[Korrelan]

so I am focussed on the parts rather than the emergent layer, which I expect will come about naturally

As usual I would suggest you start small, on a very limited dataset, just to get the gist of the problems you are likely to encounter.

I have found from experience that these types of systems are best tackled by bringing the whole system along at the same time/ pace rather than concentrating on one section only to have to re-jig later down the line due to an un-foreseen problem in another module.

Depending on your programming level you could define the roads with thick black lines and the pot holes by blobs on a simple bitmap.

You can then write the various types of logic required to map the holes and simulate fixing them.  I know this is not the end intended use for the system but it will give insights you won't gain from theorising alone.