A Novel Simple 2D Model of Eye Gaze Estimation

With the development of image process technology, camera-based eye gaze estimation methods make a possible nonintrusive way of human computer interaction (HCI). Most of them base on Pupil-Cornea Reflection Technique (PCRT) which uses extra light sources and estimates gaze point on screen by a polynomial mapping function. In this paper, the input vector of classical PCRT is adjusted to a new one which can be precisely extracted in the system using only one camera. The adjusted PCRT is called Pupil-Corner Technique (PCT). Meanwhile, a novel simple 2D Geometric Model (GM) is proposed. It bases on the geometric relationship of eyeballs and screen and simulates the movement of eyeballs when user looks through the screen with his head still. Furthermore, a fitting method is used to promote the accuracy of GM. At last, the combined technique called Geometric Fitting Technique (GFT) is compared with PCT in experiments and the results show that GFT has an acceptable accuracy which can be applied in HCI applications and future geometric model is suggested by the guide concluded from the paper to promote accuracy.