Hi Pit On Feb 26, 4:03 am, Pit Suetterlin <p.suetter...@royac.iac.es> wrote: > Hi Oskar, > > I tried the several-lines method for calibrating my lens, and it indeed worked > quite nicely. > > Oskar Sander wrote: > > > sorry for waking up an old thread. > > > I'd like to use this method to calibrate a UW camera set-up. That is, > > a camera in a housing with a wide angle lens attached on the housing. > > Obviously this set-up is very much different from the camera on land > > so I need to do it in the pool, so I need to plan the exercise in > > beforehand. > > > The pool have a lot of natural lines to use in the tiles of the pool, > > the question is if these would be good enough, or if I must make my > > own lines using a little buoy with a weight and orange string to get > > enough accuracy? > > My experience was that the lines from the spacings between tiles look nice > until you start and use them for setting the control points. The exct > placement gets a bit uncertain and thus you get a higher noise which falls > back on the derived parameters unless you have many points. > I.e., if you use them I'd make sure to take a lot of images with lines going > through the FOV at different positions and in different angles. If I remember > correct I was using something like 10 images, defining some 30 lines in > there...
Did your calibration image set cover 360 degrees? It seems to me that if you use multiple images for calibration, then the correct focal length or fov becomes a critical parameter (which is not the case for single-image straight line calibration). And it is known that the PT optimizer will almost always choose a wrong value for that unless you force it to be right by insisting on closure of a full circular image set. But the straight line control point optimization (as currently implemented) requires that the output projection be rectilinear. So naively I would suppose that only a part of your straight line data would be usable, unless the rectilinear error is computed separately for each image, ignoring the rotation that aligns it on the panosphere? Some time ago I looked into the possibility of making libpano optimize straight lines on the panosphere (where they become great circles) instead of in a rectilinear projection. My hope was to better support calibration of fish eye lenses, which is a continuing problem. My code gave worse results than the existing method, so I gave it up. However I still think Hugin needs an easy and reliable way to do straight line lens calibration. I believe that after many years of using various calibration systems, photogrammetrists finally decided the straight line method was best. And they often use naturally occurring straight lines rather than special calibration rigs. The key is software that can follow lines and estimate their positions to subpixel accuracy. The raw image of a calibration line will in general be curved, so a human has to designate which lines are straight in reality -- but not set dozens of control points on them. Maybe this would make a good GSOC project. Regards, Tom --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ To post to this group, send email to hugin-ptx@googlegroups.com To unsubscribe from this group, send email to hugin-ptx-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/hugin-ptx -~----------~----~----~----~------~----~------~--~---