Assisted Road Surface Tagging using Machine Learning
summary
An ML pipeline consumes USGS NAIP imagery plus existing OSM traces, produces a segmentation mask for road pixels, then labels each road segment paved or unpaved and returns an obscuration/confidence score. The implemented classifier reports about 96% accuracy on paved and 95% on unpaved across varying tree cover. The code runs locally now and the project asks for donated compute to batch-process nationwide and publish predicted surface tags for OSM routers (Valhalla, etc.) to improve rural routing.
AI-written summary · about
related talks
19mHow we leverage AI to build great maps at massive scale
33mWhat and where? - Machine learning for geospatial image analysis
22mUrban Artificial Intelligence in Amsterdam
46mML Use Cases at Think Big Analytics
36mClassification of Satellite Imagery
31mDetecting Road Conditions from Space Using PyTorch, Public Data, and Free Satellite Imagery
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
22mIsland Mapping: Measuring Urban Footprints with Mapillary & Open Source Tools