Design And Implementation Of Fatigue Driving Detection On Face Feature

As an increasingly popular means of transportation in daily life,cars greatly facilitate people’s production and life.With the increase in the number of cars,there are more and more traffic accidents,and fatigue driving accounts for 30%-40% of the total number of accidents.According to incomplete statistics,every driver has experienced fatigue driving.During the special period of the new crown epidemic this year,the country has also promoted self-driving travel to avoid cross-infection,so it is particularly important to be vigilant when driving and pay attention to driving safety.This paper has designed a set of fatigue driving detection system,equipped with ARM core board,infrared camera and high decibel speaker and other hardware devices,which can effectively and real-time detect whether the driver is in a fatigue state and perform fatigue alarm.This system fuse the multiple facial features collected by the camera to get the discriminant result and apply it to the actual.The specific research contents are as follows:(1)Design of fatigue driving detection system.According to the analysis of needs,the design scheme and specific technical route of the fatigue driving detection system are formulated,and the system is divided into five modules: video acquisition,face detection and positioning,feature extraction,modeling recognition and sound alarm.(2)Face detection and location.The MTCNN face detection algorithm that can adapt to various environments is adopted,which is formed by cascading three-layer sub-networks of P-Net,R-Net and O-Net,and at the same time completes border regression and rough positioning of face feature points.Using the ERT algorithm of cascade regression to locate 68 feature points on the face.(3)Facial fatigue expression recognition.A CNN + LSTM dynamic fatigue expression recognition model is proposed.This model uses CNN to extract the spatial features of a single frame of facial images,and then input to the LSTM recurrent neural network to extract the timing features of each frame of images,and finally output the results through softmax classification.(4)Facial feature extraction.Six features that are more obvious when facial fatigue is extracted: facial fatigue expression frequency FFE,maximum fatigue expression duration MDFE,PERCLOS value,blink frequency BF,yawn frequency YF,nod frequency NF.The differences between PERCLOS,BF,YF and NF under fatigue and non-fatigue conditions are compared.(5)Fatigue modeling recognition.Two models are designed.Each model first obtains the basic probability distribution of each evidence through SVM training,and then fuses according to the D-S evidence combination rule to obtain the result.The reasons for the different results of the two models are analyzed and compared.(6)System test.After testing,this system can work normally for the brightness of the light in the cab,the deflection of the driver’s head,and the face occlusion.The detection of driver fatigue driving has high recognition accuracy and real-time.Help reduce the occurrence of traffic accidents.