nearest neighbour, kinda

let’s say I have some blobtracking IDs/channels and due to background substraction not being 100% accurate, i sometimes get an 1-4 blobs per person (2 for the body, 1 for each arm, … )

So now i need to merge/average these points, but at the same time keep blobs from another person that’s standing nearby (as far as possible) untouched. How would I approach such a problem in touch?

So far I can only think of using the consolidate featured in the facet SOP, but that one snaps to the nearest point depending on the sort order of the points. So in my case it might snap to the arm instead of the main body blob … So an average position would be much nicer (facet SOP RFE?). Any ideas on how to improve the facet based setup? Or a totally different approach?

I don’t exactly get what the case is but if you can get some additional info from a processed top, you may use a sort op elevation map effect that actually moves the points in the direction of the camera as to create sort of layers to aid the Facet SOP merging thing.
maybe is obvious or not applicable, but I’m just in the process of shutting down and thought that was better to blahb this sketchy concept out before my tomorrow.

example proj?