An algorithm for an eye tracking system with self-calibration

This paper proposes a simple and highly accurate eye tracking system. The eye image is observed by a system, such as an infrared eye tracking system, where the relative position between the head and the camera is kept invariant. Calibration where the examinee is instructed to gaze at a particular marker is avoided, and the calibration is executed automatically using only the input time sequence images. In the eye image, when the relative position to the camera is kept invariant, two queues exist to estimate the eye direction, which are the center coordinate and the flatness of the pupil. There are problems in the accuracy and the stability of the detection in the flatness data, however, although the absolute direction can be estimated. In the study, the eye direction is estimated as follows. The flatness is estimated by using multiple images. The eye parameters, such as the radius and the center of the eye rotation, are corrected, and the eye direction is estimated with stability. An evaluation experiment used actual eye images, and it was found that a more accurate measurement is realized, compared to the estimation of the eye direction based on flatness. Thus, the usefulness of the proposed method, is demonstrated. © 2002 Wiley Periodicals, Inc. Syst. Comp. Jpn.