Panoramic photos with AgiSoft

[disco_pihtija]

Dear All,

I have a question. I am a master student and trying to find something for researching.
I have one idea and I need your help because almost everybody her have more experience than me.

Is it possible to make panoramic photos using GigaPan and from that photos to make 3D model in AgiSoft Pro?

I am really looking forward for your replies and your answers.

Best

Veljko

[RalfH]

Difficult. I haven't tried it, but would not be too optimistic about it. Usually, people do it the other way around: use Photoscan to create a combined image (texture) from many photographs.

[disco_pihtija]

I just finished extracting and I can only agree with you.

[GPC]

These are two very different goals with very different requirements.

A. In panoramic photography (such as gigapan), the absolute goal is to keep the nodal point of all images the same. To minimize the differences in camera position so that while pivoting between images there is zero discrepancies in perspective.

B. In structure from motion based approach, the primary goal is to observe the same scene from a variety of perspectives so that the 3D geometry can be computed.

The two objectives are the opposite of each other. One is to be zero, the other is to be maximum. You may accomplish one or the other, but it is impossible to accomplish A and then B.

What you want to do is not possible.

[RalfH]

Hmmm, I wouldn't put it quite as strongy, cs. As you say, panoramic photographs are combined in a way that the nodal point is the same for all images. Therefore, in theory, a combination of several panoramic images taken from different positions could indeed be used in a structure-from-motion approach to increase pixel resolution whil retaining full area coverage. The problem I see is camera calibration for these panoramic images. And as one could easily take many images from many positions, the panoramic-to-SfM approach may not be worth the touble. There may be a use case if only very few positions were available for taking images while at the same time needing high pixel resolution and large area coverage.

[Wishgranter]

im have seen that Google experimented with google streets panos, they managed to create 3D models from the data.... will try find the PDF.....

for now just read this http://cipa.icomos.org/fileadmin/template/doc/PRAGUE/038.pdf will try find the Google streets PDF....

[GPC]

Ahh, I may have misunderstood his intention. If he is going to take several panoramics from several different locations, than it would be possible.. really no different than what the rest of us are doing. Although those panos often have stitching errors that would hurt the accuracy of the model.

I had interpreted that he wanted to make a model from 1 large pano from 1 position.

[tezen]

Hey!

Instead of using a TriPod (or GigaPan) just hold a normal cam in your hands and move it around you an take lots of shots (do HDRs!) with a stretched arm. Don?t forget to have enough overlap (2/3) and to take some photos of the point you were standing at. After photo-alignment PhotoScan will generate a 3D-Model. It even works if you move your cam a short way like a stretched arm in this example. With the 3D-Modelling-Software of your choice (I use ZBrush) you?ve to build insert a (mostly half-) sphere around the place you?re standing and scale it up to fake the background. Don?t forget to flip the normals of the sphere. With some modelling skills you can sculpt between spaces which weren?t calculated by PhotoScan. Reload the finished model into PhotoScan and generate the texture. In most cases it looks better than Modos HDREs ( example: http://www.luxology.com/store/kits/hdre/moofe/ ). And with floating-point images (HDRs) you can use that as a light-source too.

Greetings
tezen

[mwillis]

I know of at least one researcher who is making 3D models using dual GigaPan stereo pairs.

[RalfH]

And that person is... ?

[Wishgranter]

yes, we are interested what sort of results can be archieved.....

[mwillis]

And that person is... ?

George Bevan. A chap in Canada.  Send me a PM, and I'll share his email address, if you like.

[ChristianeZhao]

Hello,
Yes you can use panorama images as a source of input. You can look at the following photo I attached in this post.

If you are interested in knowing more about using panorama images as a source, feel free to contact me @ChristianeZhao in Linkedin.

You can also check out YouTube for some of our cases that are done by the panorama.
https://www.youtube.com/channel/UCXQvtIrbrpk6tdXgee7haoQ