Driver Fatigue State Detection Based On Facial Features

With the gradual progress of the economy,the number of vehicles has increased year by year,and the traffic safety problems caused by this have become more and more serious,bringing potential huge threats to the property and life safety of drivers and pedestrians.According to relevant studies,traffic safety problems caused by driver fatigue account for more than 20% of the total accidents.In the context of the global epidemic of novel coronavirus pneumonia(COVID-19),it is difficult for drivers to wear masks to capture facial features,which brings new challenges to the task of fatigue driving detection.Based on the research status of fatigue driving detection at home and abroad,it can be assorted into three methods: detection methods based on the physiological characteristics of drivers,vehicle behavior characteristics,and facial characteristics.Comprehensive comparison,the fatigue detection method based on facial features has the advantages of real-time,high accuracy and low cost,and has become the mainstream in the rapid development of computer technology and AI.In addition,in the context of the COVID-19 epidemic,drivers wearing masks will lose some of their facial features.Therefore,this commentary presents a pilot fatigue state detection method based on convolutional Neural Network(CNN).The main research contents are as follows:(1)Determine the overall plan.According to the demand analysis,the overall framework of driver fatigue detection under mask occlusion is designed,including image acquisition,face detection,mask occlusion recognition,feature location and state analysis,and fatigue state detection.When the device detects that the driver is in a state of fatigue,it reminds him to stop and rest to ensure the safety of himself and his passengers.(2)Face detection algorithm under mask occlusion.First,the driver video is obtained through the camera,and the multi-task cascaded convolutional neural network(MTCNN)is used for face detection.The network structure has the advantages of low equipment requirements,easy training and high accuracy;then,the Mobile Net-V2 algorithm is used.For mask occlusion recognition,the algorithm has the advantages of easy portability,short running time and high recognition accuracy.(3)Facial feature location and state recognition.First,the eye corners are localized by the Shi-Tomasi algorithm;then,the eye and mouth regions are localized by the face border and eye corner coordinates;finally,the eye and mouth images are intercepted and corrected,and transmitted as input to the volume-based In the eye state recognition model(EOCR-Net)and the mouth state recognition model(MOCR-Net)constructed by the Convolutional Neural Network(CNN),the eyes are opened or closed and the mouth is opened or closed.During the classifier training process,data augmentation methods such as rotation,zoom-in and zoom-out,and illumination changes are used to expand the eye and mouth sample dataset to prevent low accuracy and poor stability due to too few samples.(4)Driver fatigue status analysis.After identifying the state of the eyes and mouth,the three parameters of yawning parameter,Percentage of Eyelid Closure over the Pupi I over Time(PERCLOS)and continuous eye-closing time are combined to judge the driver’s fatigue level,and experiments are carried out.The test consequences indicate that the pilot strain detection model presented in this commentary can bestow notices when the pilot is in a fatigue state,and can insure high precision while current-time detection.