Trajectory and Analysis Data for a Before–After Study of Left-Turn Vehicle–Pedestrian Interactions Following Median Installation

Description

This dataset contains de-identified derived vehicle and pedestrian trajectories, Python analysis scripts, reproducibility checks, and key numerical outputs supporting a before–after trajectory-based study of left-turn vehicle behavior and vehicle–pedestrian interactions at an urban intersection in Tehran, Iran. Vehicle trajectories were reconstructed from manually tracked video observations using Kinovea and analyzed using a Python-based workflow. The public package includes materials supporting the primary full-maneuver vehicle analysis, DCA/TCA calculations, TTC sensitivity analyses, multiple-comparison sensitivity, and source-video-cluster sensitivity. Raw roadway videos and field photographs are not included because they contain real-world imagery and are subject to privacy considerations. The dataset contains 346 reconstructed vehicle maneuvers, of which 344 vehicle–pedestrian interaction events had positive common temporal overlap for interaction analysis. The data should not be interpreted as demonstrating a causal improvement in pedestrian safety.

Steps to reproduce

Download and extract the dataset package while preserving the folder structure. Install Python and the required packages listed in requirements.txt. Review README.md for the project structure, analysis workflow, and file descriptions, and consult DATA_DICTIONARY.md for variable definitions and units. Run the Python scripts from the package root using the documented sequence in README.md. All repository scripts use relative paths and do not require access to the original raw roadway videos. The workflow reproduces trajectory-model diagnostics, full-maneuver vehicle metrics, turning-phase sensitivity, DCA/TCA calculations, TTC radius/threshold/horizon sensitivity analyses, false-discovery-rate sensitivity, source-video-cluster sensitivity, and the principal manuscript tables and figures. The public dataset contains 346 reconstructed vehicle maneuvers. Vehicle–pedestrian interaction analyses use 344 eligible events with positive common temporal overlap (117 before and 227 after). Raw roadway videos and field photographs are intentionally excluded because they contain real-world imagery and are subject to privacy considerations.

Institutions

• Tarbiat Modares University

  Tehran, Tehran

Categories

Licence