Overlay two Equirectangular Panoramas - Registration of two Panoramas
- I need to make Camera panoramas Match Laserscanner panoramas in the best
way....that means overlay the two equirectangular and expect a good
match [2-3px error with 20000px=360°].
At first I tought it was just a matter of translating... so we developed
an average-rigid-move... but it doesnt work well at all. Our original
panos are 20000x10000 and overlay accuracy is around 10-15px with this
I wonder if the matter is the equatorial line position that maybe
differs between scanner equirectangular and camera equirectangular.....
I can't determine where's the equator... having experience in Panos
maybe you can really be of help in addressing what's the problem.
I would even pay a consulant for this if the price is cheap and it will
make me reach the goal.
Half resolution panos that should match are here:
- PTGui is able to align the images reasonably well, though not perfectly, by assigning individual lens parameters to one of the images (the scan image). A vertical shift is indeed necessary and this amounts to around 347 pixels at the supplied resolution. The match is good in the central region but less so in the upper and lower areas. I've uploaded a zip file containing the ptgui project file and remapped scan image here:John
- Am 02.05.2014 20:18, schrieb Paolo:
> I need to make Camera panoramas Match Laserscanner panoramas in the bestIf the camera panorama is stitched of several images it probably would
> way....that means overlay the two equirectangular and expect a good
> match [2-3px error with 20000px=360°].
be better to use the laser scan as a boilerplate and stitch the camera
images to the laser scan. In PTGui this would require to set control
points manually, but with a well calibrated lens two to four points per
image should suffice.
- Thanks for your files, they work well... but it was the camera to be adapted to the scan and not vice-versa.
Even having your pts file I can't invert the thing..... Could you be so kind to give a complete set of files to do the vice-versa operation?
I did like this: I opened your pts with PtGui, swapped checks between "Image 0" and "Image 1" at the Lens Settings tab.... did the same at Optimizer Tab... run "Optimizer" -> Error too high.
Could this be done with Hugin instead of PtGui? Hugin keeps crashing here.... dunno if I am doing something wrong or Hugin uses to crash normally.
Thanks for your support
Il 03/05/2014 09:45, j.houghton@... ha scritto:
PTGui is able to align the images reasonably well, though not perfectly, by assigning individual lens parameters to one of the images (the scan image). A vertical shift is indeed necessary and this amounts to around 347 pixels at the supplied resolution. The match is good in the central region but less so in the upper and lower areas. I've uploaded a zip file containing the ptgui project file and remapped scan image here:John
- Here's the project file with the camera image aligned to the scan image.However, the hexagonal hole at the nadir suggests that camera image may be stitched from 6 shots around plus a zenith image. As Erik suggested, if you have these camera images to hand, you should get a better result by including the scan image in their stitch and aligning the images to the scan image as well as to each other. Then output a stitch with the scan image excluded.John
- Correction: Sorry, There was an error in the project file I uploaded. An amended version is at:http://ge.tt/3iBIvJg1/v/0?cJohn
- Hello John,
thanks for your effort and your support. You were really precious to let me start working with such matter.
My task is creating a procedure to adapt/remap camera taken panoramas over scan panoramas obtainer with LaserScanner equipment.
Now I am leaving PtGUI and going straight to PToptimizer/nona so that I can embed the procedure in our software.
The image that I've sent you earlier goes flawlessly with the following procedure while other panos are not working well!
The first phase of procedure is filing an optimizer.txt script with a correct initial data like detailed on following lines.
I still didn't understand how could you calculated "e" (vertical shifting between centers) parameter, I am doing like this: e=(Camera_H-Scan_H)/2... is it correct as initial value for PToptimizer?
I am giving a starting yaw calculated like this:
- calculate average of X deltas between scan.jpg and camera.jpg
- yaw=pic_width_for_360 * 360 / avg_x
then I am giving a starting pitch calculated like this:
- calculate average of Y deltas between scan.jpg and camera.jpg
- pitch=pic_height_for_180 * 180 / avg_y
I am giving a starting roll of 0. I would not know how to calculate it at the moment.
Then I fill the optimize.txt file with "c n0 N1 x... y... X... Y..." lines and on footline i leave a "v y0 r0 p0 e1" so that "e" parameter is optimized too.
After this procedure I run the PToptimizer over the optimize.txt file and I go down to catch the first line starting with "o f4 r......" and with that data I fill the stitch.txt file at both 'i' and 'o' lines.
Feeding nona with stitch.txt I create the remapped file that I would use to overlay over scan.
If you overlay with a semitransparent layer camera-remap to scan you can see there are big areas that simply don't match...... like around the nearest black/white target and the big texts on half way to the ceiling.
I've posted files here: http://clouddesigner.it/match2/match2.zip
I wonder if you can help me address where is the issue, match-procedure, camera setting or what?
Original pictures that originated the camera pano were taken with an 8mm Samyang, tripod and nodal ninja mkII.
I'm really thankful for your collaboration
Il 07/05/2014 10:03, j.houghton@... ha scritto:
Correction: Sorry, There was an error in the project file I uploaded. An amended version is at:http://ge.tt/3iBIvJg1/v/0?cJohn