Technical Design Space Analysis for Unobtrusive Driver Emotion Assessment Using Multi-Domain Context

Driver emotions play a vital role in driving safety and performance. Consequently, regulating driver emotions through empathic interfaces have been investigated thoroughly. However, the prerequisite - driver emotion sensing - is a challenging endeavor: Body-worn physiological sensors are intrusive, while facial and speech recognition only capture overt emotions. In a user study (N=27), we investigate how emotions can be unobtrusively predicted by analyzing a rich set of contextual features captured by a smartphone, including road and traffic conditions, visual scene analysis, audio, weather information, and car speed. We derive a technical design space to inform practitioners and researchers about the most indicative sensing modalities, the corresponding impact on users' privacy, and the computational cost associated with processing this data. Our analysis shows that contextual emotion recognition is significantly more robust than facial recognition, leading to an overall improvement of 7% using a leave-one-participant-out cross-validation.