The Research Of Eye Gaze Tracking Technology Based On Pupil Corneal Reflection

Gaze direction represents a person’s attention, and eye-tracking technology is a technique to calculate the direction of people’s attention. Eye tracking has vast prospects in the fields of interaction and diagnostics, such as human-computer interaction(HCI), virtual reality, eye disease diagnosis, human behavior research analysis and so on. Eye tracking technology can be divided into contact-type and non-contact-type according to whether it disturbs the users. Non-contact type is more popular since it is more convenient. Although gaze tracking technologies based on pupil-corneal reflection already obtained good research results, but it still has great space for improvement in some key issues. The main contribution of this study is as follows:A new method that used K-Means clustering and improved Hough transform for iris area localization is proposed to replace eye area localization. Because all the parameters needed for gaze points calculation are located in the iris region, which can ensure that accurate positioning of feature points extracted from human eye in the premise of remove the confounding factors such as eye lashes and eyebrows.A new feature spots search method based on gray scale is proposed. The method uses an adaptive threshold method for image segmentation, combined with the geometric characteristics of the light sources to achieve matching image spots. Compared with traditional methods, the image segmentation of adaptive threshold ensures complete spots. And geometric-characteristics of image ensure that the spots can match light sources exactly.1D line operator is used to detect the limbic boundary points. And a cycle Least Square pupil fitting method is used to reduce the effect of accuracy rate of ellipse fitting from image noise.A dynamic adaptive calibration method is proposed aimed at the two assumptions of the cross-ration model. Screen is divided into n*n sub-region in the method. And one pair of mapping parameters was calculated for each sub-region. Two parameter-matrixes are needed to each user. The current parameters are the optimal parameters obtained dynamically from the matrixes in real time. This method has avoid the overall shift problem appeared in the traditional methods, and has improve the local accuracy of the system.A group of accuracy testing experiments and two groups of application experiments are done to verify the proposed method. In the research condition of this paper, the distance error of gaze points can converge to about 20 pixels in the accuracy testing experiment. The other experiments are application experiments, which using eyes control for text input, and another application experiment is design to detecting and analyzing human attentions through eye tracking technology. Verify that the application potential of the technology in this paper.