A New Method for Pupil Detection in Gaze-Point Estimation Systems Based on Active Contours

Abstract

Eye tracking and gaze-point estimation has increasing applications in the field of human-machine interface. Although so far a number of gaze-point estimation algorithms were investigated by researchers, video-based methods can be counted as the most important and efficient category in which eye features are obtained by processing of eye images. One of the most important factors affecting on the accuracy of gaze-point estimation is high-accurate extraction of pupil boundary. In this paper, a new method based on active contours is proposed for pupil boundary extraction. Active contours are among the conventional and useful methods for image segmentation. Generally, deformable models are curves that can evolve in order to minimize the internal and external energies in image domain. The internal energy keeps the curve smooth and differentiable, while the external energy directs the curve to the desired properties. Experimental results demonstrated suitable performance of the proposed method for a number of benchmark eye-images. Also, we used our method in an eye-tracker system for pupil segmentation. Significantly good performance of that system compared to a number of other eye-trackers can be counted as another concrete evidence for high solution quality of our method.