Driver Drowsiness Detection: An Approach Based on Intelligent Brain–Computer Interfaces

Abstract

Estimating reaction times (RTs) and drowsiness states from brain signals is a notable step in creating passive brain–computer interfaces (BCIs). Prior to the deep learning era, estimating RTs and drowsiness from electroencephalogram (EEG) signals was feasible only with moderate accuracy, which led to unreliability for neuro-engineering applications. However, recent developments in machine learning algorithms, notably stationarity-based approaches and deep convolutional neural networks (CNNs), have demonstrated promising results for a class of BCI systems, e.g., motor imagery BCIs, and affective state classification. These methods have not been systematically analyzed for EEG-based driver drowsiness detection and RT prediction.