Driver Drowsiness Detection based on Variation of Skin Conductance from Wearable Device

The majority of road traffic crashes worldwide are caused by driver drowsiness. For this reason, it is necessary to recognize an incoming drowsiness status for alerting the driver as early as possible, preventing serious accidents. Variation of physiological signals can result from incipient drowsiness that the driver is unaware of, so it is worth investigating if such variation may be exploited for early drowsiness detection, in order to raise a warning. To such an aim, several studies involved mainly bulky and intrusive multimodal acquisition systems to collect driver-related information from several sensors, either worn by the individual and embedded in the car-cabin. The aim of this study is to detect the driver drowsiness through a comfortable wrist-worn device, by analysing only the Skin Conductance (SC) physiological signal. To automatically classify the drowsiness status, three ensemble algorithms have been tested, among which Random Forest results to be the best, featuring an overall accuracy of 84.1%. The obtained results prove that it is possible to classify the drowsy status of a driver from SC signals only, collected on the wrist, and motivates further research aimed at the early identification of the incipient drowsiness, for the implementation of a real-time warning system.

Automated summary

The majority of road traffic crashes are caused by driver drowsiness. This study aims to detect driver drowsiness through a comfortable wrist-worn device by analyzing only the Skin Conductance (SC) physiological signal. Three ensemble algorithms were tested, with Random Forest yielding the best results, achieving an overall accuracy of 84.1%. The results demonstrate the possibility of classifying driver drowsiness using wrist-collected SC signals, motivating further research for the implementation of real-time warning systems.