The growing number of fatigue related accidents in recent years has become a serious concern. Accidents caused by fatigue, or more precisely impaired alertness, in transportation and in mining operations involving heavy equipment can lead to substantial damage and loss of life. Preventing such fatigue related accidents is universally desirable, but requires techniques for continuously estimating and predicting the operator's alertness state. PERCLOS (percentage of eye closure) was introduced as

more »

... an alertness measure. Some years later, it was claimed to be superior in fatigue detection to any other measure, including the general Eye-Tracking Signal (ETS) and even EEG recordings. This study will show that this is not the case. To put things into the prospective a fair and objective comparison between PERCLOS, the general ETS and EEG/EOG has to be established. To achieve this purpose, a protocol was established to investigate the fatigue detection capabilities of PERCLOS, ETS, and EEG/EOG in a simple two class discrimination analysis using an ensemble of Learning Vector Quantization (LVQ) networks as a classification tool. Karolinska Sleepiness Scale (KSS) and Variation of Lane Deviation (VLD) were used in order to obtain independent class labels, whereas KSS provided subjective alertness labels while VLD provided objective alertness labels. The general ETS and the fused EEG/EOG measures contain substantially greater amounts of fatigue information than the PERCLOS measures alone. These conclusions were found to be valid for all three commercially available infrared video camera systems that were utilized in the study. The data utilized in the discrimination analysis were obtained from 16 young volunteers who participated in overnight experiments in the real car driving simulation lab at the University of Schmalkalden.