一种基于卷积神经网络的疲劳驾驶检测方法

疲劳驾驶检测具有重要的警示作用，对检测方法的准确性和实时性均有较高要求。为此，提出了一种基于卷积神经网络的疲劳驾驶检测方法。首先，针对车内特定使用环境，对MTCNN算法进行了加速优化，在保证高准确率的同时检测速度提升高达27倍。其次，在实现人脸特征点精确定位基础上，提出了一种基于稀少特征点快速准确提取目标区域图像的ERFP（extracting images based on rare feature points）方法。再次，利用构建的眼、嘴数据集EMSD（eye and mouth state date sets）完成了眼、嘴部状态分类模型的训练。最终，利用训练得到的模型，结合相应的判定算法，实现了疲劳驾驶的检测判定。实验结果表明，该方法在实车环境下对瞌睡和哈欠行为的判定准确率均达到了96%以上，且每秒可完成约50帧图像的检测，具备良好的实时性。

Fatigue driving detection method based on CNN

Fatigue driving detection has a certain warning effect, and has a high demand for the accuracy and real-time of the detection method. This paper proposed a method of fatigue driving detection based on convolution neural network. Firstly, it accelerated and optimized the MTCNN algorithm for the specific use environment in the vehicle, and increased the detection speed up to 27 times while guaranteeing high accuracy. Secondly, based on the precise location of facial feature points, it proposed an ERFP method for fast and accurate extraction of target region image based on rare feature points. Thirdly, it trained the eye and mouth state classification model by using the constructed eye and mouth data set EMSD. Finally, it realized the fatigue driving detection and judgment by using the trained model and the corresponding judgment algorithm. The experimental results show that the recognition accuracy of the proposed method for drowsing and yawning in real vehicle environment is over 96%, and it can detect about 50 frames per second, which has good real-time performance.