another jab at on site 3D acquisition!

Well this is just an update to my earlier post about experimenting with the mesh reconstruction using still pictures.

This test is looking better but still needs some more details. In my earlier tests where I just computed normals for the whole point cloud and meshed it, I figured its way more optimised if you clean up the point cloud and get rid of the junk point sets. Since I'm running Meshlab 1.2.2 I do not have the luxury of selecting and deleting the point sets within meshlab. So I just used Houdini instead!

The best part is houdini's bgeo format supports multiple datatype functionality akin to the ply format (which means you can still preserve your color information within houdini and move the data from Houdini to MEshlab pretty seamlessly).

I just figured one issue though, which again is easily resolved. The mesh generated by Meshlab appears to be black, this is because of the normals data getting screwed up. So once the mesh is imported into Houdini we could just recompute the normals which isn't time taking at all as we have the mesh.

So just append a facet sop and recompute the normals! Sweet!

Adios!

My first 3D scan! On site set acquisition.

I have been working on some projects in the uni and so though about if I could possibly do something about dummy geometry models

I used a bunch of photographs (of a friend's backyard) to create a point cloud (photosynth) and then meshed it with meshlab. Voila I had a decent dummy geometry ready in one day! Now that I know its achievable, its time to do some more tests for my actual backplate.

Still need to work on textures..trying to get the texture from the point cloud data itself using meshlab..

Below are some results of the same (mesh rendered with ambient occlusion within meshlab)