Feasibility Study on Contactless Feature Analysis for Early Drowsiness Detection in Driving Scenarios

Drowsy driving significantly impairs drivers' attention and reaction times, increasing the risk of accidents. Developing effective prevention technologies is therefore a critical task. Previous studies have highlighted several limitations: (1) Most drowsiness-detection methods rely solely on facial features such as eye blinking or yawning, limiting their ability to detect different drowsiness levels. (2) Sensor-based methods utilizing wearable devices may interfere with driving activities. (3) Binary classification of drowsiness levels is insufficient for accident prevention, as it fails to detect early signs of drowsiness. This study proposes a novel drowsiness-detection method that classifies drowsiness into three levels (alert, low vigilant, drowsy) using a non-contact, camera-based approach that integrates physiological signals and visible facial features. Conducted as a feasibility study, it evaluates the potential applicability of this method in driving situations. To evaluate generalizability, experiments were conducted with seen-subject and unseen-subject conditions, achieving accuracies of 96.7% and 75.7%, respectively. This approach provides a more comprehensive and practical solution to drowsiness detection, contributing to safer driving environments.