Advanced 1D Temporal Deep Dilated Convolutional Embedded Perceptual System for Fast Car-Driver Drowsiness Monitoring

Abstract

Recently, Advanced Driver Assistance System solutions (ADAS) are significantly contributing to the increase in driving safety levels. ADAS leverages the capability of taking active control of vehicle to prevent potentially dangerous situations. Specifically, researchers have investigated the analysis of the car driver attention level. Recent reports confirmed that there is an increasing incidence of driving crashes occurred for drowsiness or inattentiveness of the driver. In this regard, several authors suggested to monitor the car driver’s physiological status due to the well known complex correlation between the Autonomic Nervous System (ANS) and the corresponding level of attention. To carry out this study, we used an innovative bio-sensor consisting of a coupled device that includes near-infrared LED emitters and photo-detectors (Silicon PhotoMultiplier device) to assess the driver’s physiological status through the associated PhotoPlethysmGraphy (PPG) signal. We also designed an embedded time-domain hyper-filtering approach combined with a 1D Temporal Convolutional architecture with a progressive dilation setup. The proposed system performs a near real-time classification of the car driver drowsiness achieving impressive results in terms of accuracy (about 96%).