Remote eye-gaze tracking method robust to the device rotation

Abstract

A remote eye-gaze tracking (REGT) method is presented, which compensates for the error caused by device rotation. The proposed method is based on the state-of-the-art homography normalization (HN) method. Conventional REGT methods, including the HN method, suffer from a large estimation error in the presence of device rotation. However, little effort has been made to clarify the relation between the device rotation and its subsequent error. This paper introduces two factors inducing device rotation error, the discrepancy between the optical and visual axis, called angle kappa, and the change in camera location. On the basis of these factors, an efficient method for compensating for the REGT error is proposed. While the device undergoes a 360-deg rotation, a series of erroneous points of gaze (POGs) are obtained on the screen and modeled as an ellipse, and then the center of the ellipse is exploited to estimate the rotation-invariant POG. Experimental results demonstrate that the proposed REGT method can estimate the POG accurately in spite of the rotational movement of the device.