基于ResNet-50与Bi-LSTM的驾驶员疲劳检测算法研究

Research on Driver Fatigue Detection Algorithm Based on ResNet-50 and Bi-LSTM

疲劳驾驶是车祸的主要原因之一。针对现有面部疲劳检测模型存在对车内光照与面部遮挡敏感,系统疲劳检测准确低、泛化力弱,提出一种驾驶员疲劳检测算法,旨在对驾驶员面部多特征进行疲劳检测,提升网络泛化性与准确度。通过MTCNN定位人脸和人脸关键点,剪裁具有面部疲劳特征的图像,输入ResNet-50网络进行空间特征提取人脸局部疲劳水平,将疲劳水平串联输入Bi-LSTM网络进行时间特征提取,通过自适应特征融合算法对时间特征进行特征融合,通过分类器对整个视频进行疲劳分类。结果表明,该算法在YawDD疲劳检测准确率91.38%,在NTHU-DDD准确率达到89.36%;与主流的疲劳检测算法相比较,该方法准确度更高泛化能力更强。

Fatigue driving is one of the main causes of car accidents.In view of the fact that the existing facial fatigue detection models are sensitive to interior lighting and facial occlusion,and the system fatigue detection accuracy is low and generalization is weak,a driver fatigue detection algorithm is proposed.Fatigue detection on the multi-features of the driver’s face improves the generalization and accuracy of the network.Firstly,the method locates the face and key points of the face through MTCNN,crops the image with facial fatigue characteristics,and inputs it into the ResNet-50 network for spatial feature extraction to obtain the local fatigue level of the face,and then inputs the fatigue level in series into the Bi-LSTM network.Feature extraction,feature fusion of temporal features are fused by adaptive feature fusion algorithm.Finally,fatigue driving of the entire video is classified.The experimental results show that the algorithm has a correct rate of 91.38%in fatigue detection in YawDD,and 89.36%in NTHU-DDD.Compared with the mainstream fatigue detection algorithms,this method has higher accuracy and stronger generalization ability.