A Non-Intrusive Approach to Gaze Estimation

Human eye tracking and gaze estimation is a new technology which acts as an interface between the user and the computer. Its main goal is to determine where a user is looking on the computer screen. Numerous methods have been developed to accomplish this task. However, they need expensive and proprietary hardware, or require the user to wear a device on his/her head, or place limitations on the user’s head movement. In this paper, we propose an inexpensive and non-intrusive algorithm to estimate the eye gaze on the computer screen. The proposed algorithm can quickly determine the glints and pupils of the user under normal lighting conditions and estimate the gaze of the user by a pretrained scaled conjugate gradient based neural network. Extensive experimental results demonstrate the promising robustness of the algorithm on estimating the gaze of users under normal lighting conditions and with moderate head movement, regardless of whether users wear glasses or contacts.