A live updating DBSCAN approach might also work. It would sort of break in the case of dense and uniform usage across the entire map (in which case there would

A live updating DBSCAN approach might also work. It would sort of break in the case of dense and uniform usage across the entire map (in which case there would be just one cluster), but otherwise might work well.