Problem
Research questions and friction points this paper is trying to address.
accident anticipation
anomaly onset annotation
risk estimation
self-supervised learning
collision prediction
RiskProp: Collision-Anchored Self-Supervised Risk Propagation for Early Accident Anticipation
🤖 AI Summary
This work addresses the bias in existing accident prediction methods caused by subjective and inconsistent annotations of “anomaly onset frames.” To overcome this limitation, the authors propose a self-supervised risk propagation paradigm that requires only collision-frame labels. The approach generates soft targets through future frame prediction to backpropagate risk signals and incorporates an adaptive monotonicity constraint to ensure temporally non-decreasing risk estimates. Without relying on anomaly onset annotations, the method effectively models the evolution of risk over time. Evaluated on the CAP and Nexar datasets, it achieves state-of-the-art performance, producing smoother and more discriminative risk curves that significantly enhance early warning capability and model interpretability.
📝 Abstract
Accident anticipation aims to predict impending collisions from dashcam videos and trigger early alerts. Existing methods rely on binary supervision with manually annotated "anomaly onset" frames, which are subjective and inconsistent, leading to inaccurate risk estimation. In contrast, we propose RiskProp, a novel collision-anchored self-supervised risk propagation paradigm for early accident anticipation, which removes the need for anomaly onset annotations and leverages only the reliably annotated collision frame. RiskProp models temporal risk evolution through two observation-driven losses: first, since future frames contain more definitive evidence of an impending accident, we introduce a future-frame regularization loss that uses the model's next-frame prediction as a soft target to supervise the current frame, enabling backward propagation of risk signals; second, inspired by the empirical trend of rising risk before accidents, we design an adaptive monotonic constraint to encourage a non-decreasing progression over time. Experiments on CAP and Nexar demonstrate that RiskProp achieves state-of-the-art performance and produces smoother, more discriminative risk curves, improving both early anticipation and interpretability.
Innovation
Methods, ideas, or system contributions that make the work stand out.
self-supervised learning
risk propagation
collision-anchored
monotonic constraint
early accident anticipation
🔎 Similar Papers
No similar papers found.
