Head-mounted Gaze Tracking System With Head Motion

Eye gaze tracking has a vast application prospect in the field of HCI (Human Computer Interaction), because it is more direct and natural than traditional methods like mouse and keyboards. Eye tracking systems based on computer vision estimate the position of gaze points by capturing the motion information of eyes. With the popularity of lightweight wearable devices like Google Glass, head-mounted eye tracking systems will be more applicable.This paper investigates head-mounted eye tracking system that supports head motion, including several key technologies such as fast locating of pupil center, calibration procedure and gaze computation model when head moves. The main work and achievements are concluded as follows:1. Improve the classical starburst method for pupil detection based on the environment in this paper. Mainly includes:using the setting threshold instead to realize pupil segmentation, desampling and secondary locating to find the rough pupil center, simple error estimation of ellipse fitting, and denoising by iterative screening.2. Use mapping methods like regional mapping based on cross-ratio to do the calibration between eye movements and gaze points on reference screen, proving its effectiveness through experiment. And propose a solution to tiny head motion during the calibration procedure, improving the user experience.3. Build a spatial gaze computation model with head motion. Take advantage of position changes of the IR lights in the image, and implement the transformation of gaze points from reference screen to the current screen by using cross-ratio, which is invariant in projective space. Analyze the error caused by the distance between scene camera and eye, and make amendments in the camera coordinate system based on depth.4. A set of demos are designed to show the applications of eye tracking system, including soft keyboards input on the screen, plane tracking and map controlling.