A robust driver assessment method for the brain-computer interface

Brain-computer interfaces (BCI) are a valuable proposition for the long-term vision of the automotive human-machine interfaces and for increasing the personal mobility of users with physical disabilities. In this work, we do not attempt to improve the vehicle dynamics' control through BCIs. Instead, we process the signals such interface inherently collects to assess the driver's fatigue. Non-invasive electroencephalogram (EEG) based approaches for driver assessment often rely on the independent components analysis (ICA) and measure the relative power of EEG frequency bands. In the case of wireless and mobile EEG devices, especially outside the domain of medical-grade electronics, a higher number of artifacts and lower channel count can be expected. Main priorities for such devices are ergonomics and usability. On the other hand, these devices increase acceptance of BCIs and simplify testing and data collection in automobiles. This work presents a robust two-step method for driver assessment with a consumer-grade brain-computer interface, which collects artifact-rich EEG data using a limited number of low-quality saline-pad electrodes. We demonstrate that a reliable assessment of driver state in such conditions is possible, if the independent component analysis is extended through a expert system-based assessment of reliable signal components in a specific region-of-interest on the brain surface. The method does away with the manual artifact removal. We prove that the lower sensor count, lower sensor quality and mechanical vibrations which occur during the drive, can be offset through additional signal processing. We also prove that the data collected by the BCI provides additional value to the driver assistance.