Personalized Causal Factor Generalization for Subjective Risky Scene Understanding with Vision Transformer

In this study, we use the Vision Transformer (ViT) to pull out key features from driving video clips, aiming to understand how different participants perceive risks during driving. We then apply Counterfactual Causal Models to see how these features affect subjective driving risk among different individuals. By testing our approach with 10 participants, we found that it can personally identify which driving situations they find risky. By combining this causal analysis with the ViT's ability to understand scenes, our method showed better accuracy in detecting subjective risky situations.