Predictive Analysis of Driver Drowsiness Progression: Multi-Level Drowsiness Classification Using Physiological Signals
Issued Date
2024-01-01
Resource Type
Scopus ID
2-s2.0-85218193073
Journal Title
APSIPA ASC 2024 - Asia Pacific Signal and Information Processing Association Annual Summit and Conference 2024
Rights Holder(s)
SCOPUS
Bibliographic Citation
APSIPA ASC 2024 - Asia Pacific Signal and Information Processing Association Annual Summit and Conference 2024 (2024)
Suggested Citation
Dachoponchai N., Wongsawat Y., Arnin J. Predictive Analysis of Driver Drowsiness Progression: Multi-Level Drowsiness Classification Using Physiological Signals. APSIPA ASC 2024 - Asia Pacific Signal and Information Processing Association Annual Summit and Conference 2024 (2024). doi:10.1109/APSIPAASC63619.2025.10848692 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/105451
Title
Predictive Analysis of Driver Drowsiness Progression: Multi-Level Drowsiness Classification Using Physiological Signals
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Drowsiness poses a significant challenge to cognitive and motor functions, compromising safety in critical tasks such as driving and increasing the risk of traffic accidents. Existing driver drowsiness detection systems inadequately address the gradual progression of drowsiness, focusing solely on binary classifications of drowsiness. This study aims to develop a neural network model that utilizes physiological signals, including EEG and ECG, to detect multiple levels of a driver's current drowsiness (Alert, Moderately Drowsy, or Extremely Drowsy). EEG and ECG data were collected from ten participants during 1-hour simulated driving experiments, supplemented by video recordings for once-per-minute drowsiness assessments through Observer Rating of Drowsiness (ORD) by two observers, which served as the ground truth. The neural network trained on 2-channel prefrontal EEG frequency-domain features, heart-rate variability (HRV) features, and driving time achieved accuracies of 92%, 77%, and 77% for Alert, Moderately Drowsy, and Extremely Drowsy, respectively. This high performance, with reliance on minimal electrodes and simple architecture, supports its feasibility for real-time applications as an early warning system for critical drowsiness in order to promote driver safety.