Cursor control by point-of-regard estimation for a computer with integrated webcam

Abstract

The problem of eye-gaze tracking by videooculography has been receiving extensive interest throughout the years owing to the wide range of applications associated with this technology. Nonetheless, the emergence of a new paradigm referred to as pervasive eye-gaze tracking, introduces new challenges that go beyond the typical conditions for which classical video-based eye- gaze tracking methods have been developed. In this paper, we propose to deal with the problem of point-of-regard estimation from low-quality images acquired by an integrated camera inside a notebook computer. The proposed method detects the iris region from low-resolution eye region images by its intensity values rather than the shape, ensuring that this region can also be detected at different angles of rotation and under partial occlusion by the eyelids. Following the calculation of the point- of-regard from the estimated iris center coordinates, a number of Kalman filters improve upon the noisy point-of-regard estimates to smoothen the trajectory of the mouse cursor on the monitor screen. Quantitative results obtained from a validation procedure reveal a low mean error that is within the footprint of the average on-screen icon.