Design Of Driver-car Hybrid Control System Based On EEG Signal Identification

Driving fatigue as dangerous as drunk driving causes drivers’ physiological and psychological dysfunction, resulting in driver’s vision blurred and unresponsive, which is the main cause of extraordinarily serious accidents. Currently, there are many driver fatigue monitoring methods, which have their advantages and limitations. The EEG signal is regarded as fatigue detection’s gold standard, because it can directly and objectively reflect the brain’s activity. The study on EEG signals under different fatigue statuses has important theoretical and practical significance.This thesis study EEG signals collected from different fatigue experiments. Firstly, preprocessing these EEG signals to obtain the clean EEG signals which are in the range of 0.5～32Hz. The process of filtering includes three kinds of algorithms. (1) Using butterworth low-pass filter to filter out power frequency interference and large number of EMG signals whose frequency are higher than 32Hz; (2) Using wavelet transform to filter out EOG signals, EMG signals and background noise, which are mixed in the EEG signals, to obtain EEG signals in the range of 0.5～32Hz; (3) Using independent component analysis to process above signals and set EMG components and EOG components to zero, then use the inverse ICA transform to reconstruct raw EEG signals. The validity of the above de-noising pretreatments is verified by calculating the root mean square error and correlation coefficients.Secondly, this thesis study fatigue EEG signals from spectral components energy and sample entropy aspects separately. By comparing spectral components energy of the four rhythm waves including Delta,Theta,Alpha,Beta and calculating the ratio of them, along with observing the change of EEG topography, we concluded that the energy of slow wave increase and the energy of fast wave decrease under different fatigue statuses. In this thesis, we apply sample entropy to driver fatigue estimation and demonstrate the superiority of these algorithms by analysis of variance. Thirdly, we design the driver-car control system. If driver is in the mild fatigue status, the buzzer will alarm. If the driver is in the fatigue condition, the status will be sent to the transportation command center by GSM. Through fatigue experiment it can conclude that the accuracy of the above methods can reach 83%.In summary, the algorithm of EEG signals preprocessing proposed in this thesis can effectively eliminate artifact components. Spectral components energy algorithm and sample entropy algorithm can efficiently extract the characteristics of the fatigue EEG signal. The driver-car hybrid control system verify the the effectiveness of the above algorithms and provide a new way for the driver fatigue monitoring.