Abstract:
Gaze tracking is considered to be a promising component of future user interfaces. It is faster than other pointing methods and represents our thoughts and intentions in a natural manner. Gaze tracking equipment is also becoming more a ffordable whilst improving in tracking performance and usability. However, there are inherent diffi culties in using gaze tracking as an input device. The fi rst di fficulty is the lack of accuracy due to the involuntary eye movements, limited pointing resolution of eyes and gaze tracker equipment error. The second diffi culty is the inability to di fferentiate between intentional and unintentional gaze. As eyes are primarily sensory organs, even the intentional gaze may be for perception only and not for performing any action. The di fferent techniques developed to perform actions by gaze attempt to avoid unintended actions (aka inadvertent clicks). Compensating for the inaccuracy and inadvertent clicks introduces complexity in the interaction, making it slow and tiring. Consequently, existing techniques of performing actions with gaze are only considered feasible for users with limited accessibility. This thesis investigates the potential of universal gaze controlled applications by designing and evaluating gaze control techniques that are fast, easy to use and natural, while also avoiding inadvertent clicks. The behavior of gaze is quite task-speci fic, therefore we focus on a single popular task: hypertext navigation. First, we evaluated design parameters of buttons activated by dwell. Our main findings indicate that placing the content outside the buttons, completely avoids inadvertent clicks, and that longer dwell time results in less accuracy. Second, we designed and evaluated four techniques for performing actions by gaze in the context of hypertext navigation. Their performances were compared to the mouse using quantitative and qualitative measures. The Multiple Con rm turned out to be the best gaze based alternative, being as accurate as the mouse but much slower. Finally, the TouchEye technique of performing actions by gaze was developed as a signi ficant improvement over Multiple Con rm. TouchEye associates hyperlinks on a web page with con rm buttons in the margin using color. Two variants of this technique, Static Coloring and Dynamic Coloring, were evaluated and compared with the mouse. Dynamic Coloring was found to be the best, with click times much faster than those of other gaze based techniques, and number of incorrect clicks exactly the same as that of the mouse. Some users also preferred Dynamic Coloring over the mouse in the subjective rankings. The results indicate that gaze control certainly has the potential of being used as a mainstream input method.