Predicting Vehicle Passenger Stress Based on Sensory Measurements

Abstract

While driving autonomously, trust and acceptance are important human factors to consider. Detecting uncomfortable and stressful situations while driving could improve trust, driving quality and overall acceptance of autonomous vehicles through adaption of driving style and user interfaces. In this paper, we test a variety of sensors which could measure the stress of vehicle passengers in real-time. We propose a portable system that measures heart rate, skin conductance, sitting position, g-forces and subjective stress. Results show that correlations between self-reported, subjective stress and sensor values are significant and a neural network model can predict stress based on the measured sensor outputs. However, the subjective self-reported stress does not always match sensor evidence, which demonstrates the problem of subjectiveness and that finding one model that fits all test-subjects is a challenge.