Quantcast

Action spherical stereo panoramas with video DSLRs

classic Classic list List threaded Threaded
81 messages Options
12345
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Action spherical stereo panoramas with video DSLRs

panovrx-2
http://www.mediavr.com/thanaglyph1.htm

Here is a stereo panorama shot with two 5DmkIIs/10.5mm Nikkor in about 4 seconds. The cameras are in vertical orientation on a spinner gadget over my head on a monopod that I am holding. They are both shooting video -- in M mode, at 1/1250 sec, at 1000IS at 30fps at about f4 (it was quite dark). For 360 that means there are about 120 exposures for each side. I turned each of the fisheye shots into equi views 50 degree horizontal and 180 vertical.
Then I loaded all the equi views -- left and right into PTGui having named them so that left and right alternated in the image sequence it was stitching. Then I used Generate Control Points and then deleted all the points that were on any near features. Then I optimized. Then I stitched the left view by omitting all the alternate images in the output options -- and ditto for the right view -- using Smartblend.
And that was it.
One advantage to the vertical orientation is I think is that it minimizes rolling shutter distortion which otherwise would have been extreme with the cameras spinning so fast. Mmm how to make it better.

Peter M

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

DemonDuck
Not bad.  But anaglyphs are always very dark.  Maybe you
could brighten them more than you think you should and that
would be about right.

panovrx wrote:

> http://www.mediavr.com/thanaglyph1.htm
>
> Here is a stereo panorama shot with two 5DmkIIs/10.5mm Nikkor in about 4 seconds. The cameras are in vertical orientation on a spinner gadget over my head on a monopod that I am holding. They are both shooting video -- in M mode, at 1/1250 sec, at 1000IS at 30fps at about f4 (it was quite dark). For 360 that means there are about 120 exposures for each side. I turned each of the fisheye shots into equi views 50 degree horizontal and 180 vertical.
> Then I loaded all the equi views -- left and right into PTGui having named them so that left and right alternated in the image sequence it was stitching. Then I used Generate Control Points and then deleted all the points that were on any near features. Then I optimized. Then I stitched the left view by omitting all the alternate images in the output options -- and ditto for the right view -- using Smartblend.
> And that was it.
> One advantage to the vertical orientation is I think is that it minimizes rolling shutter distortion which otherwise would have been extreme with the cameras spinning so fast. Mmm how to make it better.
>
> Peter M
>
>
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Roger D Williams
In reply to this post by panovrx-2
On Thu, 25 Mar 2010 09:37:24 +0900, panovrx <[hidden email]> wrote:

> http://www.mediavr.com/thanaglyph1.htm
>
> Here is a stereo panorama shot with two 5DmkIIs/10.5mm Nikkor in about 4  
> seconds. The cameras are in vertical orientation on a spinner gadget  
> over my head on a monopod that I am holding. They are both shooting  
> video -- in M mode, at 1/1250 sec, at 1000IS at 30fps at about f4 (it  
> was quite dark). For 360 that means there are about 120 exposures for  
> each side. I turned each of the fisheye shots into equi views 50 degree  
> horizontal and 180 vertical.
> Then I loaded all the equi views -- left and right into PTGui having  
> named them so that left and right alternated in the image sequence it  
> was stitching. Then I used Generate Control Points and then deleted all  
> the points that were on any near features. Then I optimized. Then I  
> stitched the left view by omitting all the alternate images in the  
> output options -- and ditto for the right view -- using Smartblend.
> And that was it.
> One advantage to the vertical orientation is I think is that it  
> minimizes rolling shutter distortion which otherwise would have been  
> extreme with the cameras spinning so fast. Mmm how to make it better.

The thing that impressed me most about this was something that had been
discouraging me from putting all the effort that would be needed into
making stereo panoramas. My concern was that the separation between the
two lenses is so important for the stereo effect that I wondered how
zooming in and out would affect it. I mean, the RELATIVE separation
between the lenses at different zoom ratios varies so widely. But I can
see this fear was groundless. I was afraid that zooming in would just
make distant objects look like a flat backdrop, as happens when you
take stereo shots of a distant mountain range with normal eye-like
lens separation--it looks like stage scenery! In your stereo photo,
as I zoom in, the stereo effect actually strengthens, and objects that
looked close together open out, with the apparent separation increasing.

