Data

Recently Microsoft released global building footprints in ODbL.

Quality

I’ve found some challanging situations:

• high density buildings, tipically in italian villages
• high parallax imagery in downtowns
• footways or narrow streets covered by roofs

Test

I made several test and some local imports, taking in account in priority order:

• not overwrite mapper manual work
• keep consistent highway network
• check parallax false footprints
• simplify wherever possible
• overall offset if needed (Strava and GPS OSM repo references)

Preveza Case study

The town is a compendium of above quality problems. Some imagery spots are burnt by the sun and consequent low contrast was oddly computed by MS algorithms. It took about 5 hours to adjust footprints (and several highways). I applied the following workflow:

• data cropped by Preveza landuse=residential
• objects below 60 square meters filtered out
• object touching OSM buildings filtered out
• overlapping polygons manual fix
• manual rotation adjustments on many object (AI problem)
• highway adjusted by Stava heatmap

Tools

Qgis for initial conversion, crop and filtering. JOSM for offsets, rotations, ovelapping fixes.