I See What You See: Point of Gaze Estimation from Corneal Images

Abstract

Eye-gaze tracking (EGT) is an important problem with a long history and various applications. However, state-of-the-art geometric vision-based techniques still suffer from major limitations, especially (1) the requirement for calibration of a static relationship between eye camera and scene, and (2) a parallax error that occurs when the depth of the scene varies. This paper introduces a novel concept for EGT that overcomes these limitations using corneal imaging. Based on the observation that the cornea reflects the surrounding scene over a wide field of view, it is shown how to extract that information and determine the point of gaze (PoG) directly in an eye image. To realize this, a closed-form solution is developed to obtain the gaze-reflection point (GRP), where light from the PoG reflects at the corneal surface into a camera. This includes compensation for the individual offset between optical and visual axis. Quantitative and qualitative evaluation shows that the strategy achieves considerable accuracy and successfully supports depth-varying environments. The novel approach provides important practical advantages, including reduced intrusiveness and complexity, and support for flexible dynamic setups, non-planar scenes and outdoor application.