[nickreese]

Kinda hijacking the thread but... My hypothesis is that we will look back and see that Streetview imagery is a goldmine for AI and will be a path to being able to answer HARD questions about the real world.

The insane thing is there are only like 7 companies that actually have meaningful datasets.

I spent 1.5 years studying the geospatial space and went so far as buying a Mosaic51 and scanning the entire country of Andorra as a test before looking at buying the camera manufacture.

Ultimately I walked away from buying the company after issues with the family office I was working with... but long story short I believe streetview imagery will be a gold mine in the future.

If anyone is working in the space. Feel free to ping me, happy to chat and even make intros to the space. If you are training an AI, ping me as well. Happy to open my images up to the right person to make something "country scale" (160k images... every 3 meters with RTK labeled gnss data).

[CalRobert]

Anyone who uses a bike to get around will know the routine of streetviewing their route ahead of time to see how dangerous it is, if the bike lanes are, in fact, real, etc.

I also have a real estate project and want to work on AI analysis of local streetview to learn about the neighbourhood.

I've wanted to build something to automate this with AI but haven't had time.

I would love to chat!

[globular-toast]

I have always wondered if/when Google will start using their streetview data to improve mapping. They could (in theory) generate directions like "turn left after the green building" and find speed limits, road surface, width, and potentially even bus routes and stuff like that. They don't seem to, though. The routes they generate are always incredibly naive when it comes to actual road type, like "let's go up this single track road with a 20% gradient to save 2 minutes".

Curious whether you think this is more than just improving mapping/routing related stuff, though.

[anamexis]

They do use streetview data to improve mapping, particularly for things like speed limits (like you mentioned) and other signage (street names, identifying which intersections have stop signs, etc.)

[maelito]

> "let's go up this single track road with a 20% gradient to save 2 minutes"

Not sure this is a good example : elevation data should be good enough to avoid this kind of roads. See e.g. https://sonny.4lima.de for Europe.

For your general thought, see here : https://blog.google/products/maps/google-maps-101-how-we-map...

> It all starts with imagery

> Street View and satellite imagery have long been an important part of how we’re able to identify where places are in the world—it shows us where roadways, buildings, addresses and businesses are located in a region, in addition to other important details—such as the town’s speed limits or business names.

So I guess what you're proposing is already done for several years but in more subtle ways.

Remember that Google Maps doesn't have the power of OSM. Hence the need for automation.

[Raphaellll]

I did a project that attempts to generate these types of instructions: https://map2seq.schumann.pub/nllni/demo/

[apayan]

> generate directions like "turn left after the green building"

I don't know about the rest of the world, but in the Los Angeles metro area, Google Maps already gives directions like this. "Turn left after the Carl's Jr.", "Turn right after the Starbucks". I notice it's usually done in areas where street signs are hard to see, but there is a clear landmark for the driver e.g. the golden arches of a McDonald's.

[Arelius]

To be clear, streetview was originally built to improve mapping. You can thank the wealth of street names and address labels on streetview.

[maelito]

The team behind Panoramax is already applying AI analysis. See e.g. https://forum.openstreetmap.fr/t/detection-des-stationnement...

Most Panoramax discussions are in French, but you'll find links to English code.