Real-Time Pupil Localization Algorithm for Blurred Images Based on Double Constraints.

Accurate pupil localization is crucial for the eye-tracking technology used in monitoring driver fatigue. However, factors such as poor road conditions may result in blurred eye images being captured by eye-tracking devices, affecting the accuracy of pupil localization. To address the above problems, we propose a real-time pupil localization algorithm for blurred images based on double constraints. The algorithm is divided into three stages: extracting the rough pupil area based on grayscale constraints, refining the pupil region based on geometric constraints, and determining the pupil center according to geometric moments. First, the rough pupil area is adaptively extracted from the input image based on grayscale constraints. Then, the designed pupil shape index is used to refine the pupil area based on geometric constraints. Finally, the geometric moments are calculated to quickly locate the pupil center. The experimental results demonstrate that the algorithm exhibits superior localization performance in both blurred and clear images, with a localization error within 6 pixels, an accuracy exceeding 97%, and real-time performance of up to 85 fps. The proposed algorithm provides an efficient and precise solution for pupil localization, demonstrating practical applicability in the monitoring of real-world driver fatigue.