Street view-based infrastructure assessment of motorcycle hotspots: A national road network analysis in Thailand

Code accompanying the paper:

Street view-based infrastructure assessment of motorcycle hotspots: A national road network analysis in Thailand
Accident Analysis & Prevention, 2026. DOI: 10.1016/j.aap.2026.108706

This repository contains the reproducible analysis workflow used to connect motorcycle casualty hotspots on Thailand's road network with infrastructure attributes observed from street-level imagery. The workflow combines crash data preparation, OpenStreetMap road-network processing, Network Kernel Density Estimation (NKDE), Mapillary street-view sampling, vision-language-model feature extraction, and hotspot/control factor analysis.

Raw crash records, road-network extracts, street-view images, API credentials, and generated model outputs are not included because of data licensing, privacy, and file-size constraints.

Repository Structure

VisionToRisk/
|-- exploratory/
|   |-- TRAMS.ipynb
|   `-- OSM_explore.ipynb
|-- nkde/
|   |-- NKDE_TRAMS.R
|   |-- NKDE_TRAMS_LAMPANG_TAK.R
|   `-- nkde_result.ipynb
|-- vlm/
|   |-- SVI_feature_retrival.ipynb
|   `-- index.js
|-- hotspot_analysis/
|   |-- hotspot_and_control.ipynb
|   |-- hotspot_analysis.ipynb
|   `-- factor_analysis.ipynb
|-- fig/
|   |-- SHAP.png
|   |-- SHAP_bee.png
|   `-- SHAP_noAADT.png

Workflow

1. Crash and network exploration
   exploratory/TRAMS.ipynb prepares and inspects Thailand motorcycle crash records. exploratory/OSM_explore.ipynb evaluates OpenStreetMap road classes and network attributes used for network-based analysis.

2. Network hotspot detection
   nkde/NKDE_TRAMS.R and nkde/NKDE_TRAMS_LAMPANG_TAK.R implement NKDE for motorcycle casualty hotspots. nkde/nkde_result.ipynb post-processes hotspot outputs and evaluates hotspot capture using severity-weighted crash measures.

3. Street-view infrastructure extraction
   vlm/SVI_feature_retrival.ipynb samples lixel locations, retrieves Mapillary street-level imagery, and extracts infrastructure attributes with a vision-language-model workflow aligned with the paper's road-safety feature taxonomy.

4. Hotspot/control analysis
   hotspot_analysis/hotspot_and_control.ipynb constructs hotspot and control samples. hotspot_analysis/hotspot_analysis.ipynb summarizes hotspot characteristics, and hotspot_analysis/factor_analysis.ipynb estimates associations between extracted infrastructure features and hotspot status.

Data Availability

The repository intentionally excludes:

• raw TRAMS/crash records;
• downloaded OpenStreetMap or other road-network data;
• Mapillary images and API credentials;
• intermediate geospatial files;
• generated SHAP outputs, model comparison tables, and scratch figures.

Users should place local inputs in ignored data/output directories and update notebook paths for their own environment before running the workflow.

Software Notes

The analysis uses both R and Python notebooks/scripts. Core geospatial and modelling dependencies include typical scientific Python/R packages for spatial data handling, network analysis, machine learning, and visualization. Exact package versions should be recorded from the execution environment used to reproduce the paper results.

Citation

If you use this repository, please cite the paper:

@article{jongwiriyanurak_2026_street,
  title   = {Street view-based infrastructure assessment of motorcycle hotspots: A national road network analysis in Thailand},
  author  = {Jongwiriyanurak, Natchapon and Haworth, James and Ameen, Horia and Christie, Nicola and Soil{\'a}n, Mario},
  journal = {Accident Analysis \& Prevention},
  year    = {2026},
  pages   = {108706},
  doi     = {10.1016/j.aap.2026.108706},
  url     = {https://doi.org/10.1016/j.aap.2026.108706}
}