Research On Driver Fatigue State Detection Method Based On Deep Learning

In recent years,scientific and technological progress has changed people's life.As a country has grown from rich to strong,the quality of life of its citizens has also been greatly improved.However,the frequency of traffic accidents in our country is relatively high,and it hurts life and money and is not conducive to the stable development of society.Traffic accidents caused by fatigue driving take up a great proportion in the total number of traffic accidents.On the basis of statistics from relevant departments,about half of traffic accidents are caused by fatigue driving in China,causing considerable economic losses to families and society.Automobile driver fatigue detection is an important part of driving safety research.Real-time and accurate detection and judgment of the driver's fatigue state is a hot spot for researchers at home and abroad.The real-time and accuracy of detection methods are indicators for judging whether the detection method is feasible.Past and present burnout driving detection methods usually include the driver's physiological signals,operating behavior,physiological response characteristics,and vehicle status.This contact detection method will affect the driving condition of the driver,and has problems such as poor comfort,low accuracy,high price and poor promotion.In recent years,deep learning has become the focus of attention,and how to use convolutional neural networks to extract image features has become a hot topic at home and abroad.In machine learning,convolutional neural networks are feed-forward neural networks,and this algorithm has been successfully applied in the field of image recognition.Compared with traditional detection methods,it has the characteristics of local connection and weight sharing,and has better robustness,real-time performance and accuracy.This paper proposes a driver fatigue state detection method based on deep learning.The method improves on the Multi-task Cascaded Convolutional Network(MTCNN)to achieve fast and accurate face recognition and Facial feature point positioning.According to the geometric relationship between the feature points,the eye and mouth features are extracted and passed to the Eye and Mouth Fatigue Classification Network(EMFC-Net)for classification training.EMFC-Net uses multiple parameterized layers to directly learn the residual performance between input and output.Instead of typical CNN networks(such as Alexnet/VGG,etc.),it uses parameterized layers to directly try to map between input and output,and it tries to directly learn the residual representation between input and output.Experiments have proved that using the parameterized layer to directly learn the residuals between the input and output has a faster convergence rate and a higher classification effect than the mapping between the input and output of the direct learning.Then the two kinds of fatigue information are combined with the fatigue judgment rule to identify driving Whether the staff is in a state of fatigue.The experimental confirms that the proposed fatigue drive detection method has batter accuracy than the previous face location detection method and feature point location detection method,and can better extract the fatigue features of the eyes and mouth,and finally pass deep convolution the neural network performs extraction.Combining multi-task features with Perclos' s law and OMR's law to determine the driver's fatigue state,and perform fatigue warning,a large number of experiments confirm that using this method can improve the accuracy,enhance the real time and reduce the calculation rate.