Is AI-Based Toponym Extraction of Street-Level Imagery a Reliable Approach for Validating OpenStreetMap Toponyms?

This repository contains code, data and the main results of the forthcoming paper "Is AI-Based Toponym Extraction of Street-Level Imagery a Reliable Approach for Validating OpenStreetMap Toponyms?".

The published paper can be found at the following address:

https://www.scielo.br/j/bcg/a/f9FhC3C3wMKKrzznyCGfYDp/?format=html&lang=en

Aims: To develop an automated process that can confirm the existence of OSM toponyms by extracting toponyms from street-level imagery using artificial intelligence techniques and textual correspondence analysis.

Flowchart illustrating the proposed research framework.

Jupyter notebooks

These jupyter notebooks files, encompasses code blocks for fetch data, preprocessing and analysis, providing a comprehensive running example.

• ../notebooks/JN1_retrive_toponyms_OsmHistory_IndivFeat_v2.ipynb
• ../notebooks/JN2_get_Mapillary-images.ipynb
• ../notebooks/JN3_get_GSV-images.ipynb
• ../notebooks/JN4_train_Yv11_on_svt-dataset-local.ipynb
• ../notebooks/JN5_txt_detect_Yv11-keras_ocr_local.ipynb
• ../notebooks/JN6_toponyms_corresponde_analysis.ipynb

Requirements

• numpy
• time
• os
• pandas
• geopandas
• requests
• json
• folium
• shapely
• pyproj
• matplotlib
• seaborn
• mmapclassify
• statsmodels
• pysal
• mgwr
• tqdm
• sklearn
• ipywidgets

Observations

• Please note that some Jupyter Notebooks have tested on the Google Colaboratory environment.