To increase driver’s interaction with vehicles, research interest is growing to develop new approaches that allow for detecting the driver’s intention. The extraction of features from electroencephalograph (EEG) data enables establishing a new communication channel by the use of brain signals as additional interaction channel. So far, the applicability of EEG data in the context of driving is strongly limited by the robustness and ambiguity of the chosen features. The major goal of the presented approach is the robust discrimination of EEG patterns preceding intended actions of the driver for predicting upcoming manoeuvres.

A pilot study on a test track containing elements of driver safety trainings was carried out. While driving, the manoeuvres the brain activity (64/32 EEG channels) and data from the car controller area network (CAN) was recorded.

In this paper we present the bottom layer of a classification model for upcoming driver’s movements by classifying left against right foot movement as well as left and right obstacle avoidance manoeuvres as sub-classes of the classes hand and feet movements.

This way, we present two ways in which features extracted from EEG can be used:​ (1) by exploiting event-related potentials of independent components for identifying sources of consolidated neural activity, and (2) to establish the fundamentals of an approach for an EEG-based rapid-response system that can predict the upcoming action of the driver. The latter was done by an offline classification of variances in certain frequency bands of the EEG. Feature validation was implemented by spatial and functional filtering driven by independent components of the corresponding EEG datasets.