Driver fatigue detection from electroencephalogram spectrum after electrooculography artefact removal

Abstract

A driver fatigue monitoring and detection system with high accuracy could be a valuable countermeasure to decrease fatigue-related traffic accidents. This study proposes methods for drowsiness detection based on electroencephalogram (EEG) power spectrum analysis. First, a new algorithm is proposed for independent component analysis with reference (ICA-R) for electrooculography artefacts removal. Comparison is then carried out between the proposed ICA-R algorithm and an adaptive filter. Secondly, 75 EEG spectrum features are extracted from the cleaned EEG. Among all the EEG spectrum-related features, 40 key features are selected by support vector machine recursive feature elimination to improve the performance of the classifier. The validation results show that 86% of the driver's drowsiness states can be accurately detected among drivers, who participate a driving simulator study.