Robust real-time multi-user pupil detection and tracking under various illumination and large-scale head motion

A novel approach to Robust real-time multi-user pupil detection and tracking is presented, and this kind of detection and tracking behaves well under the circumstance of various illumination or large-scale head motion. Firstly, with active IR illumination, the possible positions of human pupils are depicted according to bright pupil effect and then some image pretreatment is conducted to diminish the fake pupil positions. Secondly, other than detecting human pupils directly, human faces in the image would be detected with real AdaBoost and the detected face positions would be optimized in order to save the time of whole processing. Thirdly, based on the faces detected, human pupils would be detected with real support vector machine (real SVM) and correlation matching. At last, the human pupils detected would be tracked with Kalman forecast in order to save the detection time of next image. Results from a series of experiments show that the new method could achieve real-time (30 frame per second) with a success rate of 95% for multiple users, and it is also proved that the new method is robust for illumination variation and large-scale head motion.