3d Conversion

[jlutgen]

On page 25 of the "Guide to Highlight Image Capture" in the section on Quality Tradeoffs, a reference is made to converting the RTI data into 3D. I recall that during our training class Mark stated that converting RTI data to 3D was problematic. So, do you know of a workable method for converting RTI data into 3D?

[marlin]

Jerry,

 

At this point in time, we do *not have a fast easy way of converting RTI data into 3D data, (not that we are passing on to our users).

 

I believe that it is possible in a lab environment with lots of processing power, specialized capture rigs, and numerous software processes.

 

You might want to consider capturing Photogrammetry sequence in addition to the RTI data capture. Its pretty straight forward and uses the same capture gear as H-RTI.

 

Photogrammetry will get you into the 3D model mode quickly.

 

Check this out:

http://culturalheritageimaging.org/Tech ... grammetry/

 

Heres a PDF manual from the BLM:

(look at the 'capture' chapters - starting around page 29)(we frequently work with the authors of this document - chi approved!)

http://www.blm.gov/pgdata/etc/medialib/ ... ote428.pdf

 

We have been using this software here with pretty good success:

(the more expensive professional version has got good metrics that the standard edition does not-shrug).

http://www.agisoft.ru/products/photoscan/standard/

 

Hope this helps!

 

marlin.

[jlutgen]

Great, that was helpful.

[kathrynpiquette]

Hi Jerry and all,

 

I know this is something Lindsay MacDonald has also been working on at UCL (http://www2.cege.ucl.ac.uk/staff/staffp ... taffID=814).

 

Here is a link to the paper we gave in Southampton earlier this year: https://www.ocs.soton.ac.uk/index.php/C ... r/view/606. We need to add our slides to the slideshare, which include a terrain map extracted from selected RTI 'original captures' pre-fitting. Hopefully we will get around to this in the next week or two.

 

All best - Kathryn

[cdschroer]

The issue with converting normal fields to 3D is that there is a small error in the calculus integration that accumulates over large areas. So, it is very accurate in the center and starts to warp over the course of the whole normal field. The solution for this is to combine normal fields with range data so that they can correct each other. Range data is highly accurate over large area and "noisy" over small areas (high frequency), while the RTI data is great in the high frequency but has this warping issue if you convert it. I want to make it clear that this warping is only an issue if you try to convert a normal field to a 3D surface, and is not an issue of RTI data left as a normal field.

 

Some important research on this was presented at SIGGRAPH in 2005 in a paper by Diego Nehab, et. al.

http://gfx.cs.prince...Nehab_2005_ECP/

 

Diego was a PhD candidate at Princeton at the time, but now is in a research lab in Brazil. We are looking at picking up this promising approach in collaboration with Diego. It is not currently funded work.

 

Carla

[jlutgen]

Thanks Kathryn and Carla. Interesting stuff. For my part, I will just follow the discussion and wait for progress. Hopefully in the not-to-distant future there will be a method to intergrate RTI and 3d into a single work process and viewer. In the meantime we will simply perform RTI and 3D photogrammetry alongside each other.

[GeorgeBevan]

Kathryn,

 

I`m really intrigued by Lindsay`s surface profiles (I think you put them up on your blog). Will Lindsay be publishing the algorithm he used to extract the profile from the surface normals? I'd love to test the accuracy of these profiles against another profilometry technique. Amazing stuff!!