Driver Sleepiness Detection Using LSTM Neural Network

Driver sleepiness has become one of the main reasons for traffic accidents. Previous studies have shown that two alpha-related phenomena - alpha blocking phenomenon and alpha wave attenuation-disappearance phenomenon - respectively represent two different sleepiness levels: the relaxed wakefulness and the sleep onset. Thus, we proposed a novel model to detect those two alpha-related phenomena based on EEG and EOG signals so as to determine sleepiness level. EOG and EEG signals inherently have temporal dependencies, and the sleepiness level transition is also a temporal process. Correspondingly, continuous wavelet transform represents physiological signals well, and LSTM is capable of handling long-term dependencies. Thus, our proposed dectection model utilized continuous wavelet transform and LSTM neural network for detecting driver sleepiness. The performance of our detection model are twofold: the recall and precision for detecting start and end points of alpha waves are generally high, and the LSTM classifier reaches a mean accuracy of 98.14%.