Abstract:

To improve the calculative precision of gaze direction, an eye gaze tracking system using stereovision information is designed, it composes of two charge coupled device cameras and a novel loop illuminator. Pupil position is got through doing subtraction of the two adjacent images, “bright pupil” and “dark pupil”, which are gained by illuminating users eyes alternately. Local gazing direction is the vector difference between the Purkinje image in “dark pupil” and pupil center located by ellipse fitting. Pupil tracking is designed using neural network combined with Kalman filtering, and mapping model from 3D parameters to gaze point is worked out via support vector regression. Experiment proves that this system can most eliminate the interference from users head motion.

Key words: eye gaze tracking, human-computer interaction, pupil, Purkinje image