Coded Targets and Auto-Detection ?

Answered

Comments

6 comments

  • Avatar
    Wishgranter
    Hi Götz

    Are you going to implement coded targets with auto-detection at some point?
    Yes we will implement Coded targets in our application, its a planned feature

    Personally, from what i've seen so far the algorythms (also of other recent software) are so refined, that an impressive accuracy can be accomplished without the targets. I am just working on a biggish project of an abbey choir (ruins) an when I overlayed the result manually (screenshot and scaled in CAD - how professional !) with a ground plan (theodolite survey), the maximum difference was about 1.5 cm (over 40 m or so), which I find absolutely sufficient for the purpose. In some cases, I would rather trust the model than the survey...

    1,5 cm is a good value, but its a "metrology" so it need to carefully plan to get correct results with acceptable errors

    However, I suspect that especially in my area the peers will expect some sort of more or less densely distributed targets with real coordinates to make the model appear more accurate. From the described experience I would think that really 3-4 targets with coordinates should be sufficient to orient the model in "real" space.
    Yes but again it depend on project size-scale-complexity. for small size project 3-4 should be enough for larger and mor ecomples need have more of them.
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Hi wishgranter,

    I know from your numerous and helpful contributions in the photoscan forum that you are quite deep into the theories but still have a practical grasp on things (including built heritage). So I guess I was aiming a bit for a (brief) discussion about the stability (can one say that?) or accuracy of complex models. I am aware that one can't expect miracles from a model of 3-4 shots of a plain surface, but from what i've seen so far it seems that with a good coverage of a complex geometry with a redundancy of say 20-30 cameras for most of the surface - also from different angles - the model seems to be pretty reliable. Again, I am not talking about the fringes of the scene and also I do not aim for the accuracy of automobile or aironautics industries. ;-)
    0
    Comment actions Permalink
  • Avatar
    Wishgranter
    Hi Götz

    if want have the best possible accuracy, then need add the laser scanner data as its metrological calibrated device.

    and yes as you say its better to have more images for the subject as it will better conditioned = captured from more angles can improve the alignment "stability".

    its a complex stuff around getting "good results" and have a idea of training sessions over the next year. because it need a bit of explanation + real experience...
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Hi Wishgranter,

    this is getting interesting! :-)
    I think it is an interesting approach to combine laser and image data.
    Because so far I was thinking of the two methods as rivals.
    I think that is also caused by the fact that image based models are often advertised as a low budget laser scanner alternative.
    Reading through the photoscan forum and my first steps quickly raised doubts about this general statement though, because the hard- and software requirements for getting good results from image based scans are also significant.
    I do think that the image based approach has several advantages though:
    - it is easier to improve your model bit by bit (fill in gaps)
    - also its easier to get small corners covered due tothe bulkyness of the scanners
    - you have to deal with the object much more closely than when you take a laser scan (that is only an advantage for people who also do research with the data like I do)
    - you can influence the result while you are working by getting closer or taking more photos - so it's more intuitive
    - initial costs are lower, so a step by step improvement is possible

    But from what you said it could be also a good idea to get a rough laser scan from few standpoints and then improve from that on. On the other hand, I can do a feature based theodolite survey (control points) and achieve the same.
    I just had another idea which i will post in feature request... :D
    0
    Comment actions Permalink
  • Avatar
    Wishgranter
    Hi Götz

    take a look on a nice example, its not ideal, but we can show here the benefits of the fusion of the data

    https://dl.dropboxusercontent.com/u/150 ... r_foto.mp4
    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher
    Hi Wishgranter,

    wow, especially the first one with the stairs ist quite convincing.
    However, laser-scanning just isn't an option for me at the minute... :cry:
    And I maintain that with enough images, one can make up for some of the gaps.
    Any comments on my other points?
    We should maybe get some real work done though... :lol:
    0
    Comment actions Permalink

Please sign in to leave a comment.