Road centrelines from an orthomosaic

Extract paved roads from a drone orthomosaic and reduce the raw masks to a connected network of centrelines with widths.

Source: examples/road_detection_from_aerial.py.

from rapidtools import RoadwayRegularizer, SAM3OrthoFeatureExtractor, download_dataset

[raster] = download_dataset('eaton_patch1')

raw_roads = SAM3OrthoFeatureExtractor(
    prompt='paved road, street',
    patch_size=100, unit='meters',   # roads need context, so larger tiles
    overlap_ratio=0.25,
    batch_size=4,
    threshold=0.25,                  # roads are low-contrast; accept more
    mask_threshold=0.25,
)(raster)
raw_roads.to_geojson('roads_raw.geojson')

centerlines, polygons = RoadwayRegularizer()(raw_roads)
centerlines.to_geojson('roads_centerlines.geojson')
polygons.to_geojson('roads_polygons.geojson')

Each centreline carries width_ft and azimuth_deg attributes. min_width_ft enforces a floor on sampled widths and min_network_length_ft drops disconnected stubs.