Excited about these techniques, I did some tests with VisualSFM & CMPMVS.
You don't need a stereo camera and a normal camera will do just fine. Just need a lot of overlapping pictures. Take a video and export every 10th frame as a picture worked fine as well.
Tried it on the exterior of a nearby mill. Calculation of the 3d model using these programs took several hours - 123D created the 3d model in several minutes but it had less detail (64k vertices vs 1.1M vertices) and the textures looked blurry.
The end result resembles the building but it's not really useful to me (yet):
- Nothing is flat in the 3d model - even if the real thing has flat surfaces. That will make rebuilding a low poly object in SketchUp difficult - how to decide which bump/point of the model to snap to??
- It's hard to get the (textured) model into SketchUp. Meshlab opens the model in seconds. Blender in a minute, SketchUp's ply importer is already busy for 3 hours and I'm not sure if it will ever finish. Maybe have to split up the large model in Blender into smaller ones and rebuild it in SketchUp?!
Also:
- reflective surfaces don't work well (found that one out 'scanning' a table with some objects on it)
- the software needs 'anchors' to match the individual frames. For more large uniform planes (ceiling, white walls etc) the software can't find matches so that results in large gaps.
Still, it's fun to see the software build something 3d out of pictures.
