Research On Eye Gaze Estimation Technique Base On 2D Model

With the development of video and image processing technology, camera-based gaze estimation technology has become an active field in Human Computer Interaction (HCI). Although traditional gaze estimation technology base on the Pupil-Cornea Reflection Technique (PCRT) has good user experience and acceptable accuracy, it is not wildly applied for that it needs extra light sources, user should make his head as still as possible when using the system and estimation accuracy will drop rapidly according to the head movements.To avoid using extra light sources and improve the tolerance to head movements, this dissertation adjusts PCRT to Pupil-Corner Technique (PCT) which takes inner eye corner as the static reference point and pupil-corner vector as the input of eye gaze estimation. However, the accuracy of PCT drops rapidly to head movements. To improve the accuracy of x-coordinate of the gaze point, a new gaze estimation technique called Geometric Fitting Technique (GFT) is proposed in this dissertation according to the geometric relationship of eyeballs and screen. The proposed 2D geometric model simulates the movements of sight when user looks through the screen. Furthermore, a fitting method computed after the calibration procedure is used to improve the accuracy of the geometric model.Experiments are implemented in several different conditions to compare GFT with PCT. User is required to look at several predefined points on the screen, each gaze state is captured by a camera and feature points on images are extracted manually. The results show that GFT has a better accuracy than PCT, especially when there are head movements, and a better tolerance to head movements. GFT is compared with PCT in different aspects and the reasons causing the errors of GFT are discussed at the end of the dissertation. At last, future research base on geometric model is suggested by the guide concluded from the dissertation to promote accuracy.