A Trajectory-free Crash Detection Framework with Generative Approach and Segment Map Diffusion

2025-11-17

AI Summary

Traditional crash detection methods rely heavily on vehicle trajectory estimation and tracking, resulting in poor robustness—especially under occlusion or low-sampling conditions. To address this, we propose a trajectory-agnostic, two-stage diffusion-based generative framework for crash detection. Methodologically, we introduce diffusion models to crash detection for the first time; our MapFusion architecture jointly encodes road segment graphs and temporal embeddings, while ControlNet enables traffic-context-aware controllable generation. Crucially, the model requires no trajectory inputs—only historical segment graphs—to synthesize plausible future road states; anomalies (i.e., crashes) are identified via reconstruction error. Evaluated on a real-world accident dataset, our approach achieves high detection accuracy and exhibits strong robustness across varying sampling intervals. It significantly enhances the reliability and practicality of real-time crash detection in tracker-free scenarios.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Real-time crash detection is essential for developing proactive safety management strategy and enhancing overall traffic efficiency. To address the limitations associated with trajectory acquisition and vehicle tracking, road segment maps recording the individual-level traffic dynamic data were directly served in crash detection. A novel two-stage trajectory-free crash detection framework, was present to generate the rational future road segment map and identify crashes. The first-stage diffusion-based segment map generation model, Mapfusion, conducts a noisy-to-normal process that progressively adds noise to the road segment map until the map is corrupted to pure Gaussian noise. The denoising process is guided by sequential embedding components capturing the temporal dynamics of segment map sequences. Furthermore, the generation model is designed to incorporate background context through ControlNet to enhance generation control. Crash detection is achieved by comparing the monitored segment map with the generations from diffusion model in second stage. Trained on non-crash vehicle motion data, Mapfusion successfully generates realistic road segment evolution maps based on learned motion patterns and remains robust across different sampling intervals. Experiments on real-world crashes indicate the effectiveness of the proposed two-stage method in accurately detecting crashes.

Problem

• Detects crashes without vehicle trajectory data
• Generates realistic future road segment maps
• Uses diffusion models to compare actual vs predicted maps

Innovation

• Generative diffusion model for segment map generation
• ControlNet integration for enhanced background context
• Two-stage framework comparing monitored and generated maps