Enhancing Highway Safety: Accident Detection on the A9 Test Stretch Using Roadside Sensors
DOI: 10.48550/arxiv.2502.00402
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the critical need for rapid accident detection to reduce traffic fatalities, which remain a leading cause of death for individuals aged 5–29. The authors highlight that existing machine learning models for autonomous driving struggle with rare, high-risk events like crashes due to the scarcity of real-world training data. To bridge this gap, the study introduces a hybrid accident detection framework and a novel dataset of real-world highway accidents captured on the A9 Test Bed for Autonomous Driving. The motivation is to enhance emergency response times and improve the robustness of perception systems by providing high-quality data for rare events that are difficult to capture naturally or simulate accurately. The proposed framework combines a rule-based approach with a learning-based model to detect accidents in real-time using roadside infrastructure sensors. The rule-based component analyzes vehicle trajectories against predefined thresholds, such as sudden stops, erratic lane changes, and time-to-collision metrics, to flag potential incidents. These candidates are then verified by a learning-based model, specifically a YOLOv8 network trained on the authors' custom dataset. This dataset comprises 48,144 frames captured by four roadside cameras and LiDAR sensors, featuring 294,924 2D and 93,012 3D box annotations across ten object classes. The dataset includes diverse crash scenarios, such as high-speed collisions, overturned vehicles, and fires, and is released in the OpenLABEL format to support cooperative perception and digital twin applications. Experimental evaluation demonstrates the effectiveness and efficiency of the hybrid approach. The learning-based model achieved a precision of 0.8 and a perfect recall of 1.0 on the test set, ensuring high accuracy in identifying accidents while minimizing false positives through multi-frame confirmation and multi-camera aggregation. The rule-based method operated at 95.05 FPS (10.41 ms per frame), while the learning-based approach processed frames in 16.13 ms using TensorRT acceleration, confirming suitability for real-time deployment. In a broader evaluation over 128 days, the system processed 12,290 video segments and identified various traffic anomalies, including standing vehicles and breakdowns, alongside confirmed accidents. The study also determined that an image resolution of 1280 px yielded optimal detection performance. The significance of this work lies in its contribution of a large-scale, real-world highway accident dataset, which addresses the data scarcity problem in training deep learning models for rare events. By integrating explicit traffic rules with deep learning, the framework offers a robust solution for automated accident detection that can shorten emergency response times. The authors conclude that this approach enhances traffic safety and provides a foundation for future research in predictive accident detection, multi-modal sensor fusion, and bridging the sim-to-real gap in autonomous driving systems.
Key finding
The proposed hybrid accident detection framework achieves a recall of 1.0 and precision of 0.8 on the test dataset, demonstrating high accuracy and real-time efficiency in identifying highway accidents using roadside sensors.
Methodology
dataset
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 1 | 2026-06-04 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: crash risk outcomes
- Methodological Resource: dataset resource