Design and implementation of a real-time video-oculographic gaze detection and tracking system

In this thesis, real-time video-oculographic Point of Gaze (PoG) computation methods have been investigated. A fully functional PoG system generally consists of three main components. The rst is to extract features from eye images. The second is to compute PoG based on a mathematical or geometric model using the information obtained in the rst step. The third is to design and develop a user-friendly application to operate the computer or allow fast text entry using the accuracy provided by the eye tracking system. To come up with a novel and fully functional eye tracking system at the end of the thesis, research and development for all three components of the system have been performed. In the rst step of the project, a robust and real-time algorithm has been designed to detect the boundary of the pupil in an eye image. The algorithm not only detects the pupil boundary and the center when the pupil is in clear sight, but it also succeeds even in tough occlusive cases where the pupil is partly covered by eye lashes or the eyelid. In the second step, Point of Gaze has been computed in 3D using the eye features obtained in the rst step. The major problem to solve in this step was to compensate head movements during PoG computation. An eye tracking model was developed to support this feature in the proposed system. Using the proposed solution, head movements can be tolerated by calibration and tracking for 3D PoG computation without the need for complex geometric computations. In the last step, a novel and user-friendly On Screen Keyboard (OSK) has been developed that is used together with the eye tracking system for fast textual entry. In our gaze interface, we try to enhance users' word input rates as they are typing with their glances. Experiments performed with many participants have shown that the developed OSK is easy to adapt and can signi cantly boost the text input throughput.