Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Highway Safety: Accident Detection on the A9 Test Stretch Using Roadside Sensors

Walter Zimmer, Ross Greer, Xingcheng Zhou, Rui Song, Marc Pavel, Daniel Lehmberg, Ahmed Ghita, Akshay Gopalkrishnan, Mohan Trivedi, Alois Knoll

TL;DR

The paper tackles the problem of rapid highway accident detection using roadside sensing by presenting a real-world highway accident dataset collected at the A9 Test Bed and a hybrid framework that combines rule-based trajectory cues with learning-based verification. It introduces a large-scale dataset with 294,924 2D and 93,012 3D boxes across 48,144 frames, annotated for 10 classes and designed for cooperative perception and digital-twin research, released publicly with tooling. The core contributions are the real-time accident-detection framework and the empirical demonstration that the learning-based component achieves precision ~0.8 and recall = 1.0 while operating in real time (~16 ms per frame). The work has practical impact for reducing emergency response times and enabling safer autonomous-traffic systems, with implications for perception, fusion, and predictive analytics across multimodal roadside infrastructures.

Abstract

Road traffic injuries are the leading cause of death for people aged 5-29, resulting in about 1.19 million deaths each year. To reduce these fatalities, it is essential to address human errors like speeding, drunk driving, and distractions. Additionally, faster accident detection and quicker medical response can help save lives. We propose an accident detection framework that combines a rule-based approach with a learning-based one. We introduce a dataset of real-world highway accidents featuring high-speed crash sequences. It includes 294,924 labeled 2D boxes, 93,012 labeled 3D boxes, and track IDs across 48,144 frames captured at 10 Hz using four roadside cameras and LiDAR sensors. The dataset covers ten object classes and is released in the OpenLABEL format. Our experiments and analysis demonstrate the reliability of our method.

Enhancing Highway Safety: Accident Detection on the A9 Test Stretch Using Roadside Sensors

TL;DR

The paper tackles the problem of rapid highway accident detection using roadside sensing by presenting a real-world highway accident dataset collected at the A9 Test Bed and a hybrid framework that combines rule-based trajectory cues with learning-based verification. It introduces a large-scale dataset with 294,924 2D and 93,012 3D boxes across 48,144 frames, annotated for 10 classes and designed for cooperative perception and digital-twin research, released publicly with tooling. The core contributions are the real-time accident-detection framework and the empirical demonstration that the learning-based component achieves precision ~0.8 and recall = 1.0 while operating in real time (~16 ms per frame). The work has practical impact for reducing emergency response times and enabling safer autonomous-traffic systems, with implications for perception, fusion, and predictive analytics across multimodal roadside infrastructures.

Abstract

Road traffic injuries are the leading cause of death for people aged 5-29, resulting in about 1.19 million deaths each year. To reduce these fatalities, it is essential to address human errors like speeding, drunk driving, and distractions. Additionally, faster accident detection and quicker medical response can help save lives. We propose an accident detection framework that combines a rule-based approach with a learning-based one. We introduce a dataset of real-world highway accidents featuring high-speed crash sequences. It includes 294,924 labeled 2D boxes, 93,012 labeled 3D boxes, and track IDs across 48,144 frames captured at 10 Hz using four roadside cameras and LiDAR sensors. The dataset covers ten object classes and is released in the OpenLABEL format. Our experiments and analysis demonstrate the reliability of our method.

Paper Structure

This paper contains 18 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Rule-Based Accident Detection
  4. Learning-Based Accident Detection
  5. Trajectory-Based Accident Detection
  6. Vision-Based Accident Detection
  7. Video-Based Accident Detection
  8. Accident Detection Using Synthetic Data
  9. Accident Detection Using Real Data
  10. Accident Detection Framework
  11. Rule-Based Accident Detection
  12. Learning-Based Accident Detection
  13. Evaluation
  14. Dataset
  15. Experiments
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Our dataset provides a visual representation of highway accidents with 3D box annotations, track IDs, instance masks, and vehicle trajectories. The accidents were captured using roadside cameras on the A9 Test Bed for Autonomous Driving. The scenes include crashes, overturned vehicles, and some instances of cars catching fire.
  • Figure 2: Visual comparison of the accident detection results (purple) on the test set of our accident dataset. Labeled accidents are displayed in turquoise. This test sequence shows a car accident scenario with a yellow car and a white van. We train object detection models with different input image resolutions (1920 px, 1280 px, and 960 px) and find out that an image resolution of 1280 px works best.