Implementation Of Head-mounted Gaze Tracking System And Research On Desktop Gaze Tracking System

Eye tracking, also called gaze tracking, is one of the key technologies of intelligent interactive interface in the future,which has a broad application prospects in commercial field and psychology field. so it becomes frontier field with great attention in recent years.This paper introduces the background of eye tracking system, reviews the development process of eye tracking technology, summaries eye tracking methods at home and abroad. Proposing high performance of eye tracking system should include three aspects :high precision, free head movement, good interactivity.Gaze tracking methods can be categorized into head-mounted and desktop systems. This paper designs and implements a head-mounted gaze tracking system on a PC.Three performance are mainly analyzed including time,precision,interactivity. Open CV library functions are called when extracting the center of the pupil and the corneal center in eye pictures, which reduce the processing time from 80 ms to 10 ms per frame. During tracking process, a more accurate mapping relationship established,which upgrade the degree from 0.8° to 0.6°. Besides, scene detection enables users to adjust the position of heads to prevent scene print out of camera. The system is stable, which allows the head move freely within a certain range and achieves high precision and real-time interactive.Last, this paper presents desktop systems based primarily on physiological model of the eye. First camera calibration are made, then using calibrated camera’s internal and external parameters to map the pupil center and two corneal centers in image coordinate system to the three-dimensional world coordinate system. The real location of center of cornea curvature and pupil can be estimated from the mathematical model of the eye. After reconstruction of the optic axis of the eye as the line defined by the center of cornea curvature and pupil, the visual axis and the estimation of the POG(point-of-gaze) can be estimated. The preliminary result suggests it has an accuracy of 0.9° and allows free head movment in a wide range.