Safety-Critical Learning for Long-Tail Events: The TUM Traffic Accident Dataset

2025-08-20

AI Summary

Detecting rare but high-risk long-tail traffic accidents on highways remains a critical challenge for safety-critical autonomous driving systems. Method: This paper proposes Accid3nD, a novel rule- and learning-integrated detection framework, and introduces TUMTraf-A—the first multimodal real-world dataset explicitly designed for accident modeling. Built upon roadside LiDAR-camera cooperative perception, TUMTraf-A comprises 48,144 image frames, annotated with 294,924 2D bounding boxes and 93,012 3D bounding boxes following the OpenLABEL standard. Accid3nD enhances robustness to infrequent accident events via multi-source temporal feature fusion and anomaly pattern recognition. Contribution/Results: Extensive experiments demonstrate that Accid3nD outperforms state-of-the-art methods in accident detection, establishing a reliable, interpretable paradigm for long-tail risk perception in autonomous driving.

Technology Category

• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Multimodal Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Even though a significant amount of work has been done to increase the safety of transportation networks, accidents still occur regularly. They must be understood as an unavoidable and sporadic outcome of traffic networks. We present the TUM Traffic Accident (TUMTraf-A) dataset, a collection of real-world highway accidents. It contains ten sequences of vehicle crashes at high-speed driving with 294,924 labeled 2D and 93,012 labeled 3D boxes and track IDs within 48,144 labeled frames recorded from four roadside cameras and LiDARs at 10 Hz. The dataset contains ten object classes and is provided in the OpenLABEL format. We propose Accid3nD, an accident detection model that combines a rule-based approach with a learning-based one. Experiments and ablation studies on our dataset show the robustness of our proposed method. The dataset, model, and code are available on our project website: https://tum-traffic-dataset.github.io/tumtraf-a.

Problem

• Detecting real-world highway traffic accidents
• Addressing long-tail safety-critical events
• Combining rule-based and learning-based accident detection

Innovation

• Combines rule-based and learning-based accident detection
• Uses multi-sensor roadside cameras and LiDARs
• Provides OpenLABEL dataset with 3D bounding boxes