Corneal-glint-based estimation of the binocular point of gaze without worn sensors

Abstract A theoretical framework for binocular model- and glint-based gaze estimation is presented including the formalization of several anatomical and physiological constraints. This framework is applied within a method that estimates the point of gaze on a surface that is parametrized as a triangle mesh. The scene surface may be positioned at different distances from the eye. The method is based on corneal glints and the pupil contour and uses three subject-specific eye parameters. It requires tracking both eyes and knowing the location of the scene surface. It does not require tracking the head pose. Also, eye torsion is constrained implicitly and Listing’s law is not used. The method is experimentally evaluated in a scenario with gaze targets in a 3-dim. volume, with standing subjects that did not wear any sensors, with downwards gaze directions, and with unrestricted, natural head poses. Averaging over all experiments yields a correctness better than 1.1° of visual angle.