Knowledge_PET3D_3D_Trajectory_and_Conflict_Dataset

This dataset contains all data required to reproduce the findings of the paper "Knowledge PET3D: An interpretable framework for 3D near-miss detection in thermal traffic video" (link to paper: https://doi.org/10.1016/j.eswa.2025.130821).

The dataset is organized by location and data type. It includes:
- 10 Example Videos: Ten representative thermal video examples (five per location)
- Processed Trajectory Data: The complete 3D trajectory dataset (from the full 272 hours of observation ) for both locations.
- Intermediate Analysis Outputs: All intermediate data required to replicate the analysis, including the generated cluster definitions, inferred de facto traffic rules, and calculated LOF anomaly scores
- Final Conflict Data: The final list of detected conflicts

The data was collected at two distinct urban intersections in a mid-sized European city: a complex signalized intersection (Location A) and an unsignalized right-turn slip lane (Location B) . All data was captured using AXIS Q1952-E thermal cameras (640x480, 30 fps) and is provided in georeferenced world coordinates (UTM).

The corresponding analysis code is publicly available on GitHub at: https://github.com/4rnd25/knowledge_PET3D