Detecting slow eye movement for recognizing driver's sleep onset period with EEG features.

Abstract

Slow eye movement (SEM) is reported as a reliable indicator of sleep onset period (SOP) in sleep researches, but its characteristics and functions for detecting driving fatigue have not been fully studied. Through visual observations on ten subjects' experimental data, we found that SEMs tend to occur during eye closure events (ECEs). SEMs accompanied with alpha wave's attenuation during simulated driving was observed in our study. We used box plots to analyze the distribution of durations of different ECEs to measure sleepiness level. Experimental results indicate that the ECEs with SEM have higher duration distribution, representing higher sleepiness level, especially for those accompanied by alpha wave's attenuation. This verifies that SEM can be used as a reliable indicator for recognizing driver's SOP. In light of this and considering the possible accompanying of Electroencephalograph (EEG) wave changes, we propose a new algorithm for detecting SEM, which extracted EEG power related features from occipital O2 signal to add them into features set of horizontal Electro-Oculogram (HEOG) signal. Then, maximum relevance and minimum redundancy (mRMR) method was used for feature selection and support vector machine (SVM) was used to classify the SEM class and non-SEM class. Experimental results demonstrate that using EEG power related features can improve the algorithm's accuracy by an average 1.4%. The feature P(α+θ)/β was ranked highest by mRMR among all EEG features, indicating the interactive relationship between EEG waves and SEM.