No project output coordinate system on pointclouds with images aligned to laserscan

Answered

Comments

13 comments

  • Avatar
    Zuzana Ďuríčková CR

    Hello CG,

    in the last screenshot the selected component is indeed not georeferenced. It seems that this component was created only from images which do not contain GPS coordinates. If this is the case, then RC does not have the information about the real-world position therefore could not georeference it. For non-georeferenced scenes only the grid plane option is available during export. 

    You can find some tips how to merge scans and images together in many threads on this forum, for example:

    https://support.capturingreality.com/hc/en-us/community/posts/115000783911-Align-laser-scan-with-images

    0
    Comment actions Permalink
  • Avatar
    CG

    Thanks, I'd read those threads but I figured the resulting component should be georeferenced because the laser scan is. I just want the output to use the same coordinate system as the scan and effectively use the scan as control since it has already been registered to GPS control in another program. How can I do this? 

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    The laserscans and images need to be aligned in ONE component for that to happen. Components have no connection, they are independent results, as large as RC can put them together. Is that it?

    0
    Comment actions Permalink
  • Avatar
    CG

    Right, I think I see my problem. The scans are not colorized so the images are not aligning to the scan. Some of the threads in the forum suggest that using tie points might be the only way to solve this (photos were not taken on board the scanner) - is that true?

    In general, does RC create images from the laser scans and use those as essentially just another photo, or are the images aligned to the z,y,z points in the scan pointcloud?

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    "In general, does RC create images from the laser scans and use those as essentially just another photo [...]?"

    Short answer: yes!

    There are a few posts and threads discussing that...

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Althought I would leave the "just" away - the precise spacial info is not lost in the process, just the mechanism of aligning scans and photos works via images rather than creating independent pointclouds and then aligning those.

    0
    Comment actions Permalink
  • Avatar
    CG

    So there is no way to use laser scan intensity only to align images to a scan? Tie points are the only way in this case?

    0
    Comment actions Permalink
  • Avatar
    CG

    I couldn't make out the answer clearly based on the existing threads. One thread seemed to suggest if I use control points to roughly align the scan with the photo alignment, RC would figure out the rest but another said it was not possible.

     

     

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    That's because it depends on so many factors!

    The more precise a question, the easier it is to give an answer - yours was rather broad, that's why I just sent the search-link. The general guiedeline used to be (I did not check the help section to verify if still the case) that a scan needs to contain RGB color for RC to work properly. Since scans are being converted into raster images, that makes sense. There are cases where an alignment was successful with using intensity only. I guess that would be possible if the intensity scale results in something similar enough to an RGB that RC can identify enough matches. If what you want could work needs to be tried by error. There are several possibilities but if I walk you through all of them it would probably be twice as fast if I did it because I would have to try it out myself...   :-)

    In my experience Tie Points only work if there is enough for RC to latch on to after a nudge in the right direction. I think there is other software out there where you can basiyally "make" an alignment only with manual points, but I did not get the impression RC works like that. In most cases it would be completely inefficient to do it like that anyway...

    0
    Comment actions Permalink
  • Avatar
    CG

    So I wasn't having any luck trying to match control points between the images and the scan since I was having trouble identifying common features accurately (the site is a cliff face). However, the images had georeferenced control, and my scan was georeferenced, so I toggled the 'merge georeferenced components' option in the alignment settings and successfully aligned the two that way. 

    The result was interesting though - it turns out one area of the photos had some bad control which caused it to deviate from the point cloud. I can see areas where RC managed to align the two components are very accurate (there is no doubling and the noise is lower than if I just had images), but where there is shadow in the scan pointcloud, eg. under the scanner or where RC couldn't find tie points, the resulting model suddenly rises up to where it would be without the scan data. I could try setting the weights of the photo control lower so that it would use the scan to align more than my bad GCPs but it would seem that alignment is possible without a coloured scan, it's just not ideal.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Very interesting, thanks for the feedback!

    How about deactivating the faulty coontrol? You also might have to delete older components since those influence the new alignment, sometimes preventing better results.

    0
    Comment actions Permalink

Please sign in to leave a comment.