For a considerable portion of the journey, the subjects remained in a relaxed state (Level 2) but frequently switched in and out of a more drowsy state (Level 4) with occasional episodes of pronounced drowsiness (Level 6). Most of the subjects drove both sessions at, or above, the speed limit (110 kph) and their ability to respond to a critical incident when experiencing the highest levels of drowsiness must presumably have been reduced.
A second common finding was the limited restorative value of the midpoint break. Whereas drivers are typically encouraged to take regular breaks when undertaking lengthy journeys, the data from this trial suggests that the relief from fatigue may be very short lived. For some drivers, evidence of drowsiness was only found after the break.
At the time of this writing, the offline processing of the fatigue and vehicle data to enable net generation had yet to be completed. However, once this has taken place, a further session of field trials is required to assess the accuracy of the net's predictive output. This trial may also be able to assess the effectiveness of the revised eye blink event monitoring system. Recent results published by Wierwille and his colleagues ( 1994) give strong encouragement. Wierwille reported successful detection of driver fatigue in a simulator using driver eye closure, independent observers' ratings of driver facial images, and additive models incorporating eye closure and a variety of EEG-based measures.
The initial validation of the current detection system is likely to be compared once again with a direct psychophysiological measure of fatigue. However, once the net is capable of achieving a reliable output, a second, equally important type of validation will be required. This is a correlative investigation of the net's output with the drivers' self-assessments of their own arousal. If drivers are unaware of subjective feelings of fatigue, then this is likely to have a significant impact on their readiness to respond to a warning device. It is assumed that the optimal timing for a warning output will largely depend on a comparative assessment of the drivers' objectively measured, and subjectively experienced, state of fatigue.
The authors gratefully acknowledge the contributions of Chris Alford, University of the West of England, who provided expert assistance with electrophysiological recording.