Online Recognition of the Mental States of Drivers with an fNIRS-Based Brain-Computer Interface Using Deep Neural Network

Abstract

Even with the recent advances in the development of autonomous vehicles, it is still important to be able to ascertain if a driver is fit to drive at any given time in case the auto-pilot fails. Continuous analysis and accurate recognition of a driver's brain activity can enable the deployment of countermeasures when a user appears unable to drive safely. This study presents a braincomputer interface (BCI) that analyses the brain activity of a user in real time and deduces the current driving mode of the car. The proposed BCI consists of a functional near-infrared spectroscopy (fNIRS)-based device used to measure activity in the prefrontal cortex of the user's brain and a deep neural network (DNN) for classification. First, a high-pass filter and the least-squares method are applied to data signals obtained through fNIRS to remove noise. Second, using a time series input vector extracted from the pre-processed signals, the DNN recognises the current mental states of drivers (manual driving versus auto-pilot) in real time. With an average classification accuracy of 61.7%, this study shows the potential of using DNN-based classification of fNIRS signals in developing BCIs for monitoring mental states.