I am greatly encouraged. The processing sounds quite complicated and
time consuming. Do you automate it with scripts for batch processing?

Roger W.

--
Business: www.adex-japan.com
Pleasure: www.usefilm.com/member/roger
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

RE: Action spherical stereo panoramas with video DSLRs

Keith Martin-2
In reply to this post by panovrx-2
> http://www.mediavr.com/thanaglyph1.htm

Fascinating stuff! It worked fairly well overall with the red/blue cardboard glasses I keep on my noticeboard. But it was curious (although probably inevitable) that there was some mirroring of red/blue offsets; some items were shifted one way, some the other, when they were at the same distance away.

k


[Non-text portions of this message have been removed]

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Wim Koornneef
In reply to this post by panovrx-2
Hello Peter,

Thanks for sharing your pano.
Obviously you succesfully used the video frames to create a lot of equirectangulars needed for the anaglyph pano.
I am curious, did you tried to connect small strips of the equirectangulars with StereoPhotoMaker instead of blending them in PTGui ?

The big drawback of using image frames instead of stills is the image quality, contrast, and the resolution that makes it not a pleasant experience when viewing the 3d pano fullscreen on large monitor.
I think your twin setup is more suited to shoot video panoramas because then the resolution isn't such a big deal. That said, when shooting 3d in a dynamic environment the only other solution is to use much more expensive equipment as used by professional 3D film makers.

Some time ago we discussed on this forum about the options and limits of the Loreo 3d lenses and it really is a pity that there new product with a wide angle 3D lens was taken out of production due to manufacturing problems.

Keith already mentioned the color mirroring of the left/right image in your pano, it has to do with the method of using the color channels when combining the left/right images into a combined anaglyph image. It is amazing that we don't notice anything of the mirroring when using a 3D viewer.

Here is a crop from a pano with a clear example of the color mirroring:
http://www.dmmdh.nl/forum_images/mirorring_red_cyan_colors.jpg

(this is the pano: http://www.dmmdh.nl/3d/dmmdh_haven_ouddorp_1_Anaglyph_en.html )

I hope you succeed in fine tuning the process of making 3D panos, please keep on posting them.

Wim
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
http://www.mediavr.com/woolana5.jpg 

This is a somewhat more accurate stitch than before I think --
I tried prealigning the pairs first this time by getting frames 1 (left and right) level and with identical yaw on a distant feature -- then applied those corrections to both L and R sequences of equis.

Then I made 20 degree wide alpha channels on each image -- to constrain the point finding and stitching. Then I optimised the left sequence and then applied the same values to the right sequence. One thing that occurs to me from this approach is that with machine vision sensors that often have ROI (region of interest) capability it means you could get very high frame rates by only using a vertical strip of the sensor. Say a 20*180 degree "slice" with a fisheye lens machine vision camera,


--- In [hidden email], Wim Koornneef <wim.koornneef@...> wrote:
>

> I am curious, did you tried to connect small strips of the equirectangulars
> with SteroPhotoMaker instead of blending them in PTGui ?
>

No I didnt try that because I shot them so quickly on a pole there was no way I could calibrate them so well for that to work -- but it could if it was on a tripod I think. But no good for action anyway with so narrow a coverage for any given frame

Peter M


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Wim Koornneef
Hello Peter,

The previous pano was also very good so it was hard for me to spot the improvements in the last version.

Working with TIF images and alpha masks to narrow the part of an image that is used for stitching is working fine but instead I prefer to create wide JPEG HQ compressed strips (created with an automatic Photoshop action) to prevent that the project storage size grows to enormous numbers.

You are right about the limitations of the use of connected strips, indeed you must have a very stable setup to make that possible, the yaw increment between the video frames must be very precise and I think this is only possible with a very precise and rotation speed controlled motor.

