By applying YOLOv3 based deep learning object detection technology, the proposed visible-light pupil tracking method predicts the centers of pupils effectively.  For testing the pupil tracking performance with the inference model, the precision is up to 80%, and the recall is 83%.  Besides, the average pupil tracking errors of the proposed deep-learning based design are only 4 pixels.

By using the YOLOv3 inference model to track the centers of pupils, the precision is up to 80%, and the recall is 83%.  Besides, the average horizontal and vertical pupil tracking errors of the proposed deep-learning based design are only 4 pixels, which are much less than the pupil tracking errors of the previous ellipse fitting design at visible-light conditions.

This developed technology can be applied to human-machine interactive interfaces, eye trackers, gaze tracking, and driving safety assistance systems.