Comparing accuracy of two algorithms for detecting driver drowsiness — Single source (EEG) and hybrid (EEG and body movement)

Abstract

Driver fatigue is acknowledged to have similar effects on driving performance as driving under the influence of alcohol. As such, drowsiness detection systems should prove to be valuable in-vehicle safety measures. There are many algorithms that are currently being developed for this purpose, however, they often utilise a single source of data to detect drowsiness onset. It is anticipated that using hybrid data sources would increase the accuracy of such devices. The objective of this analysis was to compare the performance of a hybrid drowsiness detection algorithm with its single source counterpart. Addition of the body movement data to form a hybrid algorithm improved drowsiness detection performance over its EEG only (single source) counterpart, such that area under the ROC curve values increased from 0.764 (single source) to 0.783 (hybrid).