Research On Fatigue State Recognition Based On EEG And Visual Detection

Fatigue refers to the phenomenon that the body’s physiological function declines.In the field of driving,the number of traffic accidents caused by fatigue is huge every year.The fatigue detection method based on single information feature is susceptible to human behavior and environmental factors,the accuracy and stability of the detection system are not high.Therefore,the current research trend of fatigue detection is comprehensive detection method combining multiple information features.This thesis combined visual features and EEG characteristics to detect human fatigue,which has important theoretical significance and application value.In the research of fatigue detection based on visual features,aiming at the problem that the current human eye localization speed is slow and the human eye state recognition rate is insufficient,a human eye localization method which combined gray-scale integral projection with Region-Convolutional Neural Network is proposed.Firstly,the gray-scale integral projection is used to roughly locate the human eye,and then put the human-eye candidate region into Region-Convolutional Neural Network Completed human eye positioning.The improved convolutional neural network is used to complete the detection of the human eye characteristic points,according to the human eye feature points,realizes the human eye state detection of single frame video;extracts the fatigue feature according to the change of human eye state in time series,and trains the single model of fatigue detection based on human visual feature through SVM.The human body fatigue state recognition based on the video stream is completed.The experimental results show that compared with the commonly used algorithms such as Gabor+SVM,the proposed human eye localization algorithm effectively improves the accuracy and rate of human eye positioning.The human eye state detection using human eye feature points has a good effect.The single-mode fatigue detection model has a classification accuracy rate of 90.35%.In the research of fatigue detection based on the characteristics of electroencephalogram(EEG),the nonlinear and non-stationary characteristics of EEG signals lead to the problem of low accuracy of fatigue detection.Based on the different of EEG signals complexity in non-fatigue and fatigue states,a fatigue feature extraction method based on EEG signal energy ratio and sample entropy is proposed.The SVMclassifier is used to train the single-mode fatigue detection model based on EEG features,and the EEG-based fatigue state recognition is completed.The experimental results show that the accuracy of classification based on the combination EEG energy ratio with sample entropy is 83.06%,which is a certain improvement compared with the single energy ratio.In the fusion study of visual and EEG information,based on the complementarity of visual and EEG information in fatigue detection,the D-S evidence theory is used to combine the two single-mode models to get the final detection model.The experimental results show that the classification accuracy rate based on the final fusion fatigue detection model is 94.28%,which is greatly improved compared with the single mode model.Finally,the fatigue detection system was designed and built,the fatigue detection model trained in this thesis was used for real-time fatigue detection.The experimental results show that the designed fatigue detection system can effectively detect the human body state in different time periods.The fatigue detection in different illumination environments,combined with the visual and EEG detection system not only has better detection accuracy than the single-mode detection system,but also the system is more stable.