Pupil center localization in NIR based on machine learning

Aiming at the problems of insufficient light, slow extraction speed of human eye image and low accuracy of pupil center location in complex environment at night, a NIR pupil center location method based on machine learning was proposed. The algorithm was optimized and improved based on Yolov5s. The C3 module of Backbone network fused with the lightweight ECA attention mechanism module and replaced the CIoU loss function with the SIoU loss function to effectively reduce redundant boxes and accelerate the convergence and regression of the prediction box. The improved Yolov5s machine learning model was used to detect the human eyes and obtain the coarse location of the pupil. Then, the acquired image is preprocessed, and the Canny operator is used to extract the pupil edge contour. The improved ellipse fitting algorithm is used to fit the ellipse, and the ellipse is corrected to accurately locate the pupil center. Experimental results show that under the condition of insufficient and uneven illumination at night, the proposed method can quickly extract the human eye image and accurately locate the pupil center, which has good robustness and real-time performance.