Driver Sleepiness Detection Based On EEG And EOG Signals

Driving fatigue is one of the main causes of traffic accidents.Most drivers have had a dozing experience during driving,and the driver's unconscious sleep can lead to serious traffic accidents with high risk.Therefore,to study how to effectively detect the driver's sleep onset period is of great scientific significance and application values for reducing the occurrence of traffic accidents.Therefore,the main purpose of this study is to find physiological indicators for the entry of sleep,and meanwhile propose corresponding detection algorithms to realize the detection of sleep onset state.Since electroencephalogram(EEG)has long been recognized as the 'gold standard' for detecting driving fatigue,and that electrooculogram(EOG)has a higher signal-to-noise ratio and is easier to acquire than EEG,we study EEG/EOG based driver sleepiness detection.Since the discovery of EEG,the study of EEG and EOG in sleep has received extensive attention and sustained development.In sleep studies,the attenuation of EEG alpha waves is considered to be the most reliable physiological marker for entering sleep.Besides,the appearance of slow eye movements also indicates the sleep onset state.Therefore,in this study,we focus on the change rules and characteristics of EEG alpha waves and slow eye movements when the driver is dozing,investigate whether they can be used as reliable physiological markers which can indicate that the driver enters sleep,and propose the corresponding detection algorithm,to achieve automatic detection of these physiological indicators.In this thesis,the main contributions are as follows:1.We design a simulated driving experiment.The purpose of the experiment is to induce the attenuation of alpha waves and the appearance of slow eye movements to examine the characteristics of these physiological indicators that can indicate the sleep onset state during driving fatigue.Secondly,in order to develop EEG/EOG-based practical wearable driving fatigue detection device,we only use a few electrodes for acquiring EEG and EOG signals.2.We find a new alpha wave attenuation-disappearance phenomenon.It refers to that alpha wave usually gradually and continuously attenuates until it completely disappears during the eye-closed period.This phenomenon occurs frequently during simulated driving,and it is a universal pattern,which is different from the intermittent pattern for alpha attenuation in sleep.This pattern indicates that the driver begins to fall asleep.3.We propose a novel method for extracting features from EEG and EOG signals using wavelet transform technique.The EEG alpha waves have significant local oscillation characteristics,and the EOG signals have local mutation characteristics.By selecting proper wavelet basis and wavelet transform methods,we achieve effective representations of local characteristics of EEG and EOG signals.4.We propose a driver sleepiness detection method based on alpha waves.Its goal is to rec-ognize the two states,relaxed wakefulness and sleep onset,by distinguishing the two alpha phenomena,alpha blocking phenomenon and alpha wave attenuation-disappearance phe-nomenon.The proposed method first uses a continuous wavelet transform based on complex Morlet mother wavelet to accurately locate the start and end points of the alpha waves.When an end point is detected,we use a long short-term memory(LSTM)network that can process time-dependent information to classify end points.Moreover,in the subject-to-subject mod-eling strategy,we adopt a generative adversarial network(CWGAN)for performing sample expansion on features extracted from two-class vertical EOG signals to improve the classifi-cation performance of the LSTM.The experimental results indicate that the proposed driver sleepiness detection method based on alpha waves can effectively detect the start and end points of alpha waves,the two alpha phenomena can be distinguished with a high precision according to the end point,and the driver's two states:relaxed wakefulness and sleep onset,are determined5.We propose a method of detecting slow eye movements for recognizing the driver's sleep onset.We turn the problem of detecting slow eye movements into an imbalanced two-class classification problem.We use various feature extraction methods such as wavelet transform to extract features from two-class data sets,and use re-sampling,SMOTE and min-max modular network methods to deal with the imbalanced two-class classification problem.