An effective and rapid localization algorithm of pupil center based on Starburst model

Starburst algorithm that combines feature-based and model-based approaches, can achieve a good tradeoff between run-time performance and accuracy for dark-pupil infrared illumination. On this basis, eye images are preprocessed using AdaBoost classifier to extract region of interest which contains pupil in this paper. Thus, we can ensure that the start point of Starburst model is located within pupil region, reducing the number of iterations of the algorithm. Meanwhile, reflection points are detected and filled effectively, eliminating interference points. At last, high density connected region clustering is used for the non-continuous edge feature points, and the ellipse fitting parameters are optimized to improve the accuracy of the pupil center localization. Finally, the effectiveness and rapidity of the algorithm proposed in this paper is validated through experiments.