Automated Pupil Center Tacking Method and Autofocusing Algorithm for Fundus Image Acquisition

Teleophthalmology and early screening of retinal diseases become more important in aging societies, leading to higher demand for compact fundus cameras. In lightweight fundus cameras, alignment of the optical axis with the pupil and accurate focusing of the fundus image are crucial for reliable image acquisition. This study develops a system for automated pupil-position tracking and fundus focusing using a portable fundus camera, a motorized stage and a PC. Pupil tracking is implemented using a YOLO-based detector, and the pupil center is defined based on the bounding box center. Fundus autofocusing is based on evaluation of image sharpness metrics under infrared illumination and application of moving average smoothing with open-loop control. The prototype achieved 15 fps control speed and an optical axis alignment error of approximately 22 pixels, corresponding to about 121 μm. The final white-light fundus image exhibited a mere 1.7% metric degradation from the best focus condition.