A Machine Vision Approach for Estimating Motion Discomfort in Simulators and in Self-Driving

Motion discomfort in highly automated vehicles and in simulators represents a persistent problem that might be mitigated if it can be monitored. In driving simulators, motion discomfort can compromise data collection. In highly automated vehicles, motion discomfort can discourage people from riding in such vehicles, undermining the potential safety benefits. Monitoring motion sickness in real-time can help mitigate its negative consequences. This report investigates the potential of machine vision techniques in estimating motion discomfort in realtime for both, simulators and highly automated vehicles. Drivers’ video data and simulator sickness scores collected in the NADS driving simulator were analyzed. The video data were reduced to the facial action units (basic units of facial expressions) and head pose estimations. While results did not show significant correlations between motion score and facial expressions, we found a significant correlation between the drivers’ head position and motion sickness severity. One important outcome of this project was a computer-aiding tool for manual coding of videos. The tool can be used to advance research on the topic of motion sickness and also in other fields and areas that rely on video analytics like affective computing.