复杂背景下的人眼状态与虹膜位置检测

Human eye state and iris position detection in a complex background

为了实现人眼区域的准确定位并在此基础上提高人眼状态检测与虹膜定位的精度和抗干扰能力,本文完善了人脸数据集,设计了人眼目标检测网络,并在此基础上研究了人眼状态与虹膜位置的检测算法。采用StyleGAN的定制化人脸图像生成方法完善了人脸数据,并依据Yolov5模型设计了人眼目标检测网络。研究了基于直方图规定化与形状特征的人眼状态检测模型,给出了准确的人眼状态判断结果,并采用直方图规定化与自适应二值化结合的方法提高了虹膜检测的精度与鲁棒性。实验表明:测得人眼目标检测网络的MAP@0.75为66.4%、测试帧率为19.28,人眼状态检测算法在睁眼、眯眼、闭眼3种状态下的检测精度分别为95%、89%、93%,虹膜定位算法可以准确检测虹膜位置并量化,本文算法具有良好的准确性与通用性,提高了人眼状态的检测精度。

In this paper,to achieve accurate positioning of the human eye region and,on this basis,improve the precision and anti-interference ability of human eye state detection and iris location,the face dataset is improved,and a human eye target detection network is designed.Further,the detection algorithm of human eye state and iris position is studied to improve human eye state detection accuracy.The customized face image generation method of StyleGAN is used to improve the face data,and the human eye target detection network is designed according to the Yolov5 model.A human eye state detection model based on histogram regularization and shape features is studied,giving accurate results of human eye state judgment.The accuracy and robustness of iris detection are improved by combining histogram regularization with adaptive binarization.As measured on the experimental dataset,the MAP@0.75 of the human eye target detection network is 66.4%,and the frame rate is 19.28.The detection accuracy of the human eye state detection algorithm is 95%,89%,and 93% in the three states of open,squinting,and closed eyes,respectively.The iris location algorithm can accurately detect and quantify the iris position.The results show that the algorithm has good accuracy and universality,increasing the precision of human eye state detection.