A robust eye gaze estimation using geometric eye features

Abstract

Gaze estimation is the process of determining the point of gaze in the space, or the visual axis of an eye. It plays an important role in representing human attention; therefore, it can be most appropriately used in Human Computer Interaction as a means of an advance computer input. Here, the focus is to develop a gaze estimation method for Human Computer Interaction using an ordinary webcam mounted on the top of the computer screen without any additional or specialized hardware. The eye center coordinates are obtained with the geometrical eye model and edge gradients. To improve the reliability, the estimates from two eye centers are combined to reduce the noise and improve the accuracy. Facial land marking is done to identify a precise reference point on the face between the nose. The ellipse fitting and RANSAC method is used to estimate the gaze coordinates and to reject the outliers. This approach can estimate the gaze coordinates with high degree of accuracy even when significant numbers of outliers are present in the data set. Several refinements such as feedback and masking, queuing and averaging are proposed to make the system more stable and useful practically. The results show that the proposed method can be successfully applied to commercial gaze tracking systems using ordinary webcams.