This project explores how artificial intelligence can help cities measure pedestrian infrastructure that is missing from conventional planning datasets. Walkability is usually evaluated through network connectivity, density, and access to destinations, but these measures often assume that every intersection is equally crossable. In reality, missing crosswalks, faded markings, long crossing distances, and limited traffic control can create serious gaps in the pedestrian network.
Using Philadelphia as a case study, this project develops a computer vision workflow that combines U-Net segmentation from aerial imagery with YOLO object detection from Google Street View. The pipeline identifies crosswalk locations, estimates crossing conditions, and detects stop signs and traffic signals across multiple neighborhoods.
The project produces a scalable first-pass inventory of pedestrian infrastructure, including 3,610 detected crosswalk polygons and thousands of street-view-based traffic control detections. Rather than replacing field surveys, this workflow offers cities a lower-cost screening tool for identifying where pedestrian infrastructure may need closer inspection, maintenance, or redesign.
