High Precision Point of Gaze Estimation Using an Add On Fixture to Mobile Devices Based on Corneal and Retinal Retro-Reflections

Abstract This work proposes a robust and convenient method for point of gaze (POG) estimation based on corneal and retinal retro-reflection detection and calculation of the glint-pupil vector. The system concept is simple and cost effective that can be attached to mobile devices while using the existing visible camera of the mobile device and additional fixture that includes an add on a near infrared (NIR) camera and a NIR light source. The overall additional system requirement is minimal, namely loos alignment tolerance of the fixture attached, and simple design. During NIR illumination of the eyes, corneal and retinal reflections are detected. An algorithm calculates the eye and the system distance, it calculates also the angle and the distance between the corneal and retinal retro-reflection thus provides the direction of the gaze and the POG estimation. Experimental results showed high precision of the estimated point of gaze that is much better than 1 degree (vertically and horizontally) in this very simple and cost effective add on system. We found out that such accuracy of POG is achieved even for coarse mechanical mounting alignment between the existing mobile device camera and the add-on system. Keywords: point of gaze; pupil detection; eyes movements tracking; digital image processing