A Framework for Gaze Region Estimation on Web Pages

Abstract

Gaze estimation and tracking systems largely depend on head-mounted cameras which make them cumbersome to use, thus, they are restricted to laboratory set up. In recent studies, researchers have attempted to estimate gaze using mouse cursor which has produced inaccurate results because factors that influence eye-gaze region have not been considered. This thesis examined existing methods for gaze estimation then proposed a framework for estimating gaze region using enhanced cursor based trackers. The work investigated mouse activities, keyboard activities and document scrolls activities on web pages which together were used to estimate gaze region in web page documents. The main tool used this study for data collection was a web application with client-side code for recording mouse, keyboard activities and document scrolls while users navigate through web page documents. Data collected was pushed to remote web server for storage and analysis. Analysis involved two stages. Stage one involved extraction of scroll activities from mouse and keyboard activities because scroll can be achieved by either of these. Stage two involved use of Lindeman, Merenda, and Gold (LMG) metric linear regression model to estimate gaze from mouse, keyboard and scroll data. The main outcome of this work was a framework for estimating users gaze. To achieve this, we first came up with a theoretical model describing relationship between mouse activities, keyboard activities, document scrolls activities and the user gaze. An algorithm that uses LMG metric linear regression model was then used to estimate users region from collection mouse, keyboard and scroll activities data. The results shows that mouse activities only cannot be relied on to estimate user’s gaze. Both keyboard and scroll activities are play a significant role in gaze over a period of time and must therefore be considered when estimating user’s gaze region.