부산대학교 도서관

Aiming at the shortcomings of current various eye tracking methods, which would result in tracking error, with improved cam shift algorithm, a rapid eye tracking method based on precise positioning iris is proposed. Firstly, the detector based on AdaBoost algorithm is used to position iris, and susan operator is used to eliminate the influence of eyeball-like factors. Secondly, with the template of iris region and analysis of the feature of object and noise's saturation, the improved cam shift algorithm based on feature fusion is used for rapid eye tacking. Allowing for the case of deformation to the target, a custom criterion is used to judge and predict center again according to its historical trajectory at the same time. Lastly, the experiment results show that this method can precisely track iris, with low error rate, its accuracy has reached more than 96% even for turning left, turning right and other conditions, and there is little difference between the position of iris estimated by tracking and its actual center, rapid speed, draw tracking time is 7.0ms and iterations are 1.6 times per frame, which are both lower than other methods, it meets the requirements of accuracy, robustness and real-time.