I need to display 3d sculptures in my application, but all I will have initially is 2d images of sculpture from different angles( angle and count of images is under our control)
1. Is there any library which can do that, e.g. take 8 images from all side and stitch it into a 3D model? Any opensource/commercial library/product is ok.
2. Alternatively how difficult it would be to do if I start from scratch? what direction you would take e.g.
Simple solution one
3D model is not a strict requirement, but the end result should be that user can rotate the object and see it from different angles so a simple solution could be to snap 100s of photos from all angles and change them as user changes angle but it would be very cumbersome to convert many sculptures.
There was a question on Super User that asked:
"Is there any free/open source software that converts photos to 3d models?"
but is has since been deleted. (It's here for anyone that has > 10K rep on SU)
From the accepted answer:
I think insight3d does exactly what you are asking for.
Having been involved in 3d graphics in the 1990s and early 2000s I would have said that you wouldn't have found anything suitable as it is a very hard problem.
However, I checked out the insight3d site and it does seem to be legitimate.
I think this is a hot research topic... I've seen many technical demos showing partial reconstructions, in domains like architecture or simple geometrical objects; so I'm not sure if you could find a drop in library that just works for every model (with just a few photographs).
Programming it from scratch is insane. Just to make the mathematical models could take a lot of time to skilled mathematicians (small improvements in the area could be worth a Ph.D.'s work), not to talk to "plain" computer scientists... and the implementation wouldn't be easy either.
In the other part, I know there exist 3d scanners that are used everyday in the games and animation industries... but I'd bet they are quite expensive, and don't fit with your requirements.
P.S. Don't take my words as definitive, as I'm just an aficionado and there's been a time since I looked the state of the art.
Have you considered using stereo cameras to take the pictures?
Microsoft's Photosynth is in the same ballpark, but may not be precise enough for your needs.
Try AgiSoft Photoscan... there's a trial version which works fully but won't export models, or save. But it's worth a try. It builds Meshes which are quite amazing. Often look better than Photosynth's point cloud results.
You can also look at Bundler and PMVS2... but they're a bit more fiddly to get used to. http://phototour.cs.washington.edu/bundler/
来源：https://stackoverflow.com/questions/1532184/convert-2d-images-to-3d-model