How are they building the 3D views? Mapping cars driving around with sensors? I was under the impression that the shapes were inferred from satellite images, but the moment I step outside the major cities, it's back to flat maps. Even the couple hand-modelled buildings previously present in maps seem to be missing here.
It is possible to use satellite imagery for photogrammetry [1], but Google is using imagery from airplanes, possibly combining it with ground level data from their street view cars.
SLAM techniques allow reconstruction of a scene from video+position data. As the aircraft takes video of the terrain, it is logging its position. Large-scale batch processing can recover the 3D positions for each image feature that will minimize some error metric.
Rural areas probably have lower-quality image data (or noisier aircraft position data).
This is exactly what I want to know. There's several towns in the UK that have all the houses in really good 3D (mostly the coastal towns) - I struggle to believe that there's a team of humans making these...
I'm not sure that's it's only LIDAR data. There are e.g. some cranes and scaffolding that also are 3D modeled. I assume it's some magical machine learning pipeline that accumulates the aerial and street view imagery, LIDAR where available, and generates a 3D model.
