Would some kind of clustering approach also work here?(K Means or DBScan maybe?)

i mean, if there's a box per triangle, that would be the smallest box, but i guess that wouldn't be very efficient or useful