Research on Pupil Center Localization Method Based on Fitting Elliptical Contours with Arc-support Line Segments

In head-mounted eye tracking system, detecting pupil center efficiently and precisely is the prime assignment. A pupil center localization method based on fitting elliptical contours with arc-support line segments is proposed to solve the problems of low detection rates of existing pupil detection algorithms in real-world scenarios caused by factors such as lighting variations, occlusions, and off-axis effects. First, use the Haar feature detector to locate the approximate position of the pupil in the eye diagram. Secondly, combine the Canny algorithm with the arc-support segment detection algorithm to extract the elliptical arc that may belong to the pupil contour, and eliminate a large number of redundant points. Then according to the distance, polarity and direction between the arc segments, the arc segments are combined into different groups, and ellipse fitting is performed by the Least Squares Method to generate the initial ellipse set. Based on the physiological characteristics of the pupil, filter out the ellipse with positive polarity, and determine the support inlier of the ellipse. Finally, the support inlier is fitted for a second time to obtain more accurate ellipse parameters, and the best ellipse is selected as the pupil contour according to the ellipse confidence level. Through comparative experiments on the public eye image dataset (LPW), the average detection rate has increased by 12.5% compared with the highest of the existing six algorithms, Even if the acquiring of eye image is affected by some shadows and light spots, the proposed method still can accurately identify the pupil, can be applied to various head-mounted eye tracking systems to provide more accurate pupil position information for line-of-sight estimation.