| In this thesis, a simple low cost eye gaze tracking system is presented which employs the cross ratios method for estimating eye gazes. A complete description of eye image collection and processing is explained. The effects of different eye image resolutions were examined using synthetically created 640x480 pixel eye images. There was some evidence that increasing eye image resolutions from 240x180 pixels to 640x480 pixels did not provide large gains in accuracy. Also examined was the effect of head position on the implemented system. It was observed that a right eye dominant user's head located 65cm from the monitor could move 10cm to the right, 17cm to the left, 5cm up and 5cm down and not greatly affect the accuracy of the eye gaze system. This movement translates to large head movements as the head can traverse across 63% of the monitor screen's width and 37% of the monitor screen's height. The limiting factor with head movement was the reliable extraction of eye features.;To facilitate the use of the system in non-cooperative environments, a simple bias compensation method is presented to eliminate a calibration routine from being preformed. The method reduced the average root mean square error of the estimated x-coordinate by 64.5 +/- 39.3 pixels (16.7 +/- 10.2mm) and the estimated y-coordinate by 53.1 +/- 14.7 pixels (13.7 +/- 3.8mm) on the monitor screen. This translates to a reduction of 28.5% +/- 17.2% for the x-coordinate and 25.7% +/- 6.6% for the y-coordinate of the eye gaze estimate. Current system limitations are also discussed that detail possible future work. |
