Innovative and unobtrusive system for real-time driver drowsiness monitoring

This PhD thesis investigates the use of physiological signals to monitor driver state, addressing human factors as primary causes of road accidents. While conventional driver monitoring systems, such as camera-based and behavior-monitoring approaches, have limitations in reliability and response time, physiological signal monitoring offers valuable insight into driver states, including drowsiness detection. Signal acquisition, however, poses challenges in automotive environments, where body-attached devices are impractical. The thesis initially explores wearable devices for monitoring, specifically the Empatica E4 wristband for capturing Photoplethysmography (PPG) and Electrodermal Activity (EDA) signals. A dedicated measurement campaign tested EDA alone and in combination with PPG, achieving 89% accuracy with EDA and 93% when combined with PPG, demonstrating these signals’ reliability for monitoring. However, limitations of wearables, such as the need to be worn and concerns with safe vehicle communication, led to the development of ANGELS, an innovative steering wheel embedded with sensors to acquire PPG and EDA directly from the driver’s hands. ANGELS offers a completely unobtrusive alternative, incorporating advanced algorithms to mitigate motion artifacts. These algorithms performed effectively, yielding a mean absolute error (MAE) of 1.19 for heart rate and 1.9 misdetected peaks per minute for EDA. Tested in collaboration with Maserati, ANGELS achieved high drowsiness classification accuracy (77.03%) using a Temporal Convolutional Network (TCN). Additional force sensors were integrated to assess road vibrations' effects on signal quality, and ANGELS was connected to commercial devices for comprehensive driver-state monitoring. The findings underscore the reliability of physiological signals in driver monitoring, with ANGELS emerging as a promising solution for continuous, unobtrusive safety monitoring.