A Novel Configuration and Algorithm of Long Distance Gaze Estimation System for Gaze Point Error Compensation.

In gaze estimation, the principal purpose is estimating coordinate of user’s gaze point on screen coordinate system. Accuracy of gaze point is depend on user’s head movement and working distance. In this thesis, proposing a long distance gaze estimation system configuration and algorithm for gaze point error compensation. Our research is novel in the following two ways: first, the multiple near-infrared illuminators configuration to generate several mapping functions for compensating gaze point error. Second, the algorithm for choosing the best mapping function in several mapping functions. The system works as follows. Firstly, capture user’s eye images by external camera and estimate the coordinate of pupil center in eye images. There are also several near infrared illuminators in front of the user for generating corneal reflections to be reference points on user’s eyes. After that, calculate the coordinate of corneal reflection points in eye images and the vector between pupil center and each corneal reflection point as an input of a mapping function which from the eye image coordinate system to screen coordinate system. After mapping, we will obtain the user’s gaze point on the screen coordinate system. Experiment results showed the performance of our proposed is better than single near-infrared illuminator configuration and the system mean gaze error is 0.54°, standard deviation is 0.48° and processing speed is 0.78fps.