View article

Human eye glints localization could significantly and directly improve accuracy of gaze tracking, especially when several glints are localized precisely at the same time. However, traditional algorithms have not ensured the accuracy of glints localization to be used for gaze tracking. Aiming at the problem in infrared video oculography that those glints reflected on the surface of eye iris could be hard to identify and localize, we propose an algorithm for precisely and exhaustively locating eye glints of high accuracy with accelerated process. Our contributions are two-folded. (1). We propose an exhaustive eye glints localization algorithm, which could guarantee both 94.9% recognition rate and < 1 pixel positional accuracy under various brightness conditions. (2). We propose methods to accelerate our algorithm, including a modified quadratic ellipse difference algorithm and video frame difference examination, which …