Driver Drowsiness Detection Using a Gyroscope Attached to a Seatbelt

In recent years, owing to the implementation of advanced safety technologies, there has been a reduction in the number of automobile accidents. However, traffic accidents caused by human error, such as driving inattentively or while drowsy, are occuring more frequently. To address this problem, automated systems that detect whether a driver is drowsy have been developed. Typically, such systems use electrode-type sensors to acquire electroencephalogram or electrocardiogram readings and perform in-cabin monitoring using cameras. However, wearing such sensors involves some discomfort for the driver, and these cameras are susceptible to noise; moreover these systems generally record personal information such as the driver's face. In this study, we propose an estimation system using heart rate variability (HRV) features acquired using noninvasive devices that cause less restriction on the driver's movement. Our proposed system obtains biometric information from gyroscopes attached to the driver's seatbelt and derives features indicating HRV. Additionally, we have trained a random forest classifier to detect drowsiness as this approach showed relatively high accuracy in prior studies. The results of an experimental evaluation confirm that the proposed system achieved an accuracy of 85.4 % in distinguishing between wakeful and drowsy states. We used the Karolinska Sleepiness Scale (KSS) to validate the results of the estimations. Our method can obtain biometric data without using invasive sensors or substantially restricting the driver's movements, and the random forest model can distinguish sleepiness.