Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zou, Yiyang, Zhao, Tianhao, Xiao, Peilun, Jin, Hongyu, Qi, Longyu, Li, Yuxuan, Liang, Liyin, Qian, Yifeng, Lai, Chunbo, Lin, Yutian, Li, Zhihui, Wu, Yu
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.27165
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915896918802432
author Zou, Yiyang
Zhao, Tianhao
Xiao, Peilun
Jin, Hongyu
Qi, Longyu
Li, Yuxuan
Liang, Liyin
Qian, Yifeng
Lai, Chunbo
Lin, Yutian
Li, Zhihui
Wu, Yu
author_facet Zou, Yiyang
Zhao, Tianhao
Xiao, Peilun
Jin, Hongyu
Qi, Longyu
Li, Yuxuan
Liang, Liyin
Qian, Yifeng
Lai, Chunbo
Lin, Yutian
Li, Zhihui
Wu, Yu
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_27165
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RiskProp: Collision-Anchored Self-Supervised Risk Propagation for Early Accident Anticipation
Zou, Yiyang
Zhao, Tianhao
Xiao, Peilun
Jin, Hongyu
Qi, Longyu
Li, Yuxuan
Liang, Liyin
Qian, Yifeng
Lai, Chunbo
Lin, Yutian
Li, Zhihui
Wu, Yu
Computer Vision and Pattern Recognition
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.
title RiskProp: Collision-Anchored Self-Supervised Risk Propagation for Early Accident Anticipation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.27165