StreetTransformer: A Computer Vision-based Tool for Analyzing and Retrieving Longitudinal Changes in Urban Streetscapes

Urban street design plays a critical role in safety, mobility, community and public welfare. Yet, systematic, large analysis of streetscape change remains challenging due to the scale, heterogeneity and unstructured nature of publicly available data. We here introduce StreetTransformer, a unified, open-source, multi-modal visual analytics system and framework for gathering, comparing, retrieving and analyzing urban streetscape state and change over time. We construct a novel spatiotemporally grounded corpus for New York City’s more than 47,000 intersections by integrating longitudinal aerial imagery, segmentation masks, civic design documents and capital reconstruction project data spanning nearly two decades of New York City history. Innovatively, this corpus includes a large cache of 33,000 unstructured pdf document pages scraped from the NYC DOT website, which have been parsed, categorized and spatially linked to intersections using OCR and LLM-integrations. Additionally, project metadata, collected from NYC Open Data, is standardized and filtered to safety-focused interventions, grounding the Documents, Images and Masks with the city’s objective records.