基于多特征融合的眼睛状态检测算法研究

Research on Eye State Detection Algorithm Based on Multi-feature Fusion

疲劳驾驶检测算法研究对提升交通安全有着重要的意义。目前,已有大量关于疲劳驾驶的文献和成果。在疲劳驾驶检测算法中,眼睛开闭状态的判断起着至关重要的作用。深度级联卷积神经网络用来检测人脸和人脸特征,利用Dlib工具快速提取驾驶员人脸特征。基于眼睛特征计算眼睛宽高比,并将眼睛宽高比、传统人眼特征的人眼虹膜等用于判断眼睛开闭的参数。该文提出一种实时地融合了EAR、虹膜等多个特征的眼睛状态检测算法,可补偿传统人眼特征的像素值比较敏感的不足,也补偿了EAR在人脸倾斜、戴眼镜、光照变换、眼睛周围有光斑等情况下非常不可靠的不足。在640*480分辨率,帧率30 fps的视频上获得平均92%的检测正确率。实验结果表明融合后的算法可在光照变换、人脸倾斜、佩戴眼镜等条件下提升检测性能,鲁棒性较高。

The research about driving drowsiness detection algorithm is of great significance to improve traffic safety.Presently,there are many literatures and achievements about driving drowsiness.In driving drowsiness detection algorithm,the judgment of eye state plays an important role.A deep cascaded convolutional neural network to detect faces and face features,and Dlib tool to quickly extract drivers’face features.Eye aspect ratio(EAR)and pupil are used to detect eye stature.We propose a real-time eye state detection algorithm that integrates EAR,pupil and other features,which can compensate for the lack of relatively sensitive pixel value of traditional human eye features and compensate for the unreliability of EAR in face tilt,glasses wearing,light transformation,light spots around the eyes and other situations.The average detection accuracy is 92% in 640*480 resolution and 30 fps video.The experiment shows that the proposed algorithm can improve the detection accuracy especially in light transformation,face tilt,glasses wearing,etc.,with high robustness.