Fatigue Driving Behavior Detection In Complex Lighting Environment

Through the investigation and summary of various traffic accident data,many traffic accidents are due to fatigue driving.Therefore,real-time synchronous monitoring of the driver’s fatigue state,and early warning when the driver is in fatigue state can effectively avoid the occurrence of traffic accidents.The existing methods can be divided into three categories: methods based on physiological information,methods based on vehicle information and methods based on computer vision.Compared with the previous two methods,which are intrusive or have shortcomings such as lag and non-real-time,computer vision-based fatigue detection methods has received more and more attention.However,the existing methods have high requirements on image quality,and cannot perform fatigue detection in complex lighting environments,so it is difficult to meet the needs of practical applications.In addition,some methods have complicated calculation process,and it is difficult to meet the real-time requirements.Regarding the above issue,this paper focuses on the study of fatigue driving detection methods based on face key points and complex lighting scenes.The m ain work of the thesis includes:(1)A fatigue detection method based on facial key points and Long Short-Term Memory network is proposed.First,face images are detected through Multi-task Cascaded Convolutional Neural Network(for short MTCNN).Second,use th e Dlib library to extract face key points from the face image,and construct the fatigue feature vector.Finally,a continuous set of fatigue feature vectors are input into the LSTM neural network to obtain the final fatigue determination result.Experimen tal results show that this method performs better than other methods.(2)Aiming at the problem of low detection rate of fatigue driving in complex lighting environment,this paper proposes a fatigue detection method under complex lighting environment by using multi-scale visual Transformer.First,an image darkening algorithm based on Gaussian distribution is proposed.The algorithm uses Gaussian function to transform the data set under normal lighting environment into a data set under complex lighting environ ment.The experimental test results show that the simulation effect is better.Secondly,after obtaining the data set under complex lighting environment,MTCNN is used to extract face images,and then a continuous set of face image sequences are input into the multi-scale visual Transformer to obtain the final fatigue judgment result.The experimental results show that the method has high detection rate and robustness of fatigue driving in complex lighting environment.(3)This paper design and develop a video-based fatigue detection system.The system is mainly divided into 6 modules: user login,function selection,file selection,video splitting,face extraction and fatigue result judgment module.The system can realize the fatigue driving detection under nor mal lighting environment and complex lighting environment.