Fatigue Classification of Ocular Indicators using Support Vector Machine

Abstract

Fatigue is one of major causes of road accident. Time on task is one of factors that worsen fatigue, however, previous literatures limited their study into short simulated driving. Moreover, drivers in Indonesia frequently experience long duration driving caused by high traffic density. This study aims to determine fatigue classification based on ocular indicators in long duration driving condition. Classification of fatigue was conducted using Support Vector Machine (SVM). Twelve subjects participated in this study, and they were told to drive for three straight hours by driving simulator. Results showed improvements of blink duration, blink rate, PERCLOS, and microsleep by the end of three hours driving. Deterioration of saccadic velocity, saccadic amplitude, and pupil diameter were also occurred by the end of three hours driving. Results from Spearman rho suggest blink duration, PERCLOS, and microsleep as parameters that significantly correlated to KSS score. Radial basis function (RBF) was used as Kernel function since it has the highest accuracy compared to linear functions. SVM model indicated validity of seven ocular indicators as fatigue classification, with accuracy above 80%.