Eye Detection for Gaze Tracker with Near Infrared Illuminator

Abstract

In this paper, an eye detection method is proposed based on the features of eye images generated in the gaze tracking environment. By configuring the corneal specular reflection detector, pupil detector, and eyelid detector in a cascade format, in actual applications reliable eye detection can be achieved even when an image of only one eye is captured because a part of the user moves out of camera range or when changes occur in the brightness and focus because the user moves back and forth. For the corneal specular reflection detector, a block setting method and a thresholding method are proposed to ensure the corneal specular reflection is not missed. Furthermore, a method of suppressing the effects of eyelashes and corneal specular reflection is proposed that allows more accurate determination of the pupil in the pupil detection process. The eyelid detector determines the presence/absence of an eyelid by using a vertical integral projection function. The eyes were detected reliably when the proposed methods were applied to the captured images in an actual gaze tracking system.