为了辅助无人值守式眼底筛查设备在不同光照环境下自动进行眼底成像, 搭建了瞳孔中心点自动定位与对准装置, 并提出了一种基于圆近似模型的算法来实现瞳孔中心点的自动定位。首先, 使用二值化、轮廓提取与填充等方法对图像进行了预处理, 以消除噪声和光斑对后续处理的影响; 接着, 使用本文提出的新型的圆近似算法来实现瞳孔中心点的精准定位; 最后, 根据定位的瞳孔中心点位置, 来控制步进电机移动平移台, 从而实现图像采集视场中心和瞳孔中心的对准。为验证有效性, 将文中算法、传统Hough圆算法及改进的Hough圆算法在定位精度、平均定位时间上进行了比较。本文算法的定位精度为93.33%, 平均定位时间为95.67 ms, 比传统Hough圆检测定位精度提高了约3.5倍, 定位时间减少了约68.86%, 比改进的Hough圆算法检测定位精度提高了约2倍, 定位时间减少了约63.11%。实验结果表明, 该系统在光照参量无法严格控制的拍摄环境下, 对瞳孔自动定位与对准的准确度和计算效率均有显著地提升。基本满足无人值守式眼底筛查设备的实时性、精确性、鲁棒性要求, 有助于无人值守式眼底筛查设备的推广。

To assist unattended fundus examination equipment in automatically performing fundus examinations under different lighting environments, an automatic pupil center location and alignment device was developed and a circle approximation algorithm was proposed to automatically locate the pupil center. First, an image was preprocessed by binarization and contour extraction, and a filling algorithm was used to eliminate the influence of noise and speckle for follow-up processing. The circle approximation algorithm proposed in this study was subsequently utilized to accurately locate the pupil center. Finally, the identified location of the pupil center was used to control the stepper motor to move the two-dimensional platform and align the image acquisition and pupil centers. To evaluate the efficiency of the proposed algorithm, the location accuracy and calculation rate were compared with those of the traditional and improved Hough circle algorithms. The precision and average consumption time of the proposed algorithm are 93.33% and 95.67 ms, respectively. Compared to that of the traditional and improved Hough circle algorithms, the precision of the proposed method improve 3.5 and 2 times, respectively, and the average consumption time is reduced by 68.86% and 63.11%, respectively. Experimental results indicate that the proposed system significantly improved the accuracy and calculation rate of automatic pupil location and alignment under conditions in which the illumination parameters cannot be strictly controlled. In summary, the proposed system is able to meet the real-time, accuracy, and robustness requirements of unattended fundus examination equipment, which has great significance for its popularization.