'Do it yourself' photogrammetry?

edited March 2010 in Technique
Hello All,

I'm toying with the idea of writing software to orthorectify images from different viewpoints and and extract 3D info from them. I know there IS such software, but I'd like to use only (free) open source stuff... Plus it's a great learning experience to try things for yourself :-)

The best component I found this far is OpenCV, a software library for image processing. It has functions to transform images (e.g. for perspective), and has interesting articles and code for lens calibration / correction.

Here's the questions:
Has anybody experience with OpenCV?
Is there other good open source sw for photogrammetry?
Are there KAPers who are writing software to process their images?

Cheers,
Hans.

Comments

  • Hi Hans!

    I had a bit of experience with OpenCV, it's a great library!I toyed around with the same idea (extracting 3D info) and I was impressed with the results. My plan is to give Qt (and PyQt4) a try and make a nice front-end for the softwares I write (most of them I just run from a console).
  • Hansel, I do not have experience with OpenCV nor am I writing photogrammetry software. I do use Photomodeler Scanner to create low-baseline stereo image models. The software works very well, and I've found it rather forgiving. It will still process image pairs that are not perfectly parallel or where the base to height ratio is less than ideal. Unfortunately, the software is not cheap. I purchased my copy under a grant.

    E-foto is a package that might be worth looking into. I've only toyed with it, but it seems pretty stable. There might be some ideas in there for you.

    @Femur: Thank you very much for the heads up on ZPR! I'm going to check this out right away. It is new to me.

    I've been really impressed by structure from motion (SfM) software like Photosynth and Bundler. On this site, there is an ongoing discussion regarding SfM. Bunder, I'm finding works very well. However, points for the point cloud are derived from SIFT. I'd like to see the ability to have greater control over SIFT's selection of key points. There are several panorama packages that also use SIFT, and these packages allow a user to specify an area on a pair of images that the algorithm should search. Hugin and Autopano have this functionality. With both packages, the entire image is analyzed by SIFT but then the user can manually specify areas where SIFT should look again. I would like to be able to have additional searches made for key points that are on features of interest. For me, SfM has real advantages for both KAP and PAP. Models can be built by different people with different cameras. With respect to camera geometries, the algorithms are very forgiving. However, if orthophoto production is the end goal--SfM is obviously not the way to go. Yet, in terms of making a nice "relatively" high resolution DEM, SfM strikes me as really good (and lots cheaper than a laser scanner).

    Hansel, I'm really looking forward to hearing more about your project.

    Regards,
    Nathan
  • Hans,

    I've toyed with the same ideas but haven't found the time to try it yet. I'll be interested in hearing about your progress.

    Dave
  • Are there any tutorials on using ZPR? Seems like a really good package to use in conjunction with Photosynth point clouds.

    Tom
  • Hi Tom,
    Unfortunately the only info is the built-in help, all the documentation is in greek and on the author's thesis, anyway Stefanos (the author) is quite a nice person and offered to answer any question by email.

    Regards,
    Mauro
  • edited March 2010
    Hello Hansel,

    Although I haven't gotten around to using OpenCV, I have heard some good stuff about it in other forums. It can extract 2D and 3D images and can run on all OS. Plus, it's made by Intel so must be good. contact lenses
  • @Nathan ... can you point me a good web tutorial (if there is any) on how to make DEM's from Ortho KAP pics with Photomodeler Scanner?
    Thanks!
    Mauro
  • Here You can find same examples of photogrammetry from Image Master software but it's no cheap :(
    http://ptolemais.pl/index.php/en_EN/modele-vrml
    http://www.youtube.com/watch?v=hCkdyP8uyCM
  • Hi All,

    Update...

    I looked at Ephoto, but didn't manage to get usefull results wit it. Besides, I believe it is a dead project? I do not want to spend serious money on photogrammetry (yet). I do not need it professionally, and have other priorities privately...

    I had some time to experiment with Opencv: really great stuff (if yo are a software engineer). I wrote a small program to calibrate my camera (correction factors for lens distortion) based on some photos of a chessboard, and another small program to remove the lens distortion from photos. It just works. I'm now experimenting with perspective transformations to 'fit' photos taken form different points of view. Its not photogrammetry yet, but it's a start. The opencv library has functions to 'handle' stereo photographs.

    I'll post here if I learn more. Please let me know if you want to join the fun.

    Cheers, Hans.
  • Thanks for the update. I agree that e-foto is probably not active. I never made much progress with it either. I'll very much look forward to hearing about what new developments you come up with on the Opencv front.

    Regards,
    Nathan
  • I dont know about OpenCV.
    I am using micmac. this is an open-source software with a dynamic and serious developping team. There is a forum where most of posts are in French, but virtually everyone there will understand English.
    The leaning curve used to be steepy and slippery but things are going much better month after month.
    The algorithms behind are rocket-prooff.
    Olivier
Sign In or Register to comment.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

In this Discussion