Wim
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
In reply to this post by panovrx-2
I made a test last night with the strip joining approach to stereo panorama making with video frames from my 5DMkII.

I have a turntable with a synchronous AC motor which has a very constant 1 revolution every 2 minutes (approx) rotation speed. I put my two 5D/10.5mm (vertical) cameras on this as close as they will go (on a common base with a 1/4" threaded section from a bolt joining both cameras). This btw gives you the minimum separation for stereo for two 5DMkII of 9.2cm
(unless you use a mirror box).

Then I carefully levelled the turntable and the cameras with a digital level. Then I shot a bit more than two minutes of video with both cameras -- at 30fps -- with the camera turning. Because it is turning so slowly, you can shoot the video continuously with shutter speeds down to about 1/60th without visible frame blurring.

Then I opened one of the videos in an editing application and noted a feature on the wall at the start and at the same position in a particular frame at the end (using a guide line in the editing application.) It turned out it shot 3582 frames for an exact 360 revolution. So each frame covers 0.1005 horizontal degrees of scene.

Then I trimmed down these 3582 fisheye frames (per side) (so PTGui wouldnt choke on the data) in Photoshop and loaded all the frames (for one side at a time) into PTGui and converted each of the frames into a 0.1005 by 180 degree equirectangular strip.
This is pretty quick (5 minutes per camera). Then I used the Mosaic tool in Stereophoto Maker to join the strips.

The panorama(s) looked good apart from broad brightness banding issues. These are to do with flare kicking in and out as bright light sources came into view -- the usual issue with scanning cameras unless the lenses are well baffled. What struck me most was the absolute perfect stitching without distortions of scene features very close to the camera  -- like 40 cm away. The video source material was obvious but the stitching was perfect. It means that if you take enough frames you can handle virtually any amount of scene depth in stereo panoramas.

Here follows an effort at describing single camera/dual camera stereo panorama capture via strip assembly:

There are two (main?) ways of strip assembly of digital stereo panoramas. The single camera (Peleg) and the dual camera approaches. With the single camera you have the camera rotating off-axis -- with the lens NPP point forward 10 cm of its proper location for instance. This means the viewpoint (the NPP)  of the camera is rotating in a circle round the camera support. Consider one particular frame. There is a scene feature, a door handle say, to the left of the centre of the view exactly 20 degrees say. As the camera rotates (counterclockwise for example's sake) there comes a point 40 degrees later where the door handle is now 20 degrees right of center frame. If these two frames are converted to level equirectangular views and strips extracted around the location of the door handles then you have two strips that look identical except for parallax difference corresponding to the different locations of the camera at the two frame capture times. So all you need to do is have one  rotation and assemble strips left or right of center from each frame and the distance from the center determines the amount of parallax and depth impression. Or you can extract multiple slices at varying distances left and right of center frame and assemble them for multiview stereo applications.

The disadvantages are obvious -- the scene must be static. Flare varies from the left and right views unless the lens is well baffled --(with twin slits).

The twin camera approach is simpler conceptually. You just have two synchronized cameras rotating in an assembly (so that their NPP points are left and right of the turntable center). You assemble central slices from each frame into left and right panoramas.

Here the scene doesnt have to be quite as static, and the levelling does not have to be so perfect. And the flare patterns will be the same in left and right panoramas. If the cameras can spin around very fast it is possible to make action scenes this way. Conceptually you could do action scenes with the single camera approach too if the rotation speed was fast enough.

You are reinventing the wheel you say -- this is just hacking DSLRs to turn them into degraded Spherons (scanning panorama cameras). The trouble with scanning cameras, apart from the expense, is that they capture only a pixel wide strip at a time (I think) and they are rarely fast enough for action except in very good light. If you are shooting with DSLRs you can retouch your panoramas later from an adjacent frame where the perspective is approximately the same, or improve light banding similarly.

But I think there is a good argument too for a halfway concept, namely small machine vision cameras with ROI (region of interest) capabilities. With these cameras frame rate at a given resolution can increase markedly when you are just capturing a central band of imagery. Two HD camera, CCD preferably,  1920 by 1080 nominally, used vertically, could capture a say, 20 pixel, wide strips at very high frame rates, even on a cheaper model.

