Model-Based Gaze Direction Estimation in Office Environment

In this paper, we present a model-based approach for gaze direction estimation in office environment. An overlapped elliptical model is used in detection of head, and Bayesian network model is used in estimation of gaze direction. The head consists of two regions which are face and hair region, and it can be represented by two overlapped ellipses. We use its spatial layout based on relative angle of two ellipses and size ratio of two ellipses as prior information for gaze direction estimation. In an image, the face regions are detected based on color and shape information, the hair regions are detected based on color information. The head is tracked by mean shift algorithm and adjustment method for image sequence. The performance of the proposed approach is illustrated on various image sequences obtained from office environment, and we show goodness of gaze direction estimation quality.