Detection and Predictive Analysis of Drowsiness Using Non-contact Doppler Sensor
DOI: 10.31803/tg-20240220092303
URL: https://hrcak.srce.hr/file/473467
archive: archived pipeline: cataloged verified
Summary
Investigates whether non-contact Doppler radar can detect and predict driver drowsiness from heart-rate-derived signals. Doppler radar measurements were validated against camera-recorded eye closures (the drowsiness ground truth) and analyzed via cross-method analysis, logistic regression, and panel logistic regression. Drowsiness detection reached p<0.001 with >95% accuracy, and panel logistic regression suitably predicted drowsiness states roughly 20-30 seconds in advance.
Key finding
Heart-rate-interval features extracted by a non-contact Doppler radar sensor support drowsiness detection at >95% accuracy and predict drowsy state onset with a 20-30s lead time, demonstrating viability of contactless physiological monitoring for fatigue.
Methodology
Empirical validation of a Doppler radar drowsiness detector. Eye-closure events on camera served as the drowsy-state ground truth; cross-tabulation, logistic regression, and panel logistic regression were applied to the radar-derived RR-interval (RRI) signal to test detection accuracy and lead time for predicting drowsiness onset.
Quality score: 5 / 5