Calibration-free eye tracking by reconstruction of the pupil ellipse in 3D space

Abstract

Most video-based eye trackers require a calibration procedure before measurement onset. In this work a stereo approach is presented that yields the position and orientation of the pupil in 3D space. This is achieved by analyzing the pupil images of two calibrated cameras and by a subsequent closed-form stereo reconstruction of the original pupil surface. Under the assumption that the gaze-vector is parallel to the pupil normal vector, the line of sight can be calculated without the need for the usual calibration that requires the user to fixate targets with known spatial locations.