Real-time eye-gaze estimation by using a virtual reference point

A human interface that uses a line of sight has been studied. In previous studies, some methods use a special light source or a hardware device and others use a software program for image processing for detecting the line. In this paper we propose a new eye-gaze input method. The method uses only a single facial image acquired under ordinary lighting condition, and it requires no laborious camera calibration. A state of eye-gaze is estimated by a virtual reference point, which is determined based on a motion of feature points on a face. We demonstrate the validity of the new method by simulation and experiments.