Detection and Predictive Analysis of Drowsiness Using Non-contact Doppler Sensor

The demand for continuous monitoring of vital signs is steadily increasing. Drowsiness occurs when individuals are tired or engaged in repetitive tasks, and driving or working in this state can lead to serious accidents. Various methods for detecting heartbeats based on Doppler sensors have been proposed due to their non-contact nature. Previous research involved developing Doppler radar sensors and verifying their reliability, with over 95 % accuracy compared to traditional ECG devices for heart rate measurement. This study proposes a method utilizing existing Doppler radar sensors to detect and predict drowsiness. To verify the test subjects' drowsy states, their faces were recorded with a camera, and the moments when their eyes were closed were validated as instances of drowsiness. Analytical methods were employed, including cross-method analysis, logistic regression analysis, and panel logistic regression analysis. The analysis revealed a p-value for drowsiness detection lower than 0.001, indicating statistical significance. Moreover, the significance of drowsiness states and stages was confirmed with an accuracy of over 95 %. Particularly, panel logistic regression analysis suggested its suitability as an indicator for predicting drowsiness states. In terms of predicting drowsiness stages and actual drowsiness states, it was observed that a time error of approximately 20-30 seconds exists. The study aimed to detect drowsiness and predict drowsiness based on data acquired through a non-contact Doppler radar sensor.