Comparing Contribution of Algorithm Based Physiological Indicators for Characterisation of Driver Drowsiness

Abstract

The algorithm based physiological characteristics of driver drowsiness – ocular parameters (derived from the frontal electroencephalogram (EEG)), EEG alpha bursts and spectral power (derived from the central and occipital sites) as well as heart rate variability (HRV) were estimated from data derived during a driving simulator experiment (30 non-professional drivers). The statistical associations of these parameters with the “gold standards” of driver drowsiness were investigated using linear regression and linear mixed models. The statistical models were also examined for a number of hybrid algorithms, which combined multiple characteristics of driver drowsiness. A combination of ocular parameters showed the strongest association (R=0.48) with the applied trained observer rating (TOR) method; followed by EEG alpha bursts indicators (R=0.30) and EEG spectrum data (R=0.21). The HRV parameters showed a weak association (R=0.04) A joint model including the eye parameters and the EEG alpha bursts resulted in the highest R=0.54 to TOR. The results indicate that a hybrid automatic algorithm, based on multiple characteristics of the eye blinks and EEG patterns, but not necessarily including the HRV measures, is likely to achieve a level of accuracy in characterising driver drowsiness similar to that of a trained observer.