Learning from the Unseen: Generative Data Augmentation for Geometric-Semantic Accident Anticipation

2026-04-29

AI Summary

This study addresses the challenges of traffic accident prediction in autonomous driving, which are hindered by the difficulty of modeling complex interactions and the scarcity of high-quality real-world accident data. The authors propose a dual-path framework: one path employs a structured prompt-driven video synthesis pipeline to generate high-fidelity driving scenarios that faithfully replicate real-world statistical distributions; the other introduces a graph neural network incorporating semantic cues to dynamically reason about spatiotemporal and semantic relationships among road participants. This work pioneers the integration of generative data augmentation with geometry-aware semantic graph reasoning for accident prediction and introduces a new, meticulously annotated benchmark dataset covering diverse regions, weather conditions, and traffic scenarios. Experiments demonstrate significant improvements in both prediction accuracy and early warning lead time on existing and newly introduced benchmarks, effectively alleviating data scarcity and enhancing system reliability.

Abstract

Anticipating traffic accidents is a critical yet unresolved problem for autonomous driving, hindered by the inherent complexity of modeling interactions between road users and the limited availability of diverse, large-scale datasets. To address these issues, we propose a dual-path framework. On the one hand, we employ a video synthesis pipeline that, guided by structured prompts, derives feature distributions from existing corpora and produces high-fidelity synthetic driving scenes consistent with the statistical patterns of real data. On the other hand, we design a graph neural network enriched with semantic cues, enabling dynamic reasoning over both spatial and semantic relations among participants. To validate the effectiveness of our approach, we release a new benchmark dataset containing standardized, finely annotated video sequences that cover a broad spectrum of regions, weather, and traffic conditions. Evaluations across existing datasets and our new benchmark confirm notable gains in both accuracy and anticipation lead time, highlighting the capacity of the proposed framework to mitigate current data bottlenecks and enhance the reliability of autonomous driving systems.

Problem

• accident anticipation
• autonomous driving
• data scarcity
• interaction modeling
• traffic safety

Innovation

• generative data augmentation
• graph neural network
• semantic reasoning
• accident anticipation
• synthetic driving scenes