Pupil and blink detection algorithms for wearable eye tracking system

Abstract To obtain the pupil center of the human eye in infrared images taken by a near-eye device for gaze tracking, pupil and blink detection algorithms are proposed. The eye-detection model and the eye feature point model are trained through the dlib library machine learning, the eye area is segmented, rough positioning of the eye area is realized, the redundant image information is removed, the number of image processing calculations are removed for the subsequent pupil positioning, and the processing time is shortened. In pupil detection, the candidate pupil contours are screened based on the gray information, shape characteristics and other pupil image information to obtain the correct pupil contour information and realize precise pupil positioning. The eye feature model is used to obtain the coordinate of the feature point of the eye, the aspect ratio of the eye is obtained by conversion, and blink statistics are performed. Experiments show that the correct rate of the pupil detection method reaches 97.24%, and the correct rate of blink detection reaches 91.59%.