Hacker News Reader

Comment by bhouston | Hacker News Reader

I think that given that Google already has 3D depth coverage from its street view machines [1], it should be possible to combine that data with some medium resolution overhead 3D scans to create something similar, and likely even higher quality at the street level.

I wonder why Google hasn't done it yet. I don't think there are any real technical limitations. It may be that getting it fast is hard and the usefulness from an end user perspective isn't there yet?

[1] http://gizmodo.com/depth-maps-hidden-in-google-street-view-c...

View on HN

Google already has done it: https://www.google.com/maps/@40.7407861,-73.987388,348a,35y,...

The difference is that Google is using low-res satellite imagery, rather than high-res drone imagery. Maybe that will change now that they've bought Titan Aerospace.

Google uses drone imagery for some cities, San Francisco being one of them, using different angles to create the 3D model of the city, and satellite imagery for the rest of the planet. It's exactly what this company is doing. From what I can see, Google Maps also uses different textures depending on the viewing angle.

Check out https://www.google.com/maps/place/San+Francisco,+CA/@37.7623... and rotate the image. Check out the trees to see the seams of the different pictures used to create the textures/models.

That's pretty low-res to be drone imagery. If you look at that same area using the Google Earth plugin, it says that it's satellite imagery:

http://i.imgur.com/qCEE7cB.jpg

(updated screenshot to be from the same angle you have)

I'm not sure why you think it's lower res. It looks about equivalent to me. The trees look a bit odd in Google maps, but the quality of the building windows and cars looks about the same.

Here's a more nuanced theory: The building fronts, cars, etc look great because they are from street view. The roofs, treetops, etc, look terrible, because they are from satellite.

The roofs, treetops, etc. look terrible because the depth data is significantly lower resolution than the photos. This causes the deformed polygons you're seeing... The actual resolution of the photos of the roofs/treetops are just as good as everything else, they're just stretched over bad 3d data.

That's too much detail for landsat.

I don't have any idea what they are using there, but they do pull in a lot of aerial photos from the U.S. government. Those photos are (largely?) taken from piloted planes though.

I think some of the cities that have come online more recently are a bit higher-res ( example: https://www.google.com/maps/place/Copenhagen/@55.6686089,12.... ). Maybe they'll reprocess San Francisco eventually.

Does that mean this isn't cool? This is off the hook. Looks very similar to the iPad versions of Berlin.. high-res images quite a smart Level-of-detail differentiating between zoomlevels. With .bin files (maybe to keep the format closed? are just raw speed?)

If this is done in an automated way they are on to something pretty great.

Also to do this, is really really hard/lots of work. So kudos for that.

Reply on news.ycombinator.com