Driver Fatigue Detection Based On Convolutional Neural Network

In recent years,with the advancement of society,people’s demand for convenient travel has continued to increase,and the global car ownership has also increased.However,the traffic accident rate has also increased due to the increase in vehicles.In a series of traffic accidents,fatigue driving accounts for 20%.How to alert the driver in advance and reduce the accident rate in the case of driver fatigue has become the focus of research.Therefore,timely detection of driver fatigue and early warning are of great research significance.In the existing fatigue detection research,methods based on physiological parameters and vehicle behavior have problems such as inconvenience,high cost,and insufficient accuracy.However,the method based on the driver’s visual characteristics is low-cost and easier to popularize in practical applications.Based on existing application research,this thesis proposes a driver fatigue detection algorithm based on convolutional neural network.The algorithm extracts facial features to classify the driver’s state,and can balance the accuracy of fatigue detection and real-time performance.details as follows:1.In order to resolve the problem of face loss caused by the complex environment such as illumination and posture in the process of driver face detection.Use the face detection method based on multi-task cascade network to improve the speed and accuracy of the driver’s face detection.It is mainly divided into three stages.In the first stage of the network,a large number of face candidate frames are extracted,the second stage of the network is further screened on the basis of these candidate frames,and the final face frame is obtained in the third stage.Through the experimental comparison between this method and some typical face detection methods,it is found that this method has strong robustness.2.In order to resolve the problem of insufficient diversity of fatigue data samples in driver fatigue detection method based on visual features.This research screens the image data in the existing driver fatigue database,and additionally collects the video data of the driver with closed eyes,yawns and normal conditions,and intercepts the video images for classification.3.For the computational consumption of the fatigue detection model,a fatigue detection method based on the Mobile Net V3 network is proposed.The network structure uses Mobile Net V3 as the basic framework,constantly trims the network,and realizes the detection function with a small amount of calculation.At the same time in ordinary convolution operations,there is a lot of redundancy in the extracted features.Therefore,in order to ensure that the detection accuracy is further improved when the amount of network calculations is almost unchanged,a parameter compression measure based on the Ghost Module is proposed.This method differentiates the features extracted by the ordinary convolutional network,one part performs the original convolution operation,and the other performs simple linear operation.The experimental results show that the accuracy of the method on the test set is95.08%,and the fatigue detection speed reaches 38 frames per second,which can effectively detect the fatigue state of the driver in real time.