Building footprints from a Bing basemap

Contents

Building footprints from a Bing basemap#

No footprint inventory? Stitch a basemap of the area, let SAM 3 find the roofs, then refine the rough masks into clean footprints against the high-resolution drone imagery.

Source: examples/building_detection_from_bing.py.

from rapidtools import (
    AerialImageryExtractor,
    BingOrthomosaicExtractor,
    BoundingBox,
    BuildingRegularizer,
    SAM3OrthoFeatureExtractor,
    download_dataset,
)

# 1. The area of interest: the extent of a drone orthomosaic.
[drone_raster] = download_dataset('eaton_patch1')
region = BoundingBox.from_raster(drone_raster)

# 2. A Bing basemap of the same area, as a GeoTIFF.
bing_tiff = BingOrthomosaicExtractor(zoom_level=19, max_workers=10)(
    region=region, output_path='eaton_patch1_bing.tiff'
)

# 3. Preliminary roofs from SAM 3.
preliminary = SAM3OrthoFeatureExtractor(
    prompt='building roof',
    patch_size=50, unit='meters', overlap_ratio=0.25,
    batch_size=4, threshold=0.50, mask_threshold=0.40,
)(bing_tiff)
preliminary.to_geojson('eaton_patch1_buildings_preliminary.geojson')

# 4. Crops around each preliminary footprint, then regularization.
with_crops = AerialImageryExtractor(
    dataset=bing_tiff, save_directory='output/building_crops',
    buffer_asset='20 m', force_square_image=True,
)(preliminary)
final = BuildingRegularizer(batch_size=8)(with_crops)
final.to_geojson('eaton_patch1_buildings_final.geojson')

Notes#

  • Point AerialImageryExtractor at the drone raster instead of the Bing TIFF in step 4 to regularize against post-event imagery.

  • examples/vegetation_detection_from_aerial.py uses the same extractor with a 'vegetation' prompt and low thresholds to map canopy.

Contents