Land Transportation Dataset for Urban Traffic Safety Reasoning
The Land Transportation Dataset (LTD) has been launched by researchers as a comprehensive open-source vision-language resource aimed at facilitating open-ended reasoning in urban traffic settings. Featuring 11,600 meticulously curated visual question answering (VQA) pairs sourced from various roadside cameras, LTD encompasses a wide range of road geometries and traffic situations. This dataset fills a critical void in city-scale traffic analysis, contrasting with current studies that primarily emphasize microscopic autonomous driving. The initiative seeks to enhance foundation models for interpreting diverse roadside camera data, ultimately promoting safer urban transportation. The dataset can be accessed on arXiv.
Key facts
- LTD is a large-scale open-source vision-language dataset
- Contains 11.6K high-quality VQA pairs
- Data collected from heterogeneous roadside cameras
- Spans diverse road geometries and traffic scenarios
- Addresses gap in city-scale traffic analysis
- Existing research focused on microscopic autonomous driving
- Aims to enable foundation models for roadside camera reasoning
- Published on arXiv with ID 2604.22260
Entities
Institutions
- arXiv