Robust Feature Extraction for Non-contact Gaze Tracking with Eyeglasses

As a new approach for human-computer interaction, the non-contact gaze tracking technology is facing an increasing demand for its practical application in daily interaction. Especially in respect of users with eyeglasses, the gaze tracking that is adaptive to interference of eyeglasses is attracting more attentions. In this paper, a robust algorithm of feature extraction for noncontact gaze tracking with eyeglasses is presented. The algorithm uses a pyramidal multi-scale screening strategy to locate the pupil at first, and then utilizes knowledge of essential characteristics to distinguish the real cornealreflections from interference of eyeglasses. On this basis, the feature extraction of gaze is more robust and accurate. An experiment has been taken to evaluate the algorithm's performance to the interferences in some different situations. Comparing to a former algorithm that has been used, the success rate of the proposed algorithm has risen significantly. Therefore, the algorithm exhibits great potential for practical applications.