Point-of-Regard Measurement via Iris Contour with One Eye from Single Image

Eye gaze tracking is a technique which is commonly used in human-computer interaction. By determining the eye gaze, the point-of-regard can be estimated by intersecting the gaze line and the target plane. A particular assumption is that irises are regarded as ellipses rather than circles in our approach. Besides, only one eye is required in the processes of the approach. The unique eye gaze can be computed via some image processing techniques such as edge detection and ellipse fitting and exploiting geometric properties of circles and ellipses in the 3D space. One of the key phases in our approach is the ellipse fitting process due to the noise of the image. Conventional algorithms of ellipse fitting are sensitive to noise. In this paper we have improved the randomized Hough transform to reduce the effect of noise. It is also confirmed to be robust by experimenting with real images. The estimation results of point-of-regard are illustrated and four distinct clusters can be separated robustly with both the influences of resolution and eyelids.