Object Reconstruction from Photographs

Object reconstruction is using a computer to build up a 3D model from a dataset.

In this case, the researchers used a bunch of photographs of buildings (like Notre Dame Cathedral) from flickr:
[uwnews.washington.edu]
They start out with random photos with no explicit camera position data and end up with a (slightly lumpy) digitalized model of the building.

This is applicable to RepRap because we'd be able to use a digital camera by itself as a 3D scanner; digitalizing a desktop-sized object just by taking a bunch of pictures of it

This is the journal article associated with the press release:
[uwnews.org]

This is cool! I've always been fascinated by this sort of technology. I'm waiting for the day when I can take out my digital video camera, wave it around in front of an arbitrary object, hit the "print" button conveniently placed on it, and have a plastic copy of it ready in a few hours. Full color in the copy is a bonus.

Looks interesting.

Questions

"They start out with random photos with no explicit camera position data and end up with a (slightly lumpy) digitalized model of the building. "

Can I take this to imply that if the camera positions were known that the results would be better/easier to compute ??

cheers

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

Interesting. I've downloaded the paper and bookmarked their website. Hopefully there's something in there to help me with my goal for my PhD of creating a webapp that does simple 3D scanning of an object from 5 or 6 photos for use by 'normal' people.

Hey Reece

That wouldn't be a web cam mounted on a Darwin on a rotating mount instead of an extruder would it ??????

Cool PhD

aka47

---

Necessity hopefully becomes the absentee parent of successfully invented children.

That would be the ultimate aim but to start with I'm thinking of a website where you can upload and orient pics taken with a digital camera form the left, right, front, back, top, and optionally bottom.

Sounds a lot like [www.arc3d.be] from KULeuven. Its is a webservice.
Other free or DIY 3d scanners: [blog.makezine.com]
And meshlab.sourceforge.net/ works very well for optimizing 3d scan data.

Another option, someone has released a point cloud extractor for Microsoft's Photosynth (photosynth.net), search the web and you'll find it.
It should be quite easy to build a convex hull from this point cloud, smooth off the noisy points in polygon reduction software, then use a 3D library that exports to STL to dump the meshes out.
The important thing here is that Microsoft just bought the talent, the code that Photosynth is based on was originally released under the GPL - a free software 3D scanner using photographs at any orientation should be possible.

That photosynth thing works best if you put it on a plate with a steppermotor under it, and then make the machine take one step at a time while taking photographs (from webcam for example). If you do it correctly, you will end up with lots of pictures. I tried it with a loose hand and taking pictures from all kinds of angles, including above, but that will end up in blurry 3D.

I think you could get fairly accurate results with a acceleration sensor and gps. Of course this doesn't work for existing photographs. The better way is a second camera though.
But there is nothing against combining the 2 since a position wouldn't hurt anyway since otherwise the error can accumulate .

That insight3d software seems to be similar - and opensource

DeuxVis Wrote:
-------------------------------------------------------
> That insight3d software seems to be similar - and opensource

Nice find, but it doesn't seem to have been updated since 23/7/2009. I think things have moved on a bit since then.

I think you'd love Autodesk Photofly project. It takes your set of pictures and it returns you a 3D mesh of the object/building/room.

[labs.autodesk.com]

Stunning!

misan