A.Pérez, M.L.Córdoba, A.García, R.Méndez, M.L.Muñoz, J.L.Pedraza, F.Sánchez.
Universidad Politécnica de Madrid. Facultad de Informática
Departamento de Arquitectura y Tecnología de Sistemas Informáticos
Boadilla del Monte, 288660 Madrid
Spain
e-mail: aperez@fi.upm.es |
http://www.datsi.fi.upm.es |
Keywords: Pattern recognition, image processing, eye-tracking, gaze-tracking
Abstract
This paper describes a real-time contact-free eye-gaze tracking system that bases its accuracy in a very precise estimation of the pupil centre. The eye camera follows the head movements maintaining the pupil centred in the image. When a tracking error is produced, the image from a camera with a wider field of view is used to locate the eye and quickly recover the tracking process. Four infrared light sources, synchronised with the shutter of the eye camera, are used to produce corneal glints. Its special shape has been exploited to allow the optimisation of the image processing algorithms developed for this system. Special care has been taken in limiting the illumination power and working way below the dangerous levels. After a calibration procedure, the line of gaze is determined by using the pupil-glint vector. The glints are validated using the iris outline with the purpose of avoiding the glint distortion due to the changes in the curvature on the ocular globe. The proposed algorithms determine the pupil centre with sub-pixel resolution, minimising the measurement error in the pupil-glint vector.