Sorry, an error occurred while loading the content.

## RE: [PanoToolsNG] Re: Stereo distortions in stereo panoramas

Expand Messages
• I feel like I m answering questions no one has asked me. But hey - I guess it would be wise to shoot at least 20, even with a landscape. However, the rule is
Message 1 of 18 , Feb 1, 2011
• 0 Attachment
I feel like I'm answering questions no one has asked me. But hey - I guess it would be wise to shoot at least 20, even with a landscape. However, the rule is quite robust ; the further the object is the less overlap you need and in theory you can get away with the normal overlap of 4 to 6 shots.

To: PanoToolsNG@yahoogroups.com
From: enridp@...
Date: Tue, 1 Feb 2011 14:07:00 +0000
Subject: [PanoToolsNG] Re: Stereo distortions in stereo panoramas

I see, thanks Peter for the explanation!

So... if we are shooting far objects, like a landscape, could we shoot only the common 4 images (for sigma8mm) for example?

Or shooting less pictures has another disadvantage in addition to parallax errors (for near objects)?

Regards !

Enrique.

--- In PanoToolsNG@yahoogroups.com, "panovrx" <panovrx@...> wrote:

>

>

> --- In PanoToolsNG@yahoogroups.com, "enridp" <enridp@> wrote:

> >

> > Peter can you teach me why we need so many shoots even with two cameras? I'm really confused by that.

> > I was thinking that using two cameras had two advantages over the method with one camera:

> > 1) You can shoot less pictures (<10) and therefore, faster.

> > 2) You can take action panoramas because you are making the 3D from two images taked at the same time.

> >

> >

> Panoramas must be stitched from images taken from approximately the same point in space. So with a normal panorama you need to find the No Parallax Point and rotate the camera round that point. With a 2 camera stereo rig both cameras are rotating around a point between them -- neither are rotating around any where near their NPPs. So to minimize the resultant parallax errors between the successive shots for each camera you need lots of shots and just use a central slice from each shot to blend to make the L and R panoramas. The closer the subject to the cameras the more shots you need -- somewhere between 30 and 300 maybe.

>

> PeterM

>

> PeterM

>

[Non-text portions of this message have been removed]
• For single camera (off-axis l/r slice assembly) stereo pano sweep capture Sony now has a rival -- Morpho, Inc (shades of the Matrix)
Message 2 of 18 , Feb 1, 2011
• 0 Attachment
For single camera (off-axis l/r slice assembly) stereo pano sweep capture Sony now has a rival -- Morpho, Inc (shades of the Matrix)
http://www.morphoinc.com/en/products/Panorama3D.html

--- In PanoToolsNG@yahoogroups.com, "panovrx" <panovrx@...> wrote:
>
> Here is an example of a couple of the stereo distortions I was thinking of:
> http://www.mediavr.com/verticaldisparity1.jpg
>
> You see the vertical separation of non-central features is extreme. It should be zero. Notice also the much exaggerated horizontal disparity in that part of the picture. This is too extreme to resolve so effectively the lateral periphery of all stereo panoramas viewed the usually warped ways are too separated to view. This is regardless of whether you use anaglyph, interlaced etc viewing methods. If the panorama is unwarped (ie. not rendered as a steerable view), eg. anaglyph equirectangular then these distortions do not exist but of course the view is unrealistic in other ways.
>
> The Abstract linked to the item on "Acquisition of stereo panoramas for display in VR environments" on
>
> http://spie.org//app/program/index.cfm?fuseaction=conferencedetail&export_id=x16280&ID=x16223&redir=x16223.xml&conference_id=927621&event_id=925953&programtrack_id=931449
>
> seems to say they have worked out a solution for this.
>
> Here is another street panorama
> http://www.mediavr.com/thana1bana.htm
> .. on a pole at about 4m with twin5d/10.5mm 50 shots in 12 sec at about 20cm separation (jpg because the camera cant do a fast 50 shot burst in Raw). To the left of the stairs there is a stitch error which I couldnt fix (the pole moved around too much I think) .. in 2d it would be fixable but that kind of stitching fixing is much harder to do perfectly in stereo sometimes.
>
> Peter
>
>
>
>
>
>
• ... Thanks, Peter. Their work is most impressive. I must admit my first reaction on discovering that they were based in Tokyo, within easy reach from our
Message 3 of 18 , Feb 1, 2011
• 0 Attachment
On Wed, 02 Feb 2011 07:51:27 +0900, panovrx <panovrx@...> wrote:

> For single camera (off-axis l/r slice assembly) stereo pano sweep
> capture Sony now has a rival -- Morpho, Inc (shades of the Matrix)
> http://www.morphoinc.com/en/products/Panorama3D.html

Thanks, Peter. Their work is most impressive. I must admit my first
reaction on discovering that they were based in Tokyo, within easy
reach from our condo, was that they might have need of my
translation skills! It would be fun to work on such leading-edge
technology.

