Abstract
We have studied a practical drowsiness detection system for professional drivers of heavy-duty trucks to prevent traffic accidents. We measured the electrocardiographic data and visual behaviors, and also the vehicle behaviors in actual sleep-inducing driving situations. Five persons participated in experiment for two days. We extracted amounts of time evaluated as sleepy state and an alert state. The systems performance for separating these two states was evaluated. As a result, drowsiness estimated by these three indices was significantly higher in the sleepy state than in the alert state. We also found that accuracy of drowsiness judgment improved when combining the indices.
