Harnessing Electrocardiography Signals for Driver State Classification Using Multi-layered Neural Networks

Abstract

Driving under conditions of cognitive overload or drowsiness poses serious safety risks and is recognized as a major cause of vehicle collisions. Thus, timely detection of the driver’s state is crucial for preventing accidents. This study proposed the utilization of electrocardiography (ECG) data in conjunction with multi-layered neural network (MNN) models to determine the driver’s state. ECG signals were obtained from 67 participants during simulated driving scenarios that induced either cognitive load or drowsiness. The study considered five driver states: drowsiness, fighting-off drowsiness, normal, medium cognitive load, and high cognitive load. Statistical analysis revealed significant changes in ECG measurements as the driver’s attentiveness levels varied from low (drowsiness) to high (cognitive overload). Multiple MNN models were developed to address individual variations in heart response and achieved classification accuracies exceeding 95%. These findings demonstrated the potential of ECG signal utilization for driver’s state detection to prevent vehicle accidents.