Gaze Estimation by 2D Geometrical Analysis of Low Resolution Images

Abstract

In this research, an eye gaze detection method is developed to realize a HCI (Human Computer Interface) through determining the position of the Iris of eye as an input. This investigation is conducted for utilizing the ordinarily available components such as a low resolution webcam which is built-in any electronic gadgets such as laptop, smart phone, and tablet avoiding the need to buy extra hardware to interact with the machine. A complete gaze detection mechanism needs three basic steps to be taken, detection of eye socket, location of iris along with its center point estimation, and a reference setting mechanism through which point of gaze could be identified. Multiple approaches are investigated to do three mentioned steps and develop a less computationally expensive and effective gaze detection mechanism. For computational simplicity, extracted eye image is considered a 2D (Two-Dimensional) plane and different ways of setting reference points for gaze point identification are explored. Solving issues along the way, research is concluded by developing a remote gaze detection method based on the ‘Chehra’ which is an open source facial landmark localization technique. The results depicted ‘Chehra’ based method as better than the other investigated methods. The approach yielded results with 83% accuracy.