It is possible to stitch a 3D panorama from as few as 4 views taken with a stereo pair of cameras. I routinely shoot them 6 around, never more than 12. The keys to success are, first, to align the the views 'at infinity' -- that is, to the true direction and roll of the views -- and second to warp the images to fit, warping the left and right images of each view as nearly the same as possible, so that stereo disparities are preserved in the pano pair.
Alignment at infinity can be approached in several ways. Perhaps the best is to use a stable high-precision rig that reproduces camera attitudes to subpixel accuracy, and create a master alignment template used for all stitches. But that is very expensive and technically demanding. The same idea can be applied with ordinary equipment if you mount one camera on-nodal and use that image series to generate a template. But this doubles the error in the other sphere, compared with having the axis halfway between the lens pupils. Peter Murphy (a master) often adds a third, small on-axis camera to a symmetric pair, for the sole purpose of generating an alignment template. Another way is to try to select a set of control points that makes a good alignment template. This will contain only points on distant parts of the scene. You can often make it better by combining control points from left and right cameras. Such a CP set is not so good for the warp-to-fit step, but the goal here is just to align the views correctly. For warp-to-fit, I use all the CPs linking adjacent images (but not more distant ones) and apply PTGui's viewpoint correction to all images, with the y,p,r alignment locked at the template values. I'm not current with what Hugin can do along these lines, but I'd guess it may be even more capable. One could use the template idea here too, by actually combining all left and right CPs and saving out a warp template to be used for the final stitches. I have developed a helper program for PTGui, called PT3D, that automates this kind of control point fiddling and combines both fine alignment and warp-to-fit in a single step that processes left and right spheres together. It usually gives alignments that can be stitched successfully if you put in enough effort on blend masks. If you have PTGui and are interested to try it, it is available under commercial license at http://panini-pro.com/PT3D/ At the outset I designed PT3D to be compatible with Hugin as well, but by now it has many specific dependencies on PTGui. However I am open to the idea of an opensource version specifically for Hugin, if someone wants to develop it. On Monday, April 25, 2016 at 3:13:51 PM UTC-4, Erik Krause wrote: > > Am 22.04.2016 um 15:24 schrieb Simon Bethke: > > I am not saying that non-NPP issues are a shortcomunig of Hugin. I am > > saying, that morphing the controlpoints in a way a texture is mapped on > a > > wireframe model (using baricentric coordinates in a triangle) will allow > > stitching non-NPP panoramas without visible seams. > > Thomas Sharpless (inventor of the pannini projection) is attempting > something like this: > http://ivrpa.org/news/practical-stereo-stitching-with-pt3d-quebec-2016/ > > -- > Erik Krause > http://www.erik-krause.de > -- A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ --- You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. To unsubscribe from this group and stop receiving emails from it, send an email to hugin-ptx+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/40510965-5346-4827-a9c4-4a84cc60726b%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
