Open Collections

A non-contact video-oculograph for tracking gaze in a human computer interface

University of British Columbia

Video-based eye tracking devices that can detect where a person is looking without requiring the user to wear anything can be effective components of a human computer interface. However, issues such as speed, accuracy, cost and case of use have so far limited the widespread, practical application of existing devices. No pre-existing device that does not require user contact is able to accurately track the eye (and hence where a person is looking) in real time in the presence of large head movements. This thesis attempts to overcome these limitations by presenting a novel design for a video-oculograph. It is a non-contact (i.e., it does not require the user to wear anything) human computer interface device, and uses two cameras to maintain accurate, real-time tracking of a person's eye in the presence of significant head motion. Image analysis techniques are used to obtain very accurate locations of the pupil and corneal reflection. All the computations are performed in software and the device only requires simple, compact optics and electronics attached to the user's computer via a serial port and IEEE 1394 interface, making the device very cost effective. An implementation of the design to track a user's fixations on a computer monitor is also evaluated in this thesis. Two methods of estimating the user's point of gaze on a computer monitor were evaluated. Using functional approximation, the gaze was estimated to within 5.2% of the monitor's width and 9.6% of the monitor's height. Using a direct, analytical approach, the gaze was estimated to within 22.0% horizontally and 27.8% vertically. Evidence was found to support further investigation into effective calibration procedures that would significantly improve the ability of the system to estimate the user's point of gaze. The implemented system - called GTD (Gaze Tracking Device) - is capable of reliably tracking the user's eye in real-time (nine frames per second) in the presence of natural head movements as fast as 100°/s horizontally and 77°/s vertically. It is able to track the location of the eye to within 0.758 pixels horizontally and 0.492 pixels vertically, and is robust to changes in eye colour and shape, ambient lighting and the use of eyeglasses.

Text

2012-04-19

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/42119

Electrical and Computer Engineering

University of British Columbia

Graduate