The boundary extraction process works by first determining cluster points that belong to the boundary. This unordered set of boundary points is then spatially sorted into a contiguous boundary using what we call "Partitioned Neighborhoods" (PNs). The result (first attached image which shows a zoomed-in area of a particular data set taken from downtown Springfield, Oregon) shows the boundaries of three identified buildings, but they have a large number of extraneous points. I have been experimenting with various point thinning algorithms. Each operates as a filter using a "boundary points in - boundary points out" style of interface. This allows them to be plugged together and used in any order; even running the same filter multiple times with or without others filters being interspersed. The second attached image shows the result of using three different filters, one after the other. Preliminary results look promising. Large numbers of extraneous points get removed, and the overall shape pretty much retains its overall structure.
OK - so maybe I don't know how to do attachments here. You can see them here:
http://people.eecs.ku.edu/~miller/Wo...ent_Zoomed.png
http://people.eecs.ku.edu/~miller/Wo...eps_Zoomed.png
OK - so maybe I don't know how to do attachments here. You can see them here:
http://people.eecs.ku.edu/~miller/Wo...ent_Zoomed.png
http://people.eecs.ku.edu/~miller/Wo...eps_Zoomed.png
Comment