Poly2tri

[hjanetzek]

Please review :)

Compared to jts conforming delaunay, poly2tri only adds vertices where necessary for triangulation so it results in much less vertices which reduces obj file size and less vertices to interpolate which should make it faster. Also the results look better on complex roof as the interpolation finds not the best segments sometimes (or there are none)

vertices: 2266, faces: 3773

vertices: 3124, faces: 5203

http://city.informatik.uni-bremen.de/~jeff/dom8.osm

[hjanetzek]

an update to the last commit:

/* join colinear segments, but not the nodes that are connected to ridge/edges

• often there are nodes that are only added to join one building to another
• but these interfere with proper triangulation. */ <- I mean correct height interpolation of the triangulation

visible at the center and left green domer

[tordanik]

First off: I agree with you that poly2tri's constrained "Delaunay" triangulation is often preferable to using the algorithms available in JTS, so I will eventually merge this.

What makes this a bit more complex is that I've actually experimented with that same library myself recently for a different part of OSM2World (SRTM terrain heights), and have therefore written my own wrapper code. This means I would like to take the time to compare our approaches and pick the best parts from each. I will probably will not get around to doing so before the end of next week, though.

By the way, a drawback I've noticed while working with poly2tri is that it is very touchy when it comes to the input. If the data it isn't perfectly prepared (filtering duplicate vertices and so on), it is prone to just crashing with stack overflows and stuff - which cannot even be properly caught like exceptions from JTS can. How successful have you been in reliably avoiding these?

[hjanetzek]

I have seen this problem with poly2tri in another project but there I was using the default polygon triangulation. The big issue here was to figure out that TPoints with same coordinates need to be unique objects to make constraints work. It was just my last guess before I would have dropped that library and have used Triangle via jni.. To your question: no stack overflows so far :)

[tordanik]

I'm finally getting around to working on this again, sorry for the delay.

The poly2tri topic is not completely finished yet as I will have to merge my own code with yours. I may open an issue for this task once the code is ready for that; I hope that you will help out there a bit. :)