TOWARD DROWSINESS DETECTION USING NON-HAIR- BEARING EEG-BASED BRAIN-COMPUTER INTERFACES

Abstract

Drowsy driving is one of the major causes that lead to fatal accidents worldwide. For the past two decades, many studies have explored the feasibility and practicality of drowsiness detection using electroencephalogram (EEG)-based brain-computer interface (BCI) systems. However, on the pathway of transitioning laboratory-oriented BCI into real-world environments, one chief challenge is to obtain high-quality EEG with convenience and long-term wearing comfort. Recently, acquiring EEG from non-hair-bearing (NHB) scalp areas has been proposed as an alternative solution to avoid many of the technical limitations resulted from the interference of hair between electrodes and the skin. Furthermore, our pilot study has shown that informative drowsiness-related EEG features are accessible from the NHB areas. This study extends the previous work to quantitatively evaluate the performance of drowsiness detection using cross-session validation with widely studied machine-learning classifiers. The offline results showed no significant difference between the accuracy of drowsiness detection using the NHB EEG and the whole-scalp EEG across all subjects (p=0.31). The findings of this study demonstrate the efficacy and practicality of the NHB EEG for drowsiness detection, and could catalyze further explorations and developments of many other real-world BCI applications.

Existing System

KNN based classifier

Proposed System

The current study presented the efficacy of using EEG features that are easily accessible from the NHB areas of the scalp for assessing driving drowsiness. To explore the amount of drowsiness-related information available in the NHB EEG, we quantitatively showed that the spectral differences between alertness and drowsiness in the pre-event (lane-deviation) EEG obtained from the NHB areas are slightly weaker than that obtained from the AC areas. Nonetheless, the drowsiness related information from the NHB EEG was sufficient to provide comparable drowsiness detection accuracy to that of using the information from the whole-scalp EEG. In general, replacing the whole-scalp recording with the NHB montage is an important and practical step toward real-world BCIs, as there are considerable advantages on the efficiency of sensors, the flexibility of mechanical design, and the improvement of user experience.

Architecture