5 replies to this topic

[yasmin_bd]

I have a sequences od 2D images (jpg files). Is it possible to reconstruct a 3D surface from these images? Is any software tool avaiable for this?

[slicer4ever]

do you have a depth map of the scene?, if so it's pretty trivial(although it also depends on the complexity of the 3D surface you want to create).

 

it's possible to do it with color images, but the results i've seen are used to re-construct entire city's with 1000's of photo's here

 

more information about what you have, and exactly what you want to accomplish would be helpful in pointing you in the right direction.

Edited by slicer4ever, 15 January 2013 - 11:25 PM.

[Ohforf sake]

Yes, this is possible, although I would consider it an ongoing research topic. You can google for "Structure from motion" to see some cool reconstructions.

 

I can't name any commercial software tools for this (which doesn't mean that they don't exist), but I believe there is a free software bundle for PMVS (http://grail.cs.washington.edu/software/pmvs/) which is a very good algorithm for this task.

 

If you want to write this yourself, there is a wide range of different approaches and choosing the right one would depend on your actual usecase. Do you need speed (realtime) or quality, do you have markers, even indoor vs. outdoor can make a difference in some cases.

[Arkhyl]

Look for Bundler (http://phototour.cs.washington.edu/bundler/) for the Structure from Motion part (retrieving of cameras position, orientation, ...) and after that PMVS to generate a point cloud of the surface (Bundler gives a script to generate PMVS input from its own output). Then you can try http://www.cs.jhu.edu/~misha/Code/PoissonRecon/Version4.51/ or maybe Meshlab to obtain a triangulation of the point cloud.

 

There is also 123D Catch (http://www.123dapp.com/catch) that generate a triangular model and a texture atlas from images, but I never been able to obtain as good results as with Bundler + PMVS.

[yasmin_bd]

I need to write down the problem:

I have a video captured by an edible camera which had been taken by a patient, then it took videos , pictures of the internal digestive tract. I need to reconstruct the 3D model of the digestive tract from the videos. Here it should be noted that the digestive tract is a bit tubular in shape and is twisted every now and then.

 

1. At this moment I have just the videos with me. I don't have any information about the camera orientation or anything. But if needed, I can manage to get it.
2. Can bundler be installed on Microsoft Visual Studios 2010? I found from the link that it can be installed on linux, windows (cygwin). I am not familiar with Windows (cygwin). I need some help with installation on MSVC 2010.
3. I am familair with C++,OpenGL, 3D modelling; but don't have much knowledge on image processing. Is it ok to work with blunder?

Please give me some suggestion so that I can start working on the project.

[Ohforf sake]

Well, I'm not sure if anyone ever tried to map the digestive system with this but it sounds like an iteresting usecase :-) One problem you will encounter is drift. Bundler will extract the camera's motion from the movement of feature points in the video. This is of course not 100% exact and errors will accumulate. This is why you usually try to make a video sequence, where the start and endpoints match, also called loop closure. In your case this might be a little ... unhygienic. So you might end up with intestines which are not correctly coiled up.