Research On Fatigue Driving Detection Algorithm Based On Deep Convolutional Neural Network

Fatigue driving is one of the main causes of traffic accidents,so it is of great significance to study the driver fatigue detection method,timely detect whether the driver is tired,and give a warning to the driver when the driver is tired to reduce the traffic accidents caused by fatigue driving.At present,most of the driver fatigue detection is based on the visual features of the driver’s face to identify the fatigue state,but when extracting the driver fatigue features,it will be affected by factors such as light,face angle and so on,making the fatigue detection accuracy of this method relatively low.In order to solve the above problems,a fatigue driving detection method based on deep convolutional neural network was proposed.The deep convolutional neural network can directly learn the visual features of fatigue from the driver’s image.Compared with the features extracted by hand,the deep convolutional neural network has better robustness and prediction accuracy to the changes of lighting,posture and other conditions.Firstly,the optimized MTCNN algorithm is used to detect the face and locate the feature points of the face,and then the eye and mouth regions are obtained by the geometric relationship between the feature points.In combination with deep learning,a fatigue expression training model was established to extract the features of human eyes,mouth and facial expressions under the three states of non-fatigue,mild fatigue and severe fatigue of drivers.Finally,the fatigue expression recognition was realized through supervised training and learning.The main research contents are as follows:(1)Face detection and feature point calibration are realized by MTCNN algorithm after deep separable convolution optimization.In the case of a small decrease in detection accuracy,the calculation amount is reduced by 8 to 9 times.For over-pixel driver pictures,the detection speed is improved and the accuracy of face detection is improved,which lays a foundation for fatigue detection.(2)The fusion of visual features of the eyes,mouth and face was proposed as a method to identify the fatigue state of drivers.The deep convolutional neural network(DNN)model was used to extract and identify the three fatigue indexes of the eyes,mouth and face.The experimental results show that,aiming at the problem that the fatigue recognition rate is low only based on the eye features when the driver is wearing ordinary glasses and sunglasses,the fusion detection method achieves better fatigue recognition effect.(3)An improved fatigue detection network model is proposed.In convolution model of GoogLeNet layer joined the latest Drop Block regularization algorithm neural network technology,suitable for convolution layer randomly discarded strategy,depth of convolution neural network model of the network layer parameters is overmuch,had a fitting and gradient diffusion problem,to obtain the very good solution,compared with the Dropout can effectively remove the semantic information,improve the robustness of the model.(4)The softmax loss function and the center loss function were used as the supervision signal of GoogLeNet model training to train,learn and finally identify the three visual characteristics of drivers.The combination of the two loss functions not only enlarges the inter-class spacing,but also reduces the inter-class spacing,and improves the problem of poor fatigue state classification and recognition due to the over-large inter-class spacing.Experimental results show that the optimized MTCNN face detection algorithm is faster and more accurate.For the fatigue driving detection of deep convolutional neural network improved by Drop Block technology and center loss function,the fatigue recognition rate is higher and the robustness is stronger.Meanwhile,the effectiveness and feasibility of the proposed algorithm are verified.