Abstract

During the dynamic driving process, classification of mental workload for drivers is a complex task due to multiple factors, including human, vehicle, road, and the environment. Many current studies focus on the impacts of a single factor, such as physiological signal, lane changing, and so on, but these results are usually unsatisfactory due to the incapacity to account for other factors. In response, this work proposes a multi-factor quantification and analysis method to classify the driver's mental workload by incorporating physiological signal, traffic flow, and environment. For physiological signals, we use sensors to gather the driver's Electrocardiogram (ECG) and Electrodermal Activity (EDA). In order to quantify the traffic flow and environment, such as traffic volume, space headway, and weather. We use detection algorithms in images from the car's front and rear driving recorders in the real-world driving experiment on the expressway. Furthermore, we can estimate the distance between surrounding vehicles and the experimental vehicle using the detected car coordinates and camera parameters. This work then applies deep learning algorithms to classify the front driving recorder images into several weather classifications. Finally, the quantified multi-factor data can be utilized to classify the driver's mental workload. Drivers, on the other hand, annotate the data using the NASA Task Load Index Scale's results (NASA-TLX). In our experiments, we compare the classification results of single-factor and multi-factor data with three machine learning algorithms: neural networks (NN), support vector machines (SVM), and random forests (RF). The results demonstrate that for traffic volume and space headway from drive recorders, the detection performance of traffic volume in a monocular camera can reach up to 87.3%. Within sight distance, the accuracy of space headway can reach up to 87.1%. The RF algorithm achieves an outstanding classification result with strong stability by incorporating ECG, EDA, traffic flow, and environmental factors, and its classification accuracy achieves 97.8%. These findings further demonstrate the efficacy of the proposed multi-factor quantification and analysis method.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call