Towards automated mapping of pedestrian path networks at scale
summary
A pipeline converts satellite imagery plus existing street graphs into 4,800 annotated samples and trains a multi-input segmentation network to infer sidewalk, curb bulb and crossing pixels, then snaps and refines a graph to match predictions; an on‑device addon with stereo camera and GPS produces street‑level segmentations and merges them into OSM‑compatible output. The system produces pedestrian networks for six entire counties, exports standard Geo/OSM JSON, and in pilots reduces survey time by >80%.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:49Problem and significance
- 3:12Solution overview
- 5:50Inference pipeline
- 8:38Challenges with satellite data
- 9:24Street-level mapping system
- 11:56Pilot study results
- 13:42Summary and conclusion
related talks
19mHow we leverage AI to build great maps at massive scale
07mUrban Map Inference by Pervasive Vehicular Sensing Systems with Complementary Mobility
16mOptimized Routing with Satellite Imagery and Computer Vision
41mLarge Scale Field Delineation
22mUrban Artificial Intelligence in Amsterdam
28mEstimating Traffic Flow to find New Drive Thru Locations
01h 36mFrom Maps to Models: An End-to-End Journey in Geospatial Machine Learning with Python
33mWhat and where? - Machine learning for geospatial image analysis