Research On Head Movement Problem In Gaze Tracking System

The gaze tracking has been a hot research direction which has a wide range of applications in psychology, commerce, human-computer interaction, military and other fields. With the development of gaze tracking technology, the improvement of the system accuracy and allowing users the wide range of head movement have been the keys of the research in recent years.Firstly, the paper describes the background, development and application of gaze tracking technology, and then introduces the key methods in gaze tracking technology. In addition, the solutions of head movement problem in head-mounted and desktop gaze tracking systems are summarized.In desktop gaze tracking system, it is difficult to extract pupil characteristic parameter exactly when the range of head movement becomes wider. The paper gives a detail introduction of the pick-up algorithm of pupil characteristic parameter based on bright and dark pupil technology, which is suitable in the case that user's head moves in a range of 20cm front and back. In this algorithm the method of removing false edge points solves the problem that pupil is overlapped. Using the corresponding relationship between two adjacent frames settles the problem that it is not easy to locate pupil when users wears glasses. Meanwhile, a stable facula segment algorithm is presented. The result of experiment shows that the rate of pick-up accuracy is 90 percents and makes the system high stability.In head-mounted gaze tracking system, the range of user's head movement is small. In order to improve the range of head movement, the paper presents a head-mounted gaze tracking system based on three-dimensional model of the eyeball. In the system, in order to improve the range of head movement, a 6D posture sensor is used to follow the tracks of head movement. Firstly, the whole calibrating process of the system is introduced and we can get real-time external parameters of camera in World Coordinate System by the positional relation between demarcated camera and sensor. Secondly, a simplified three-dimensional model of eyeball is built in which we can calculate the optic axis according to the ootics and geometric principle and then the optic axis is transformed into visual axis by single-point calibration so that the fixation point can be estimated. Finally, the effect of the error of sensor and pupil characteristic parameter on system accuracy is analyzed with anolog data and the feasibility of this system is verified.