Research on feature points positioning in non-contact eye-gaze tracking system

The accuracy and the real time ability of positioning the feature points, including the four reference glints and the center of the eye pupil, are essential to the eye-gaze tracking system based on the cross-ratio. In the paper, in order to handle the two issues in the images to be processed, that is, the similar gray values and the low resolution, an algorithm of feature points positioning with strong adaptability is proposed, namely, using the improved Otsu method for detecting the glints, and adopting a method combining emitting rays with ellipse fitting for positioning the center of the pupil. Some experiments are performed on the non-contact eye-gaze tracking system on a VC ++ 6.0 platform, and the results show that: when the eye is located 70cm in front of the screen (33cm times 26cm), 30 times 25 points can be distinguished, and the processing requires only 16ms on average.