AN IMPROVED PUPIL LOCALIZATION TECHNIQUE FOR REAL-TIME VIDEO-OCULOGRAPHY UNDER EXTREME EYELID OCCLUSION

摘要

Video-oculography (VOG) is a useful tool for detecting eye movement abnormalities as well as for human-computer interaction. Performance of a real-time VOG system relies on how the center of pupil is estimated under various occlusions. However, the accuracy of the state-of-the-art algorithms decreases significantly with extreme eyelid occlusion, although they can estimate pupil center under different illumination conditions and irregularity of pupil contour. To handle this problem, we propose a novel method to localize pupil in a real-time VOG during extreme eyelid occlusion. We have improved the geometric ellipse fitting method by incorporating a cumulative histogram processing for robust binarization and the random sample consensus (RANSAC) for removing outliers. Experimental results with 90% eyelid occlusion show that the proposed algorithm increases the accuracy of pupil localization up to 51% compared with a state-of-the-art method. During real-time pupil tracking, the proposed algorithm achieves significant accuracy improvement (N = 9000 frames; p ＜ .05) with less than 10 ms of computational time per frame. The proposed method shows promising results for accurate measurement of horizontal and vertical eye movements.