Research And Implementation Of Fatigue Driving Detection System Based On Deep Learning

With the continuous development of economy,society and science and technology,the personal ownership of motor vehicles in China and even in the world is increasing year by year,and the incidence of traffic accidents is also increasing rapidly.A large number of traffic accident analysis results show that fatigue driving is one of the main causes of accidents.Therefore,how to effectively detect fatigue driving and avoid traffic accidents caused by fatigue driving has important research significance and wide application value.The occurrence of fatigue driving has certain time sequence characteristics,and the relevant information will be lost from the image alone.It has important reference value for fatigue driving detection to extract time sequence information from related video clips and make full use of time sequence information and space information.Convolutional neural network is a popular method in the field of video image analysis.It is widely used because of its higher accuracy and intelligence than human.Based on the convolution neural network,the main research contents are as follows:(1)This paper proposes a fatigue driving detection method based on the fusion two-stream P3DResNet,which solves the problem of how to effectively use the spatial and temporal features in video data,but the independent two-dimensional convolution can not use the temporal information,and the direct three-dimensional convolution depends on too much hardware.The algorithm first preprocesses the video data,and then constructs a model.The model is composed of RGB flow,optical flow and fusion module.Both RGB flow and optical flow use p3 dresnet to extract the spatial and temporal features,and add fusion module after the appropriate convolution layer,and finally realize the detection of fatigue driving.Experimental results show that the fusion of two stream p3 dresnet model can improve the accuracy of fatigue driving detection without increasing the model parameters.(2)This paper proposes a fatigue driving detection method based on cross non-local attention mechanism,which solves the problem that the common convolution can only extract the local correlation information,and can’t use the relationship between pixels in a long distance,which leads to low detection accuracy.This method first uses the migration learning based on VGG-Face net model to realize face detection,and then adds a cross non local attention mechanism based on the fusion of dual stream p3 dresnet.By calculating the correlation between pixels and all other pixels,the key information in the image is found,and finally the detection of fatigue driving is realized.Experiments show that this method can improve the accuracy of fatigue driving detection.(3)This paper designs and develops a fatigue driving detection system based on the above-mentioned detection methods,which realizes the function of face detection and fatigue driving detection.It is mainly divided into three modules: data preprocessing,fatigue driving detection and evaluation.The preprocessing module includes the function of face detection.The test shows that the system can effectively detect fatigue driving and has practical application value.