[Talk-ca] Merging OSM + Geobase

[William Lachance]

On Fri, 2009-03-27 at 08:19 -0600, James Ewen wrote:

> > Right, so what I'm proposing here is that we don't skip the existing
> > roads, but merge them. In the case of the HRM, I'd propose simply
> > replacing the existing positioning data with the geobase information,
> > which should solve the issue of stitching roads together.
> 
> That was decided against before the import experiment began. How would
> you propose merging the two ways? We went over this previously, and
> the consensus was to leave OSM data in place, and only import the
> GeoBase data where there was no existing data. The problem is that you
> either use the OSM nodes and ways, the GeoBase nodes and ways, or some
> hybrid of the two. If you are going to create a hybrid, then there's a
> whole new set of decisions to be made.

I do apologize if it seems like I'm bringing up the same old stuff all
over again. I really did try to go through the archives before posting
here, but all I could really find as far as decisions were concerned was
this thread from November which seemed to conclude that we should keep
OSM information intact no matter what:

http://lists.openstreetmap.org/pipermail/talk-ca/2008-November/000247.html

I see the argument of not wanting to blow away people's hard work, but
from my (admittedly NS-centric view), that approach will lead to severe
limitations in terms of the quality and consistency of the results. If
you look at the Halifax region, there's great swathes of OSM data in
Dartmouth which have no naming information whatsoever. To me it just
makes no sense to just throw out the rich set of NS geobase data for
these areas, just because someone did a first pass at mapping it out.

To be clear, creating a hybrid between the two data sets is exactly what
I'm proposing to do. I do agree that some sorting out of which
information should be kept around would be required, and the decisions
might be different depending on which area you're talking about. The
right formula may only be obvious after some experimentation.

> > Maybe I'm not understanding something, but why couldn't you just add
> a
> > new node to both ways at the point of intersection? I don't see why that
> > would look like a mini off ramp -- on the contrary, it wouldn't change
> > the actual geometry of either way at all!
> 
> Well, you wouldn't want or need to add two nodes at the exact same
> spot. Just add a single node and connect the two ways to that node. As
> to your second statement, have you actually looked at the same images
> of Camrose yet? Some of the ways stop short of the OSM ways. It would
> be simple to add a segment between the end of the Geobase way, and the
> existing OSM way. However some of the GeoBase ways extend across the
> OSM way. By connecting the end of the GeoBase way to the OSM way, you
> will be backtracking on the GeoBase way. The map representation would
> show that the GeoBase road crosses over the OSM road, and then doubles
> back on itself before you get to the intersection.
> 
...

Ok, after looking at Camrose, I see what you're saying. I don't see an
easy way of avoiding these issues if we're going to stitch two different
geometry sets together (which seems to be a requirement for at least
some areas). I can't claim to know the right solution(s) to them right
now, but I'll keep this in mind as I run my experiments.
-- 
William Lachance <wrlach at gmail.com>