Different approaches and models have been used to detect driver’s stress and affective states, using physiology, facial expression, and self-reports. Including contextual information is expected to help improve the accuracy of the models.
This project focuses on vision-based extraction of driving environmental context. Thanks to recent advances in machine learning with shared real-world datasets, we believe it is now feasible to train an automated system to predict a driving-induced state of stress from a visual scene.
This work may help not only with predicting driver stress in real-time applications but also in expanding the utility of other unlabeled data sets for additional research.