Abstract
In recent studies of a gaze tracking system using 3D model-based methods, the optical axis of the eye is estimated without user calibration. The remaining problem for achieving implicit user calibration is to estimate the difference between the optical axis and visual axis of the eye (angle kappa). In this paper, we propose an implicit user calibration method using saliency maps around the optical axis of the eye. We assume that the peak of the average of a sequence of saliency maps indicates the visual axis of the eye in the eye coordinate system. The angle kappa is estimated as the difference between the optical axis of the eye and the peak of the average of saliency maps. We developed a prototype system with two cameras and two IR-LEDs. The experimental result showed that an accuracy of 1.5 degrees was achieved using the deep learning-based saliency maps.
