Robust Pupil Localization Algorithm Based on Circular Hough Transform for Extreme Pupil Occlusion

Pupil localization has been the core of eye tracking research. Accuracy of an eye tracking application relies on how center of the pupil is localized under various occlusions-such as eyelids and eyebrows. Previous works have preserved the accuracy of pupil localization when the eyelid covers the pupil up to 70% occlusion. However, the existing algorithm fails to maintain accuracy of pupil localization when the coverage of the occlusion is 70% or more. To deal with this problem, we propose a robust pupil localization algorithm based on circular Hough transform. The proposed algorithm was tested on 37 images taken from CASIA Iris dataset with various simulated eyelid occlusions. Our method achieved significant accuracy compared with previous state-of-the-art method-yielding 90.71% accuracy on 90% eyelid occlusion. The experimental results demonstrate that the proposed algorithm is promising to handle extreme eyelid occlusion in various eye tracking applications.