only use roof lines which are inside the building outline

[hjanetzek]

Hey, good work with live editor for roof lines. One issue I found is that ridges that are part of another building(b) but connected in one node to building (a) are also considered to be part of a. I think roof lines should only be considered if they are completely inside or connect in two nodes (i.e. overlap with the building outline)

[hjanetzek]

see relation 147095 for example (admittedly most of the mapping is wrong use of roof lines imo) http://www.openstreetmap.org/browse/relation/147095

[kendzi]

It is wonderful example that roof line tagging schema is not so clear as it looks like at begin. For example it is hard to guest to which building part should that roof edge belongs: http://www.openstreetmap.org/browse/way/223907268

Currently it is rendered almost as expected, but if we remove edges which don't lies inside building part, that edge won't be render.

Personally I think that edges and ridges should be connected only with building parts which they belongs. So if edge belongs to building “A” and don't belongs to building “B” I should not be connected with building “B” at all.

Some more interesting errors:

• edge with out start point (flaying start point) http://www.openstreetmap.org/browse/way/223907262
• ridge not connected with outline http://www.openstreetmap.org/browse/way/223907265 http://www.openstreetmap.org/browse/way/223907257

[hjanetzek]

No question this is an example of a particular bad use of roof-lines. I was talking about http://www.openstreetmap.org/browse/way/223907249 where the ridge node that connects also to the neighbour building part interferes.

My algorithm currently completely ignores roof-lines which are not connected to the building outline. So roof-lines must subdivide the outline into faces. Not sure if one should handle other cases at all. http://opensciencemap.org/~jeff/test/#scale=19,rot=307,tilt=37,lat=48.137,lon=11.575

If there was a plugin for josm to check whether roof faces are planar (or which would allow to move roof-lines only along directions where faces stay planar) more people would use roof-lines correctly. I think one could support it by automatically creating the straight-skeleton ways for the outline which can than be refined while keeping the constraints of flat site. Not an easy project for sure. btw this tool looks also like an interesting approach to model roof-lines http://twak.blogspot.de/2011/09/siteplan-source-release.html

[kendzi]

Show how roof lines are parsed currently and it is not look good. I will try to add your rule for roof lines.

I watch Siteplan and it base on modified skeleton algorithm this mean it is far away from roof line tagging schema.