An Efficient Eye Localization Method Based on Geometrical Approach and SUSAN Operator

Abstract

Currently, eye localization has been a popular research field, amount of methodologies and algorithms were proposed. However, most of them focused on improving detection's precision under static environment whereas required a high demand on computational resources, which leads to an obvious defect—low efficient performance in real-time video streaming eye detection. To solve this problem, an efficient eye localization method is proposed, which uses geometrical approach method and simple SUSAN operator to achieve a high efficient and accurate method for eye center and corners localization. According to the results of our experiments, the proposed algorithm can hold above 95% accuracy during whole video streaming with 90FPS, even video frames are blurred or dithering, which is enough feasible for real application, such as gaze tracking. And for static environment, the proposed algorithm also keeps an average level compared with other eye localization algorithms.