manual improvement of alignment / sparce point cloud

Answered

Comments

24 comments

  • Avatar
    chris
    +1

    i agree better tools for this would be helpful.
    0
    Comment actions Permalink
  • Avatar
    chris
    I'd like to see an align selected cameras as well
    0
    Comment actions Permalink
  • Avatar
    ssh4
    Control points (Tie Point) is created for this.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Vladlen, I DID use control points!
    Actually, I am becoming quite good at using them.
    But it is VERY time consuming and if then the result is not even close to the competitor, where it took considerably less time in spite of the extremely longer calculation times, then I think it is reasonable to ask if such a tool would not make sense in RC.
    I think it could speed up overall processing times (the manual part) a LOT.
    And why limit the fantastic speed of RC only to the automatic processing?

    I would even be willing to switch back and forth between different programs, but I tried and tried and did not manage to figure out how. I think the aligning process is just too integrated into the cores of the different programs so that it isn't possible to align it in one and then take tie points / spares point cloud to another one to continue...
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    chris, what do you mean by align selected cameras?
    Add those to the largest chunk?
    Because you could just deactivate all the others, but I'm pretty sure that's not what you mean...
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    No comment at all from a member of the RC-Team?
    0
    Comment actions Permalink
  • Avatar
    Wishgranter
    Hi Götz Echtenacher

    No comments as Vladlen say it, if images are not properly aligning, then you need to use TIE points to help the alignment. Just properly placed on enough images and its work. Sometimes you need palce 10´s of them, depending on scene composition.
    0
    Comment actions Permalink
  • Avatar
    Vektra
    Is it possible to implement the option of displaying user tie points on all photos (as Agi) so that we see a potential error immediately? :)
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Yes Wishgranter, that's exactly my point.
    10s of tie points take time, sometimes considerably.
    Gradual selection doesn't.

    What I mean is at least a comment of WHY it will not be implemented, if it is not possible with your algorythms or too complicated.
    Because otherwise this question will just be asked over and over again (not neccessarily by me :-)).
    If there is a good reason, it is easier to accept...

    @Zlatan:
    That is already possible - look in the blue IMAGE ribbon which pops up when you select a 2D view.
    There are several options including what you are looking for...
    0
    Comment actions Permalink
  • Avatar
    Vektra
    Götz Echtenacher wrote:
    That is already possible - look in the blue IMAGE ribbon which pops up when you select a 2D view.
    There are several options including what you are looking for...


    If you're thinking of ray of sight, that's not what I'm asking :D
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    No that's not what I meant. :D
    There is also "Tie Points" which, if I interpret your question properly, should be exactly what you are looking for.
    If not, look at "Show Matches" and "Residuals", both helpful tools.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Ok, I shall make another attempt.
    Maybe somebody from RC will have mercy... :-)

    Manual alignment optimization will

    [ ] be available at some point in the near future.
    [ ] be considered, but is difficult to implement and there are more pressing issues.
    [ ] never be implemented because we firmly believe that it is easier for us to teach and users to learn optimal image taking.
    [ ] oh look, a Giraffe!
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    I'm not giving up!

    I realize the setting "max repro error" ican be used to influence the results but that is too much trial and error and it often results in component split.
    From trying around a bit I got the feeling that the setting does not influence feature detection but is essentially a filter after that step.
    Is that correct?
    So should it not be possible to incorporate some filter for the tie points after initial alignment?
    0
    Comment actions Permalink
  • Avatar
    Lucia CR
    Well, one of our experts recommends you to keep adding control points (2D-2D at least 10, 2D-3D at least 7), and if it does not help, increase the point weight... So the manual alignment optimisation is actually implemented.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    CPs were enough. If I add more I might become a case for an institution...
    And same argument - CPs takes hours (worst case) and a slider takes seconds!
    My cases are very often small scale historic buildings and trying to find suitable spots for CPs aon irregularly weathered stone or plaster can be quite frustrating. I am thrilled when there is fancy bird poop to target. :mrgreen:

    Ah, point weight. I've managed to avoid that one so far - at the time it wasn't covered yet in the help section.
    But I'll give it a try next time.
    What's the range that makes sense - 10-20 or even 100?

