Study Of Gaze Tracking Algorithm Based On Monocular Camera In Natural Light

With the development of computer technology, human-computer interaction theory and technology gets more and more concern and attention. Eye gaze tracking technology, which can obtain the line of sight directly, is of great significance in driving human-computer interaction of technology change, as well as machine vision and psychology research. Gaze tracking technology based on monocular camera in natural light, which has no need for active infrared light and complex hardware-based calibration, can be more widely applied. However, there are still many basic theoretical problems in this field have not been solved well, which are accurate extraction of eye feature information and the head movement tolerance ability of gazing point mapping model. Therefore, according to theses question, a gaze tracking algorithm which can tolerance head free movement in natural light is proposed. The main contents are as follows:(1) Eye movement features extraction in natural light. According to the effects of illumination change and head movement, a face features initial location method based on shape regression is used to determine the regions of interest of eye movement feature points. Next, according to specific physical shape of each corner features, a screening method based on corner score calculation is used to locate the inner corners of eyes and mouth corners. Outer corners of eyes is detected by a curve fitting method. For iris feature, a sliding window based method is proposed for iris edge extraction and iris center positioning.(2) Gaze mapping model construction with ability of head movement tolerance. Firstly, eye movement vector based on eyes data is constructed by iris center and eye corners. Secondly, according to the relation of eye features points position on image while head moving, a method based on projective mapping correction is proposed to adjust the eye movement vector and reduce the impact of head movement. Finally, relation between head movement and gaze point deviation is obtained by support vector regression. The gaze mapping model which can satisfy free head movement is constructed after the estimate results of gaze points are compensated based on support vector regression model.The eye features extraction methods in the paper can effectively locate the eye corners and mouth corners in natural light, and the accuracy rate of iris feature detection method can be up to 95.6% with different eye position and head posture. The accuracy of gaze tracking, which combined with eye movement vector correction and gaze point compensation models, is about 2.67°. Experiment results show that the proposed gaze tracking algorithm can solve the problems better when gaze tracking based on monocular camera in natural light.