Eye searching technique for video-based eye-gaze detection

In our video-based eye-gaze detection method, which was made for communicators for the physically disadvantaged, before starting eye gaze detection, an attendant had to manually adjust the direction of the mirror in front of the camera so that the camera would get the user´s eye. To remove this burden, we designed an automatic eye searching technique. In our system, the eye image is zoomed in for high eye-gaze resolution. Firstly, in the search procedure, to quickly find the eye, which determines the eye gaze, the eye image was zoomed out so that the camera visual field would become wide. Secondly, while the mirror was rotated stepwise in a rectangular region near the subject´s face, it was judged whether both of the two pupils existed in the camera frame. If they were judged to exist, the mirror was rotated so that only one pupil would come to the image center. Finally, the pupil image was zoomed in. The result showed that the search for the eye is possible. Although the time necessary for the search was long in the present stage of this study, other strategies for searching would make it much shorter