Roger W.

--
Work: www.adex-japan.com
• ... Hello Enrique, For a 3D pano with no objects closeby and only far away objects you have to seperate the camera lenses wider to get more parallax otherwise
Message 4 of 18 , Feb 2, 2011
• 0 Attachment
enridp wrote:
> ... if we are shooting far objects, like a landscape, could we shoot only
> the common 4 images (for sigma8mm) for example?...

Hello Enrique,

For a 3D pano with no objects closeby and only far away objects you have to
seperate the camera lenses wider to get more parallax otherwise there is no
depth in the 3D view.

So instead of 9 cm, 25, 50 cm or even more lens seperation, depending on the
distance to far away objects, can be needed.
With such extreme lens seperations (hyper stereo) the NPP off set is really,
really huge and with just 4 images around it will cause huge stitch errors
in each of the left and right pano.
To tackle this probem you really have to shoot a lot more then 4 images.

In general; the more images you shoot the smaller the stitch errors will be,
this is because the total amount of parallax error wil be evenly spread over
all images. The number of images that are needed depends on the amount of
stitching errors that *you* find acceptable......

Wim
--
View this message in context: http://panotoolsng.586017.n4.nabble.com/Stereo-distortions-in-stereo-panoramas-tp3246722p3253601.html
Sent from the PanoToolsNG mailing list archive at Nabble.com.
• Thanks Wim !!! You are right about the extra separation (and hence the extra parallax error) needed for very far objects. But I m a bit confused now of why
Message 5 of 18 , Feb 2, 2011
• 0 Attachment
Thanks Wim !!!
You are right about the extra separation (and hence the extra parallax error) needed for very far objects.

But I'm a bit confused now of why earthmine is using two cameras, and with big separation:

http://www.freeimagehosting.net/image.php?3f93db70f8.jpg

They have the best "street view" system in the market I think, with a very accurate 3D (although there are only videos about earthmine, I couldn't find any online demo or something like that).

--- In PanoToolsNG@yahoogroups.com, Wim Koornneef <wim.koornneef@...> wrote:
>
>
>
> enridp wrote:
> > ... if we are shooting far objects, like a landscape, could we shoot only
> > the common 4 images (for sigma8mm) for example?...
>
> Hello Enrique,
>
> For a 3D pano with no objects closeby and only far away objects you have to
> seperate the camera lenses wider to get more parallax otherwise there is no
> depth in the 3D view.
>
> So instead of 9 cm, 25, 50 cm or even more lens seperation, depending on the
> distance to far away objects, can be needed.
> With such extreme lens seperations (hyper stereo) the NPP off set is really,
> really huge and with just 4 images around it will cause huge stitch errors
> in each of the left and right pano.
> To tackle this probem you really have to shoot a lot more then 4 images.
>
> In general; the more images you shoot the smaller the stitch errors will be,
> this is because the total amount of parallax error wil be evenly spread over
> all images. The number of images that are needed depends on the amount of
> stitching errors that *you* find acceptable......
>
> Wim
> --
> View this message in context: http://panotoolsng.586017.n4.nabble.com/Stereo-distortions-in-stereo-panoramas-tp3246722p3253601.html
> Sent from the PanoToolsNG mailing list archive at Nabble.com.
>
• ... That s not for stereo, but for accurately measuring distances through triangulation. Note that *your* eyes are arranged in a horizontal configuration.
Message 6 of 18 , Feb 2, 2011
• 0 Attachment
> But I'm a bit confused now of why earthmine is using two cameras,
> and with big separation:
>
> http://www.freeimagehosting.net/image.php?3f93db70f8.jpg

That's not for stereo, but for accurately measuring distances through triangulation. Note that *your* eyes are arranged in a horizontal configuration.
• Yes, but why is so important the orientation of the camera?, you can rotate your head 90 degrees and still viewing in 3D. Is not possible to arrage the camera
Message 7 of 18 , Feb 2, 2011
• 0 Attachment
Yes, but why is so important the orientation of the camera?, you can rotate your head 90 degrees and still viewing in 3D.
Is not possible to arrage the camera in a vertical configuration like earthmine? in tha way we will be rotating both cameras from their NPP.

