Use ML and CV and image recognition to do something cool with google maps. It's just an interesting idea, use CV to analyze google maps data.
As someone working with something vaguely related to this sort of thing. Let me just say, use Open Street Maps. Google mapping data and the licensing thereof is a huge pain and often straight up impossible. Don't lock yourself into developing something for a platform where you will be forbidden from selling or even distributing the fruits of your labor. Plus with OSM you get direct access to the underlying vector data, which in most cases will save you an arduous and error prone processing step
Other than that, the whole field is wide open for all kinds of awesome ideas.
As someone working with something vaguely related to this sort of thing. Let me just say, use Open Street Maps. Google mapping data and the licensing thereof is a huge pain and often straight up impossible. Don't lock yourself into developing something for a platform where you will be forbidden from selling or even distributing the fruits of your labor. Plus with OSM you get direct access to the underlying vector data, which in most cases will save you an arduous and error prone processing step
Other than that, the whole field is wide open for all kinds of awesome ideas.