A Study Of Calibration Techniques On Eye Gaze Tracking System With Head Fixed

Eye gaze tracking technology has got more and more attention as a new way of interaction, meanwhile, it attracts lots of researchers in different fields to participate in this project. Nowdays, eye tracking technology has been applied widely in many areas,such as sports, military and psychology. This dissertation is mainly concerned with the problem of the calibration of the eye gaze tracking system with head fixed. The author’s major contributions are outlined as follows:1. The way to simplify the calibration process has been studied. A one-point calibration method has been adopted. Firstly, the inside corner and the center of Purkinje image are extracted and compared with the ones captured at the first calibration stage,in the meantime,the transition matrix between two eye images is able to be obtained. Then the new P-CR vectors can be calculated with the transition matrix. Finally, by the new P-CR vectors and the mapping function achieved at the first calibration stage, gaze can be estimated.This method can hardly affect system’s performance and avoid conducting the cumbersome calibration process each time. So it greatly simplifies the calibration process.2. Calibration algorithm has been studied. A calibration method based on smooth pursuit has been proposed. In this method, the user’s eyes follows a moving target along a certain path, during this process, the camera captures eye imags and P-CR vectors can be obtained with the image processing algorithm, in addition, the trajectory of moving object is also recorded by computer. Here, every P-CR vector and coordinate of path at the same time can constitute a pair of data. Then outliers among all data will be eliminated with Spearman rank correlation and the rest of data will be used to computer the mapping function between P-CR vectors and coordinates of path. The experimental results demonstrates that this method needs less time and avoids affecting the system’s accuracy even though the user is interruptted in the calibration process, so it is a better way of interacting with a computer.3. The approach to reject efficaciously outliers that appear in the calibration process based on smooth pursuit has been studied. Random Sample Consensus(RANSAC) algorithm instead of Spearman rank correlation to select inliers has been presented. Experiments indicate that RANSAC algorithm is able to get rid of outliers to improve accuracy of the system.