Tracking eyes becomes an important component to unleash new ways of human-machine interactions in augmented and virtual reality (AR/VR). To make the eye tracking system responsible, eye tracking systems need to operate at a real-time rate (> 30Hz). However, from our experiments, modern gaze tracking algorithms operate at most 5 Hz on mobile processors. In this talk, we present a real-time eye tracking algorithm that operates at 30 Hz on a mobile processor. Our algorithm achieves sub-0.5° gaze accuracy, while requiring only 30K parameters, which is one to two orders of magnitude smaller than state-of-the-art algorithms.
Feng, Yu; Goulding-Hotta, Nathan; Khan, Asif; Reyserhove, Hans; and Zhu, Yuhao
"A Real-Time Gaze Estimation Framework for Mobile Devices,"
Frameless: Vol. 4:
1, Article 3.
Available at: https://scholarworks.rit.edu/frameless/vol4/iss1/3