[Talk-ca] RoadMatcher

[Steve Singer]

I've been playing around with RoadMatcher 
(http://www.jump-project.org/project.php?PID=RM&SID=OVER) and thought I'd 
share some of my findings.

I've put both a planet.osm dump and some of the NRN provinces into a PostGIS 
database and have pointed roadmatcher at it.  I use auto-match only (no 
adjustments) and generate a listing of OSM ways that are matched to a NRN way.

With this listing we would in theory have the option of

1. Excluding those NRN road segments from the import.  I've modified my version 
of geobase2osm.py to do this and have generated test some test .osm files.  If 
we go this route, then we will need to think about how we will get road names, 
street addresses etc into the excluded ways.

2. We could use that output to delete the osm ways because we see them as 
likely conflicts with the NRN data.

One question is how well does road matcher work, and is it worth the effort in 
running.  I don't have a metric of measuring 'how well' RoadMatcher works.  Nor 
do I yet know what the best RoadMatcher settings are.

I have taken road matcher and generated 'matches' for some tiles and 
excluded these NRN segments from my geobase2osm.py import.

You can bring the .osm files into JOSM (with Open) and view them, comparing 
them to the exisitng OSM ways.  In a perfect world these .osm files won't 
have any overlapping ways with the exisitng OSM data (in practice this isn't 
the case)

NTS 090F, covers the Tofino BC area)
---------
http://www.mediafire.com/?zmtm2tov001

It did a decent job of matching many of the smaller roads.

NTS 030M05 (Hamilton to Oakville  Ontario)
----------------------------------------
http://www.mediafire.com/?dymoxmdlw11


In the spot checks I did roadmatcher missed many matches I was expecting it 
to make. I don't understand why it made some matches but failed to match 
some others.  In other cases the Ontario NRN geometry is strange (they 
sometimes model a residential street with 2 segments, then switch to one 
segment then switch back to two (all in between an intersection).


(NTS 074D , covers Fort McMurray, Albert)
--------------------
http://www.mediafire.com/?4qtwcgnbn1m

The matching seemed better here than Ontario but it is hard to know which 
ways are duplicates because they are newer than my canada.osm dump and which 
road matcher is missing.

Warning: I've discovered that some of roads in OSM were added after I 
grabbed my canada.osm dump (a number of weeks ago).  This accounts for some 
roads you'd expect to match but didn't but in some cases roadmatcher failed 
to match for no good reason I can see.




It seems to do particularly bad on matching the long low resolution traces of 
major highways but the success rate on shorter ways is more promising.

Also, it looks like it is possible to run roadmatcher in an automated fashion 
for this type of match.  It would require some scripting though and preping 
of input files.

My intent in posting these .osm files is to show what RoadMatcher is capable 
of NOT to get feedback on the tagging of the converted ways (I don't think 
the vesrion of geobase2osm.py incorporates Jason's recent changes)

Please don't try to upload these files to the OSM database (be patient).


Steve