--- In PanoToolsNG@yahoogroups.com, "Aldo Hoeben" <aldo@...> wrote:
>
> > But I'm a bit confused now of why earthmine is using two cameras,
> > and with big separation:
> >
> > http://www.freeimagehosting.net/image.php?3f93db70f8.jpg
>
> That's not for stereo, but for accurately measuring distances through triangulation. Note that *your* eyes are arranged in a horizontal configuration.
>
• ... If that is how you want to view your images, then by all means arrange your cameras in a vertical orientation instead of horizontal. But it sounds mighty
Message 8 of 18 , Feb 3, 2011
• 0 Attachment
> Yes, but why is so important the orientation of the camera?, you
> can rotate your head 90 degrees and still viewing in 3D.

If that is how you want to view your images, then by all means arrange your cameras in a vertical orientation instead of horizontal. But it sounds mighty uncomfortable to me ;-)

Earthmine uses the two cameras to be able extract actual 3d geometry from the pair of images. To do this with accuracy, the requirement is to know the exact distance between the two cameras, but it is not all that important how the cameras are arranged. If however they would put the cameras side-to-side, one camera would obscure part of the view of the other camera and vice versa. Ofcourse this still happens with the vertical arrangement, but for this application the zenith and nadir are not that interesting.

Stereo capture and viewing is something altogether different. For that the cameras need to mimick what the eyes do. IE: the two cameras (arranged horizontally) would need to rotate around a single shared point. If you were to use two one-shot systems, you would have the aforementioned problem that the two cameras see eachother when looking to the sides (left camera sees the right camera when looking right). Additionally things will get screwy if you look backwards in the panorama; the left and right "eye" will be swapped. Two one-shots are just not going to work.

Having said all that, if earthmine would indeed create accurate 3d geometry from their capture, it would be possible - in theory - to create stereo pairs out of that. But that requires A LOT of processing, and even then the result would be a far cry from what Peter has achieved lately with his stereo spinner work.
• Yes you are right, If we have the cameras in vertical configuration then the red and cyan will be up and down not at the sides, even after rotation of the
Message 9 of 18 , Feb 3, 2011
• 0 Attachment
Yes you are right, If we have the cameras in vertical configuration then the red and cyan will be up and down not at the sides, even after rotation of the images.
(an easy test, rotate your head with any Stereo Image and you can see how the effect is lost)

And the Stereo images from images with 3D information (distance of each pixel) could work, but the effect will not "volumetric" I think.
I mean, we will see different planes at different distances, not objects with volume.
That's what happens with "Fake3D" movies:
http://realorfake3d.com/

--- In PanoToolsNG@yahoogroups.com, "Aldo Hoeben" <aldo@...> wrote:
>
> > Yes, but why is so important the orientation of the camera?, you
> > can rotate your head 90 degrees and still viewing in 3D.
>
> If that is how you want to view your images, then by all means arrange your cameras in a vertical orientation instead of horizontal. But it sounds mighty uncomfortable to me ;-)
>
> Earthmine uses the two cameras to be able extract actual 3d geometry from the pair of images. To do this with accuracy, the requirement is to know the exact distance between the two cameras, but it is not all that important how the cameras are arranged. If however they would put the cameras side-to-side, one camera would obscure part of the view of the other camera and vice versa. Ofcourse this still happens with the vertical arrangement, but for this application the zenith and nadir are not that interesting.
>
> Stereo capture and viewing is something altogether different. For that the cameras need to mimick what the eyes do. IE: the two cameras (arranged horizontally) would need to rotate around a single shared point. If you were to use two one-shot systems, you would have the aforementioned problem that the two cameras see eachother when looking to the sides (left camera sees the right camera when looking right). Additionally things will get screwy if you look backwards in the panorama; the left and right "eye" will be swapped. Two one-shots are just not going to work.
>
> Having said all that, if earthmine would indeed create accurate 3d geometry from their capture, it would be possible - in theory - to create stereo pairs out of that. But that requires A LOT of processing, and even then the result would be a far cry from what Peter has achieved lately with his stereo spinner work.
>
Your message has been successfully submitted and would be delivered to recipients shortly.