A methodology for using street view imagery, scene text detection, and GeoDBMS in urban geography studies

The city is an open book from which, until recently, we could not read more than excerpts. However, with the advent of API integrated-street view imagery platforms, deep learning text detection-recognition-script identification (DRSid) with computer vision using a Node.js application called ‘street-reader’, as well as the geospatial tools like ArcGIS Pro needed to analyze trends in big data, it is now possible to acquire near-total linguistic datasets of the linguistic landscape with high rates of accuracy and reproducibility. This methodology is designed to vastly scale up data collection, which allows for insights into temporal and spatial trends not previously possible. Data handling issues related to each step in the methodology are investigated and documented, including: imagery resolution, Google Street View API request metadata, double detection errors, benchmarking results with native speakers, issues related to watermarks, scene text detection errors, managing data with PostgreSQL and PostGIS, encoding practices, join practices, OSM layer integration, and Clopper-Pearson significance testing.