A High Detection Rate Pupil Detection Algorithm Based on Contour Circularity Evaluation

A pupil detection algorithm that exhibits a high detection rate is presented. This work builds on two previously proposed algorithms and includes a procedure that provides an initial global adaptive threshold that is further adjusted by a feedback loop. The processing steps involve a blurring phase, two morphological operations, contour determination using Laplacian, a convex-hull enclosure as well as a novel contour processing solution that ensures that the contour of maximum circularity is always chosen as the pupil. Finally, the Least Square Fit to Ellipse (LSFE) procedure is used to accurately estimate the position of the pupil center. The proposed algorithm achieves a 95% average detection rate at five pixels for two representative databases.