Automatic Eye Gaze Estimation with Limited Supervision

Visual perception is an essential cognitive process. Recent research has focused on understanding visual perception through the development of intelligent systems. Designing these systems require accurate estimation of gaze direction. Even with significant progress in the past few years, automatic gaze estimation still remains challenging especially due to the constrained environment based data acquisition setup. To overcome the limitation, this research attempts to develop automatic methods that advance the algorithm for gaze estimation in an unconstrained environment with limited supervision. The proposed methods arising from this thesis have potential applications in gaze based user interfaces and behavioural studies.