Shooting a subject from all directions, similar to cartography

[piccolbo]

Hi,

I was thinking recently about all the different techniques that are made possible or more powerful by combining multiple images (panorama, hdr, focus stacking, etc) and it seemed to me one had not been tried yet: shoot the same subject from many directions. Sounds like the dual of panorama photography, whereby one shoots in many different directions from one position. Kind of swapping the photographer and the subject. Like most ideas I come up with, it's not new. In fact it's about the same, geometrically, that's done when mapping the globe to a surface as in cartography. Not even new in Hugin circles, see https://groups.google.com/forum/#!searchin/hugin-ptx/globe/hugin-ptx/kju8CPAgOOY/TUfB8AAHqJwJ. Despite doing my best to follow the advice in that thread and plenty of experience with panorama photography in Hugin, I end up with results like this:

This is my process: The object, a fruit bowl, is on a stool with a dark background, camera on a tripod. Shot 17 pics at approximately regular angular intervals by rotating the subject, not moving the camera. The distance remains consistent this way. Light is soft, a little stronger from the left. Adjusted curves in Aperture for all pics, imported in Hugin. Use cpfind to find control points with option --linearmatch to reduce spurious finds. For most pairs, control points are on the bowl, for a few on the background and support. Fix manually. Add  horizontal line to each frame to avoid snake like stitch. Optimize positions first, then release a few parameters at a time to see what happen. Image above is obtained by optimizing everything. a, c, d,e, g are nontheless 0, TrX, TrY, TrZ almost 0. Plane yaw and pitch are mostly low, with a couple of outlier images. Worst distance is 25, most < 4.  I am aware that parallax changes will make it challenging to map the fruit in a continuous way. I am just focusing on the bowl and no cps are on the actual fruit. Does anyone understand what is going on, let alone how to fix it? Thanks

Antonio

[Yuval Levy]

On 15-05-03 06:43 PM, piccolbo wrote:
> <https://lh3.googleusercontent.com/-iyuyJUGhLGU/VUafJWL6qNI/AAAAAAAAON0/yvlmQmuSgEI/s1600/_1070218%2B-%2B_1070238.tif>

> Does anyone understand what is going on, let alone how to fix it? Thanks

Ciao Antonio,

the described process would only work if your bowl was a perfect cylinder.

there are a few possible solutions, which all boil down to simulating a
line scan camera. Not difficult, giving your controlled set up.

The simplest solution that comes to my mind: take a lot more pictures at
shorter angular intervals. Run them through an imagemagick script to
slice a few pixels from the middle. Try with narrower or wider slices
until they match well enough.

Yuv

[Jim Watters]

On 2015-05-03 7:43 PM, piccolbo wrote:
> This is my process: The object, a fruit bowl, is on a stool with a dark
> background, camera on a tripod. Shot 17 pics at approximately regular angular
> intervals by rotating the subject, not moving the camera. The distance remains

> consistent this way. [snip]

> Add horizontal line to each frame to avoid snake like stitch. Optimize
> positions first, then release a few parameters at a time to see what happen.
> Image above is obtained by optimizing everything. a, c, d,e, g are nontheless
> 0, TrX, TrY, TrZ almost 0. Plane yaw and pitch are mostly low, with a couple
> of outlier images. Worst distance is 25, most < 4. I am aware that parallax
> changes will make it challenging to map the fruit in a continuous way. I am
> just focusing on the bowl and no cps are on the actual fruit. Does anyone
> understand what is going on, let alone how to fix it? Thanks
>
> Antonio

I would add more horizontal control points to force the edges to be a horizontal
straight line. I would expect some large a,b,c numbers to distort each image.
Constraining the image to be a cylinder 360 by joining the L & R will help with
the optimization. After you can add end images. Or remove images to give it an end.

Any control points away from the seam of the two images should also be changed
to horizontal control points. This is not to make them a line but to throw away
the horizontal shift that would happen because of the parallax error and only
use the vertical shift to optimize.

I have been using horizontal control points in a similar way to help stitch
stereo images from multi camera rig.

Not sure if you should need any TrX, TrY, TrZ I would only add them at the end.

There is enough repeating pattern that automatic control point generation would
cause problems.

--
Jim Watters
http://photocreations.ca

[Erik Krause]

Am 04.05.2015 um 00:43 schrieb piccolbo:
> Like most ideas I come up with, it's not new. In fact it's about the
> same, geometrically, that's done when mapping the globe to a surface as
> in cartography.

Igor Marx did this with a full spherical strawberry:
http://www.360cities.net/image/strawberry-panorama
but I suspect that this wasn't as hard as your fruit bowl since the
structure is much more forgiving. Some leaves are missing also.

We tried this many years ago with a painting on a pipe head, which was
roughly ellipsoidal, without much success. The point is: For anything
but a cylinder or a sphere you need specialized mapping functions that
are different for different directions. In hugin / panotools you can
only get approximations if you use many narrow stripes.

Best would be a slit scan camera and a slowly rotating turntable. If you
have a Canon EOS camera chances are good that Magic Lantern runs on it.
You could try one of the nightly builds where you find a slit scan
camera in the silent picture module. There is an IPhone app also:
https://www.youtube.com/watch?v=csSDpzXwvRI
as well as for Android:
http://de.appszoom.com/android_applications/slit-scan

--
Erik Krause
http://www.erik-krause.de

[Mark Phillips]

There are many web sites that allow one to upload 6-8 pictures and the software will create the 3D image for you. There is also some open source software you can download - http://insight3d.sourceforge.net/insight3d_tutorial.pdf. I have also seen a tutorial for Gimp, where you can create an animated gif based on exposures from different angles to show the object rotating on the page.

I am experimenting with these options for a family memorabilia project. I am capturing 3-D images of certain objects such as trophies to display on a family web site. I had hoped Hugin would be able to create these images, but I didn't find any useful tutorials. I assume from the previous posts that the software cannot do this.

Mark

--
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+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/5547E7AE.1060505%40gmx.de.

For more options, visit https://groups.google.com/d/optout.

[piccolbo]

Thanks everybody, that pretty much settles it.

Antonio