    And sorry for the next one, but what do you mean by 2D-2D and 2D-3D?
    You mean each CP in 10 images and 7 CPs overall?
    0
    Comment actions Permalink
  • Avatar
    Lucia CR
    Götz Echtenacher wrote:
    CPs were enough. If I add more I might become a case for an institution...
    And same argument - CPs takes hours (worst case) and a slider takes seconds!
    My cases are very often small scale historic buildings and trying to find suitable spots for CPs aon irregularly weathered stone or plaster can be quite frustrating. I am thrilled when there is fancy bird poop to target. :mrgreen:

    Ah, point weight. I've managed to avoid that one so far - at the time it wasn't covered yet in the help section.
    But I'll give it a try next time.
    What's the range that makes sense - 10-20 or even 100?

    Aligning a spherical globe held in foam
    Aligning components breaking up
    See also 'Control Points' tutorial in the software Help.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Hi lubenko,

    The help section I know of course!
    There is only a general explanation.
    Cheers, didn't know those two yet.
    Helpful.
    Why not put that into help? :D

    I did try it already and found that it can be crancked up to 200 in difficult cases.
    That screws the whole rest of the model up but at least it helps enormously to identify errors in CP placement.
    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant
    Thanks, useful informations here !!!
    I agree, a feature to filter tie points after alignment (like in PS) would be much less time consuming than this trial and error game with the max reprojection error.
    And by the way, what is the good value now (it used to be 3, now I guess this is 1 ?) ?
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Hi Jonathan,

    What do you mean by "good value"?
    The preset? That would be 10.

    The thing is also that I find that I am using up a big portion of the time gained by placing those CPs.
    If it weren't for the huge added benefit of much better resource management... ;-)
    0
    Comment actions Permalink
  • Avatar
    Lucia CR
    Götz Echtenacher wrote:
    ...what do you mean by 2D-2D and 2D-3D?
    You mean each CP in 10 images and 7 CPs overall?

    You can place control points (CPs) among 2D pictures without 3D coordinates. So the recommendation here is at least 10 CPs when placing them just in the 2D, and at least 7 CPs when placing them also in the 3D.
    0
    Comment actions Permalink
  • Avatar
    ssh4
    To be honest, for fix topicstarter issue in this project will be enough group by EXIF cameras, align several times for refine alignment and use downsampled 2-4 times images in meshing. And result will be near the same (same but better) as in Competitor.

    Also better do not touch weight of control points. default 10 is good for 99.99% of projects.
    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant
    What do you mean by "good value"?
    The preset? That would be 10.

    I mean the "max feature reprojection error" on alignment. I am usually putting it between 1 and 3.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Oh, that one.
    Two.
    Factory settings can be restored by the upper of the tiny buttons on the lower left of each settings window (plus save and restore customs).

    In my (topicstarter) project, the max reprojection error I get after cleaning it with the competitor is 0,5.
    When I use that in RC, I get several components.
    So no good.

    I usually use normal with factor 2.
    More will not help because I will loose detail in the corners.

    I have not tried to use exif grouping yet because in my cases it often does more damage than good.
    But I shall give it a try when I get around to it again.
    0
    Comment actions Permalink
  • Avatar
    Bodo S.

    Hello everyone,

    using CR for survey-grade scanning, I'm very interested in making sure, that accuracy is as best as possible in all captured areas.

    Having control over the TIE-POINTS which are used for camera alignment and being able to manually take some tie-points out -either by hand/lasso selection or via a tool that allows gradual deletion of tie-points based on their estimated precision value- would make a lot of sense IMHO.

    Two examples:

    In one scene, there is occasional car traffic around a large industrial building. No chance to not have them on some of the lets say 2.000 images in the set during a couple of hours of aerial flight. Those things are appearing, moving and disappearing, leaving tie point fragments here and there, which I think, should not be used for global alignment. I wish I could manually delete those tie points with the lasso tool.

    In another scene there is a lot of vegetation while the wind is blowing quite a bit. Now I would rather want the cameras to be aligned to static points than permanently moving foliage. I wish I could delete those green points, based on a colour picker tool with some settings for the inclusion of similar points (tones of green).

    I'd like to know, if my assumption is correct when I think that manually cleaning the tie-point-cloud can lead to higher accuracy i.e. via more realistic camera calibration.

    What do the devs think about that?

    Best regards, Bodo

    0
    Comment actions Permalink

Please sign in to leave a comment.