A Fine-Grained Lightweight Urban Signalized-Intersection Dataset of Dense Conflict Trajectories

Area of Science:

• Transportation Engineering
• Computer Vision
• Data Science

Background:

• Traffic participant trajectory data is crucial for urban intersection analysis and policy optimization.
• Existing drone-based datasets lack representativeness, richness, and fidelity.

Purpose of the Study:

• Introduce FLUID, a fine-grained trajectory dataset for urban signalized intersections.
• Present a lightweight, full-pipeline framework for drone-based trajectory processing.
• Enhance traffic analysis, behavior modeling, and autonomous driving research.

Main Methods:

• Collected approximately 5 hours of data across three intersection types.
• Captured over 20,000 traffic participants (TPs) in 8 categories.
• Developed a drone-based processing framework with high spatio-temporal accuracy.

Main Results:

• FLUID dataset features dense conflicts (2.8/min), with 15% of vehicles involved.
• Dataset includes trajectories, signals, maps, and raw videos.
• Validated spatio-temporal accuracy against existing platforms and ground truth.

Conclusions:

• FLUID provides a valuable resource for understanding traffic interactions and behaviors.
• The dataset supports research in human preference mining, traffic modeling, and autonomous driving.
• The processing framework offers efficient, high-fidelity trajectory data acquisition.