[OSM-dev] Determining if two linestrings are "similar"

Paweł Paprota ppawel at fastmail.fm
Mon Jan 7 11:10:07 GMT 2013


Hi Jukka,

Thanks for the links, I will take a look.

I've done some performance testing and all three (ST_Buffer,
ST_SnapToGrid, ST_Distance "minimax") solutions incur a significant
performance penalty with the massive amount of data that OWL needs to
process, i.e.:

Without this heuristic I get 20-25k changesets / hour.

With this heuristic I get 5-10k changesets / hour (depending on the
implementation).

Right now I employ a simple bounding box + linestring length heuristic
which has pretty much no performance penalty and work for > 99% of
cases. I still need to find some good test cases to check if I don't
omit some important changes but for now it works.

Paweł




More information about the dev mailing list