It seems to me that having a huge database of known locations of vehicles could be used to do some cool (but likely immoral) things, specially coupled with the fact that several state agencies all over the world make it free and easy to retrieve licensing and ownership information for vehicles. It'd also be relatively much easier to do compared to facial recognition.
The wildest idea I can think of is using reverse camera on every car and recognizing the plate of the car which is tail gating you. Now, that's more interesting, but how much of a fleet you would need to track down a random car.
Not something car manufacturers would do either, even with their love of telemetry.
(About provate property: https://www.security.nl/posting/406881/Juridische+vraag%3A+m... )
> I'd also look at things like using neural nets to detect race and/or underlying genetic structure from visual, audio and other metadata
I don't see the use in sufficiently cosmopolitan cities, which are the only places with the camera density to do this. I think there's a better case for (using audio recordings) predicting social circles, because close associates tend to pick up both tonal patterns and vocabulary patterns.
At some stage license plates will have to disappear, but then you move to facial recognition for cars and other meta data, they probably have unique sounds for instance.
I'd try and do a app like https://en.wikipedia.org/wiki/List_of_barcode_games
You should know how to monetize it.
(Obviously facial recognition works equally well here, but there is some regulatory attention, although none that would stop you)
I'd also look at things like using neural nets to detect race and/or underlying genetic structure from visual, audio and other metadata.