Peter M











--- In [hidden email], "panovrx" <panovrx@...> wrote:

>
> http://www.mediavr.com/woolana5.jpg 
>
> This is a somewhat more accurate stitch than before I think --
> I tried prealigning the pairs first this time by getting frames 1 (left and right) level and with identical yaw on a distant feature -- then applied those corrections to both L and R sequences of equis.
>
> Then I made 20 degree wide alpha channels on each image -- to constrain the point finding and stitching. Then I optimised the left sequence and then applied the same values to the right sequence. One thing that occurs to me from this approach is that with machine vision sensors that often have ROI (region of interest) capability it means you could get very high frame rates by only using a vertical strip of the sensor. Say a 20*180 degree "slice" with a fisheye lens machine vision camera,
>
>
> --- In [hidden email], Wim Koornneef <wim.koornneef@> wrote:
> >
>
> > I am curious, did you tried to connect small strips of the equirectangulars
> > with SteroPhotoMaker instead of blending them in PTGui ?
> >
>
> No I didnt try that because I shot them so quickly on a pole there was no way I could calibrate them so well for that to work -- but it could if it was on a tripod I think. But no good for action anyway with so narrow a coverage for any given frame
>
> Peter M
>


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

hdrv
Peter,

A 0.1005 degree wide strip means how many pixels ? 3 or 4 I guess with  
your camera setup.

Romuald


Le 8 avr. 10 à 03:08, panovrx a écrit :

