Can electroencephalography improve road safety? An EEG-based study of driver's perception of traffic light signals in a virtual environment

Abstract

Virtual traffic light environment was simulated by exposing the test subject to images of traffic lights to study his cognitive responses. Electroencephalogram (EEG) was collected from a driver in this environment, pre-processed and decomposed into EEG rhythms with wavelet transform. Epochs related to individual visual stimuli were extracted. Minimum and maximum values, standard deviation, skewness, kurtosis, and variance were used as feature vectors for classification with K-nearest neighbour (KNN) and neural network classifiers to discriminate between different traffic light colours. Classification accuracy was 84.05% and 86.94% for KNN and NN classifiers respectively, while the highest performance was observed for images of yellow lights. We conclude that drivers may perceive different traffic lights differently and that their perception results in distinct neurological activities reflected in EEG. Therefore, EEG-based detection of traffic lights may be possible that may be implemented in future automotive BCI systems expanding cars' assistive driving capabilities.