Estimation Method for Vision Based Human Interface

In this paper, we propose a method to estimate posture of human head, which is suitable for constructing vision-based human interface. By the proposed method, the position and the direction of the head are estimated from the image of singular marker put on the head of a user. As the feature, moments of image up to second degree are used. Posture is estimated by estimation system which consists of observation model based on local analysis and compensation function. Our method has the following characteristics important for posture estimation method as an element of general use human interfaces. Firstly, it does not constrain user, since it is vision based method. Secondly, a user only have to put singular marker on the top of the head, and thus easily puts on and takes off the marker. Thirdly, because the features to represent the marker are moments up to second degree, we can expect considerable robustness to noise and less computational cost. The experimental results shows validity of our method.