>
> I made a test last night with the strip joining approach to stereo  
> panorama making with video frames from my 5DMkII.
>
> I have a turntable with a synchronous AC motor which has a very  
> constant 1 revolution every 2 minutes (approx) rotation speed. I put  
> my two 5D/10.5mm (vertical) cameras on this as close as they will go  
> (on a common base with a 1/4" threaded section from a bolt joining  
> both cameras). This btw gives you the minimum separation for stereo  
> for two 5DMkII of 9.2cm
> (unless you use a mirror box).
>
> Then I carefully levelled the turntable and the cameras with a  
> digital level. Then I shot a bit more than two minutes of video with  
> both cameras -- at 30fps -- with the camera turning. Because it is  
> turning so slowly, you can shoot the video continuously with shutter  
> speeds down to about 1/60th without visible frame blurring.
>
> Then I opened one of the videos in an editing application and noted  
> a feature on the wall at the start and at the same position in a  
> particular frame at the end (using a guide line in the editing  
> application.) It turned out it shot 3582 frames for an exact 360  
> revolution. So each frame covers 0.1005 horizontal degrees of scene.
>
> Then I trimmed down these 3582 fisheye frames (per side) (so PTGui  
> wouldnt choke on the data) in Photoshop and loaded all the frames  
> (for one side at a time) into PTGui and converted each of the frames  
> into a 0.1005 by 180 degree equirectangular strip.
> This is pretty quick (5 minutes per camera). Then I used the Mosaic  
> tool in Stereophoto Maker to join the strips.
>
> The panorama(s) looked good apart from broad brightness banding  
> issues. These are to do with flare kicking in and out as bright  
> light sources came into view -- the usual issue with scanning  
> cameras unless the lenses are well baffled. What struck me most was  
> the absolute perfect stitching without distortions of scene features  
> very close to the camera  -- like 40 cm away. The video source  
> material was obvious but the stitching was perfect. It means that if  
> you take enough frames you can handle virtually any amount of scene  
> depth in stereo panoramas.
>
> Here follows an effort at describing single camera/dual camera  
> stereo panorama capture via strip assembly:
>
> There are two (main?) ways of strip assembly of digital stereo  
> panoramas. The single camera (Peleg) and the dual camera approaches.  
> With the single camera you have the camera rotating off-axis -- with  
> the lens NPP point forward 10 cm of its proper location for  
> instance. This means the viewpoint (the NPP)  of the camera is  
> rotating in a circle round the camera support. Consider one  
> particular frame. There is a scene feature, a door handle say, to  
> the left of the centre of the view exactly 20 degrees say. As the  
> camera rotates (counterclockwise for example's sake) there comes a  
> point 40 degrees later where the door handle is now 20 degrees right  
> of center frame. If these two frames are converted to level  
> equirectangular views and strips extracted around the location of  
> the door handles then you have two strips that look identical except  
> for parallax difference corresponding to the different locations of  
> the camera at the two frame capture times. So all you need to do is  
> have on
> e  rotation and assemble strips left or right of center from each  
> frame and the distance from the center determines the amount of  
> parallax and depth impression. Or you can extract multiple slices at  
> varying distances left and right of center frame and assemble them  
> for multiview stereo applications.
>
> The disadvantages are obvious -- the scene must be static. Flare  
> varies from the left and right views unless the lens is well baffled  
> --(with twin slits).
>
> The twin camera approach is simpler conceptually. You just have two  
> synchronized cameras rotating in an assembly (so that their NPP  
> points are left and right of the turntable center). You assemble  
> central slices from each frame into left and right panoramas.
>
> Here the scene doesnt have to be quite as static, and the levelling  
> does not have to be so perfect. And the flare patterns will be the  
> same in left and right panoramas. If the cameras can spin around  
> very fast it is possible to make action scenes this way.  
> Conceptually you could do action scenes with the single camera  
> approach too if the rotation speed was fast enough.
>
> You are reinventing the wheel you say -- this is just hacking DSLRs  
> to turn them into degraded Spherons (scanning panorama cameras). The  
> trouble with scanning cameras, apart from the expense, is that they  
> capture only a pixel wide strip at a time (I think) and they are  
> rarely fast enough for action except in very good light. If you are  
> shooting with DSLRs you can retouch your panoramas later from an  
> adjacent frame where the perspective is approximately the same, or  
> improve light banding similarly.
>
> But I think there is a good argument too for a halfway concept,  
> namely small machine vision cameras with ROI (region of interest)  
> capabilities. With these cameras frame rate at a given resolution  
> can increase markedly when you are just capturing a central band of  
> imagery. Two HD camera, CCD preferably,  1920 by 1080 nominally,  
> used vertically, could capture a say, 20 pixel, wide strips at very  
> high frame rates, even on a cheaper model.
>
> Peter M
>
>
>
>
>
>
>
>
>
>
>
> --- In [hidden email], "panovrx" <panovrx@...> wrote:
>>
>> http://www.mediavr.com/woolana5.jpg
>>
>> This is a somewhat more accurate stitch than before I think --
>> I tried prealigning the pairs first this time by getting frames 1  
>> (left and right) level and with identical yaw on a distant feature  
>> -- then applied those corrections to both L and R sequences of equis.
>>
>> Then I made 20 degree wide alpha channels on each image -- to  
>> constrain the point finding and stitching. Then I optimised the  
>> left sequence and then applied the same values to the right  
>> sequence. One thing that occurs to me from this approach is that  
>> with machine vision sensors that often have ROI (region of  
>> interest) capability it means you could get very high frame rates  
>> by only using a vertical strip of the sensor. Say a 20*180 degree  
>> "slice" with a fisheye lens machine vision camera,
>>
>>
>> --- In [hidden email], Wim Koornneef <wim.koornneef@>  
>> wrote:
>>>
>>
>>> I am curious, did you tried to connect small strips of the  
>>> equirectangulars
>>> with SteroPhotoMaker instead of blending them in PTGui ?
>>>
>>
>> No I didnt try that because I shot them so quickly on a pole there  
>> was no way I could calibrate them so well for that to work -- but  
>> it could if it was on a tripod I think. But no good for action  
>> anyway with so narrow a coverage for any given frame
>>
>> Peter M
>>
>
>
>
>
> ------------------------------------
>
> --
>
>
>
>



Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Wim Koornneef
In reply to this post by panovrx-2
Hello Peter,

Thanks for the very well documented test for shooting 3D with a dual DSLR camera setup in video mode.
You used a lot of image frames of the videos and you couldn't see errors at the connected seems, I think this proves the general rule that when connecting strips (instead of blending strips) "more is better".
That said I am pretty sure that with 1/2 of even 1/10 of the huge number of images frames you would also get a very good result.

