EXIF sorting or Calibration Groups

Comments

3 comments

  • Avatar
    Götz Echtenacher

    1. Yes

    2. Good question, I don't get it either.

    3. Not sure what you mean...

    4. Hmmm?

    5. No, it just uses focal length since that influences the distortion most.

    6. Yes. Yes.

    0
    Comment actions Permalink
  • Avatar
    Tom Foster

    2. I have it - maybe.

    Looking at 1D views of a set of 23 photos after alignment, I see that what in Priors (camera EXIF data) is shown as

    FL 28; Principal point x 0.0; Principal point y 0.0.

    in Registration is shown varying between

    FL 29.186 to 29.394; Principal point x .124 to -.087; Principal point y -.012 to -.264.

    The Lens distortion Radials 1, 2, 3 show similar variation.

    In automatic Calibation Groups sorting, it seems that RC must work hard to deduce these figures for each photo - and gets answers quite scattered, and showing bias in one direction or another.

    By specifying authoritatively that the camera EXIF data is in fact correct, maybe RC is instructed to use the camera EXIF data instead of spending time and resources working it out for each photo from scratch. Even if all photos are at same zoom on a single camera.

    It may be that the camera gives EXIF data that is itself is inaccurate (e.g. FL28 could mean anything between 27.51 and 28.49) but at least it's consistent from photo to photo, which might improve triangulation of projections.

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Hi Tom,

    you are right in so far as the number the exif gives as focal length is only a rough guideline.

    But it is good enough for RC to determine that everything eith the this same number will likely have very similar distortions. So it treats ALL images in one group as the same, using the info of ALL the images in the group which ic much better thatn doing it individually for each images, since they often include areas with low or no features at all. So doing that will ensure that the alignment (in terms of the un-distortion of the images) will be very close to perfect. After one is satisfied with everything, one can then ungroup everything and align once more (WITHOUT deleting the previous alignment) and the RC will slightly adjust each image so it will reflect slight differences.

    When you align without grouping, difficult images might reault in a wrong un-distrotion model which can lead to problems in the geometry or even break-up of components.

    0
    Comment actions Permalink

Please sign in to leave a comment.