Kuroyume0161

12 December 2006, 08:34 AM

Here's a topic that you won't find on Google (or anywhere else).

Consider this scenario: You have a Wavefront OBJ polygonal mesh with groups. The groups are seamed up against each other - that is, there are duplicate vertices at meeting polygons between groups. These duplicate vertices must be removed. The first thing you think is - vertex welding! Wrong! The vertices must NOT be touched in any way, shape, or form (literally people). Their number and order must remain consistent for deformations and morphs. But the polygons' vertex indices are free to be altered to 'weld' the polygonal seams.

I've searched Google (using several dozen term variations), IEEE, and ACM with no results of interest. My library of 3D texts is void on the topic.

I do have a veritable 'brute-force' algorithm in place that does this very well - but it gets really, really slow as the vertex/polygon count increases (say, over 40,000). There has to be a better way - and this is evident in the applications where it is done. My feeling is that there is a structural methodology to increase the performance of the task - i.e.: arrange the polygons and vertices (and other working data for the welding process) so that the seam welding is more optimal. But hints to what this better way is are not evident. The closest paper that I've encountered is by a Japanese guy from 1961 ("Welding of polygons and the type of Riemann surfaces") which deals with Conformal Mapping. Except for the fact that there is no algorithmic discussion, the math involves ultra-super-complex maths (Riemannian spaces, triple integrals, and all sorts of other fun stuff), and never really seems to talk specifically about 'welding polygons', it is the best found so far... ;)

Anybody have anything?

Thank you very much,

Robert

Consider this scenario: You have a Wavefront OBJ polygonal mesh with groups. The groups are seamed up against each other - that is, there are duplicate vertices at meeting polygons between groups. These duplicate vertices must be removed. The first thing you think is - vertex welding! Wrong! The vertices must NOT be touched in any way, shape, or form (literally people). Their number and order must remain consistent for deformations and morphs. But the polygons' vertex indices are free to be altered to 'weld' the polygonal seams.

I've searched Google (using several dozen term variations), IEEE, and ACM with no results of interest. My library of 3D texts is void on the topic.

I do have a veritable 'brute-force' algorithm in place that does this very well - but it gets really, really slow as the vertex/polygon count increases (say, over 40,000). There has to be a better way - and this is evident in the applications where it is done. My feeling is that there is a structural methodology to increase the performance of the task - i.e.: arrange the polygons and vertices (and other working data for the welding process) so that the seam welding is more optimal. But hints to what this better way is are not evident. The closest paper that I've encountered is by a Japanese guy from 1961 ("Welding of polygons and the type of Riemann surfaces") which deals with Conformal Mapping. Except for the fact that there is no algorithmic discussion, the math involves ultra-super-complex maths (Riemannian spaces, triple integrals, and all sorts of other fun stuff), and never really seems to talk specifically about 'welding polygons', it is the best found so far... ;)

Anybody have anything?

Thank you very much,

Robert