Abstract: For rapidly and accurately detecting drivers drowsiness, a drowsiness detection method based on brain network features is proposed in this paper. Firstly a real driving experimental environment is selected to collect the electroencephalogram (EEG) signals of drivers in real time and the wavelet packet decomposition and reconstruction are conducted on the EEG signals collected with their rhythm signals extracted. Then, the connection matrix is constructed by calculating the phase lag index of each lead and the brain network features of each rhythm are extracted. Finally through the artificial neural network regression analysis on the subjective drowsiness of driver and the features extracted, the complicated relationship between them is obtained with a correlation factor of 90.27%. The results verify the feasibility of drowsiness assessment method based on functional connectivity, opening up a novel way for brain dynamics modeling under different mental states. The method proposed uses fewer electrodes to detect drowsiness by wearing EEG equipment so is handy and economical, having a great significance to the development of driver drowsiness detection system.

Key words:  drowsy driving, EEG signal, brain network, neural network