Abstract: For interference problems of the glasses in human eyes location, this paper fully exploits the eye structure information, combined with a fixed feature of glasses, proposes a new method for precise location of the human eyes. It obtains eyes regions by using AdaBoost detection algorithm. And on the basis of this, firstly, this paper divides the eye-eyebrow bar regions using gray projection function in the eyes rectangular regions. Secondly, it achieves binary processing of the eye-eyebrow bar regions by using adaptive binary circle fitting. Finally, it determines the exact position of the human eyes through the binary image circle fitting center. This method takes full advantage of the class structure of the eye round, and effectively avoids the impact of the glasses in the human eyes location. The experimental results show that the algorithm has good positioning results in the case of with and without glasses.

Key words: AdaBoost algorithm, grey projection function, binary circle fitting, eyes location

摘要： 针对眼镜框对眼睛定位的干扰问题，充分挖掘眼部类圆形的结构信息，并结合眼镜框的固有特征，提出了一种新的人眼精确定位方法。它在AdaBoost检测算法已经得到的眼部区域的基础上，在眼部矩形区域内利用灰度投影法分割眉眼条形区域，运用自适应二值化圆拟合对眉眼条形区域实现二值化处理，通过二值化图像圆拟合的圆心来确定人眼精确位置。该方法充分利用了眼部类圆形的结构特征，有效地避免了眼镜框对人眼定位的影响。实验结果表明，该算法在有眼镜框和无眼镜框的情况下都具有很好的定位效果。

关键词: AdaBoost算法, 灰度投影算法, 二值化圆拟合, 人眼定位