摘要
针对驾驶环境下光照变化和头部倾斜旋转对人眼检测的影响,提出一种基于Zernike矩特征的人眼定位与状态识别方法。首先分析了图像Zernike矩幅值的旋转不变性及其相位角度信息在人眼定位与状态识别方面的适用性。其次为了提高人眼定位的精度,采用Relief算法分析筛选出更为适合描述人眼形状信息的Zernike矩组成特征向量,并结合相对误差距离测度的向量相似性匹配方法来实现人眼区域的检测与定位。随后提出一种基于Zernike矩相位信息的快速旋转角度估计算法,用以补偿修正人眼图像的旋转角度,消除头部倾斜旋转对人眼状态识别的影响。最后对角度修正后的人眼图像采用投影法计算得到最终的人眼开合度状态。实验结果表明,该方法能在一定程度上提高人眼定位的精度和鲁棒性,并且所提出的旋转角度修正算法可使得人眼状态识别的准确率大幅提高。
A method for eye location and status recognition is proposed to deal with the problems caused by illumi-nation changes and head rotation in driving environment.Firstly, according to the rotational invariant property of Zernike moments＆#39;magnitude and the angle information in their phases , the applicability of Zernike moments for eye location and status recognition is analyzed.Secondly, in order to improve the accuracy of eye location, the Relief algorithm is used for selecting the Zernike moments, which are more appropriate for describing the shape of eye im-ages, to compose the feature vector, and then a vector similarity matching method based on relative error distance measurement is described to achieve the eye detection and location .Thirdly, according to the phase information of Zernike moments, a rapid rotation angle estimation algorithm is proposed for correcting the rotation angle of eye im-age and thus eliminating the effect of head rotation on the eye state recognition .At last, the projection method is used on the angle corrected eye image to obtain the final result of eye opening degree .The experimental results show that this method can improve the precision and robustness of eye location to a certain extent , and the proposed rotation angle correction algorithm can significantly improve the accuracy of eye state recognition .
出处
《电子测量与仪器学报》
CSCD
北大核心
2015年第3期390-398,共9页
Journal of Electronic Measurement and Instrumentation
基金
国家自然科学基金(13AZD088)
中国铁路总公司科技研究开发计划(2012x0007)资助项目
