多姿态眼球中的瞳孔定位算法

摘要

This paper describes a pupil localization approach that is capable of processing multi-view eyeball under natural light environment. In the proposed method, the Adaboost algorithm is used to detect human face and the multi-resolution ASM method is used to locate the facial feature points. Then eye images can be obtained by four eye's feature points. After the location of the eyes, a sliding window technique is employed to analyze the eyes' gray information distribution. The local area covered by the sliding window with lowest gray intensity is assumed to be the pupil candidate area. Finally, a similar local area analysis technique is developed to adjust the position of the pupil candidate area. Experiments on multi-view eyeballs show that the proposed method can effectively localize the pupil no matter the pupil is complete or incomplete.%针对当前基于普通光源的瞳孔定位算法主要适用于眼睛开度较大、眼球较完整的问题,提出一种适用于多姿态眼球的瞳孔定位算法.该算法在采用Adaboost进行人脸检测后,利用多分辨率ASM技术进行面部关键点分析并实现眼睛定位；在眼睛定位的基础上,利用滑动窗口技术遍历整个眼部图像,并根据眼睛的灰度分布特点采用2级灰度信息分析的方法进行瞳孔定位.实验结果表明:在光线分布比较合理的情况下,采用文中算法不仅可以在眼睛开度正常、瞳孔较完整的情况下进行瞳孔定位,当瞳孔位于眼睛边缘、眼睛开度较小等瞳孔不完整的情况下也具有良好的定位效果.