Driver drowsiness detection based on classification of surface electromyography features in a driving simulator.

Abstract

Driver drowsiness is a significant cause of fatal crashes every year in the world. In this research, driver's drowsiness is detected by classifying surface electromyography signal features. The tests are conducted on 13 healthy subjects in a driving simulator with a monotonous route. The surface electromyography signal from the upper arm and shoulder muscles are measured including mid deltoid, clavicular portion of the pectoralis major, and triceps and biceps long heads. Signals are separated into 30-s epochs. Five features including range, variance, relative spectral power, kurtosis, and shape factor are extracted. The Observer Rating of Drowsiness evaluates the level of drowsiness. A binormal function is fitted for each feature. For classification, six classifiers are applied. The results show that the k-nearest neighbor classifier predicts drowsiness by 90% accuracy, 82% precision, 77% sensitivity, and 92% specificity.