NIR-based gaze tracking with fast pupil ellipse fitting for real-time wearable eye trackers

Abstract

In this work, a NIR (near infrared ray) based fast pupil ellipse fitting based gaze tracking system is developed for the wearable eye tracker. By a near-field and side-view eye camera, the contour of pupil in captured images is an ellipse generally, and that of pupil shape is not always circular. After pre-processing, the pupil contour is recognized by the two-stage binarizations, and the binaried pupil contour is applied to select the possible candidate points for pupil ellipse fittings. After the Random Sample Consensus (RANSAC) based fast pupil ellipse fitting, the centers of pupils are estimated effectively, and the gaze tracking is worked efficiently after calibrations. By experiments, the average estimated errors of pupil ellipse centers are smaller than 2 pixels. At training mode, the average horizontal and vertical accuracies of gaze tracking are 0.66 and 1.84 degrees, respectively. At testing mode, the average horizontal and vertical accuracies of gaze tracking are 1.13 and 2.24 degrees, respectively. Finally, the proposed main function performs up to 266.7 frames/sec by a personal computer with 3.4GHz operational frequency.