Robust eye tracking and location method based on Particle filtering algorithm

Abstract

In this paper we present a fast vision-based eye-gaze tracking method based on Particle filtering algorithm in the condition of near-infrared light and single-camera, against to the requirement of real-time eye tracking in engineering, and the fact that presently most of eye tracking methods in video are not precise, target easy to lose. In the initialize step, we use a high accuracy cascaded classifier trained by AdaBoost algorithm to get the primitive information of eye region. Considering the eye region information in the last frame image is valuable to the next frame image analysis, the particle filter algorithm is adopted to accomplish the eye region tracking. Experimental validations show that the processing time for each single frame is effectively reduced by using the constraints between the last and next frames, for it reduce the search range of the human eye. Finally, we design a segmentation method with double thresholds to extract the pupil and Purkinje bright spot from contours, which conduce to pupil positioning and distinguish the eye region.