Classifying Driver Distraction with Textile Electrocardiograms.

Abstract

Textile sensor-based vital sign assessment plays an important role in continuous monitoring due to its unobtrusive and non-invasiveness. Textile electrocardiography (ECG) sensors allow mental wellbeing assessments in drivers during driving. In this study, we assess the effectiveness of a single-lead ECG obtained from a non-medical-grade ECG shirt for detecting driver distraction due to induced stress. Using ECG shirts, a single-lead ECG (256Hz, 12 bits) is acquired from N=10 healthy volunteers having driving licenses in three distinct driving situations (Baseline, Texting, Calling) in a controlled environment. ECG data is manually checked, and segmented into short durations (10, 30, 60 seconds). These segments are applied to a customized convolution neural network (ccNN). The proposed approach is able to classify the driver's distraction with ccNN yielding a weighted F-Score of 0.65 and an average accuracy of 67.12% on the validation set. Leave-One-Subject-Out Cross-Validation results showed weighted F-Scores ranging from 0.53 to 0.75. Thus, a single-lead, wearable textile ECG provides informative insights into a driver's mental wellbeing.