Computational Modeling Of Gaze Tracking

Eyes are very important for human beings. People get information through eyes and eye movement indicates one’s psychological state at the same time. Gaze tracking technology can be used in all kinds of area, including psychological analysis, medical, military, entertainment etc.. It is so popular that many researchers have carried out a lot of researches on this technology.This paper introduces the research background, development and status of gaze tracking technology. It also reviews the methods and principles of gaze tracking. Gaze tracking becomes more convenient and accurate with video images-based gaze tracking method, which is widely used in gaze tracking research and eye trackers. This paper presents a regression-based gaze tracking method with a camera and a infrared light. We build up the gaze tracking system and analyze factors which will affect the accuracy of the system.We build up the gaze tracking system with a infrared light, a infrared filter, a CCD camera, displays, stents, head support etc.. At the same time, the algorithms of the system are optimized, including pupil extraction, calibration and gazing point calculation. Pretreatment of the eye image is needless in this system.Finally, we measure the accuracy of this gaze tracking system. Factors which will influence the accuracy of this system are classified into two parts, including external factors and internal factors. The study find that the impact of small involuntary eye movement of system accuracy is not obvious in watching state. Meanwhile, changes in light levels of indoor lighting can not have a significant impact on the accuracy of the system. In terms of hardware, improving eye image capture resolution helps to improve system accuracy. In terms of software algorithm, using sub-pixel edge detection technology can also improve the accuracy of the system. At the same time, using different regression functions in the calibration process will also affect the system accuracy. The accuracy of system based on 9-parametric regression function is better than that of system based on 6-parameter regression function. In addition, experiments are designed according to two different types of SMI eye trackers. The error of gaze tracking accuracy in X and Y direction usually ranges from 0.5 to 2 degrees. The system meets the requirement and can be used to estimate the fixation point accurately.