Mind that they don't only take pictures. The cars are full of sensors. From GPS to verify coordinates of roads, over Wifi to scan SSIDs for locating devices without GPS to LIDAR to scan 3D features and probably a lot more ...
They already know the photos have the objects they are interested in. Are they still working on motorcycles buses and signs?
The recaptcha part doesn't even depend on recognition, it's just a statistical comparison with what other humans clicked on given the same prompt + some pointer analysis and browser fingerprinting.