I fully agree with the disadvantages of the single camera approach you mentioned but please keep in mind that a lot of people don't have 2 cameras with fisheyese lenses available so when taking the first steps into shooting 3D panos the single camera method is very good to develop knowledge and insight in the complexity of making 3D panos (we know that making a good 3D pano is more then just putting 2 images together..)
Because a lot of people are using a Nodal Ninja panohead I wrote an article about how to shoot and make a 3D pano with a single camera, a fisheye lens and a NN panohead and posted the article on the Nodal Ninja forum.
 
http://www.nodalninja.com/forum/index.php/topic,865.0.html

About the width of a strip of a scanning camera, I don't have such a camera but a tech person of Seitz told me that the width of the image they capture with their scanning cameras is not fixed to just 1 px and can be much more then that depending on the rotation speed of the camera, the iso setting, the shutter speed and some other parameters.

Peter thanks again and keep on posting,
Wim
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
In reply to this post by hdrv

--- In [hidden email], rvHD <vromuald@...> wrote:
>
> Peter,
>
> A 0.1005 degree wide strip means how many pixels ? 3 or 4 I guess with  
> your camera setup.
>
> Romuald
>

2 pixels wide 3582 pixels high for 0.1005 degrees by 180 degrees high equi strips



More thoughts on this stuff:
-- if you have a really solid setup, and can level everything exactly and a very constant motor, then if your frame timing is very constant it should be possible to shoot continuous rotation strip assembly stereo panoramas without calibration, except for working out the number of frames for 360 + a bit degrees rotation.

Suppose you have a 1 frame per second approx intervalometer timer and it is very consistent. And your turntable turns at a very steady rate. So you look at your frames and you notice that frames 1 and frames 120 are pointing at approximately the same area, but frame 120 is slightly advanced relative to frame 1.  You set the yaw of frame 1 to 0 degrees and calibrate the yaw of frame 120 -- say it is -0.57 degrees. So 120 frames = 360.57 degrees in terms of strips So each equi strip must be 360/360.57 of the normal pixel width for 120 frames in 360 degree rotations.

With video shooting the camera provides the intervalometer firing off shots at a very steady 30 fps. The same principle can apply I think to slower intervalometer regular stills timed sequences. The video would be a lot more forgiving though as the strips are so narrow.

If you want to use Smartblend then you could use the same idea to generate equi strips say exactly 6 degrees wide. Then in PTgui you set the yaw manually to 0, 3, 6, etc degrees etc and stitch those strips.

Must do some more tests.

Peter M






Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Wim Koornneef
This post was updated on .
Hello Peter,

Your idea to stitch without optimizing the yaw will work, all you need is a precise rotation and interval.

I have a question, could you try (or perhaps you already tried) to stitch the video frames without any optimizing and warping at all ?
So just taking out small strips of 2 px width of the middle of each fisheye video frame and then connect them together ?

The unwarped 2 px width strips should be almost similar (*) to the warped equirectangular version so the connected result should be almost equal.
If this is true then this would simplify the stitching procedure a lot.

(*) when taking 1 px out of the exact optical center of the fisheye image there is no difference at all (**) with an equirectangular and I think 2 px will also do fine but this has to be tested.

(**) there will be a small difference because the image is not corrected for lens errors but I don't expect that this will be problematic for the viewing experience but again, this has to be tested.

Wim
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

microfunguy
In reply to this post by panovrx-2

Peter wrote:

> More thoughts on this stuff:

Many thanks Peter, very useful stuff.

As you know, I have a single-camera rotation rig that I will be using in continuous-rotation mode with the Panasonic Lumix FZ3 (in its unique manual video, MJPEG mode) and a stepper mode for use with a portrait format Canon A620.

I am modifying my SDM software so that raw image strips (about 100 pixels wide) are saved rather than the entire frame.

Resulting panorama will be about 20,000 x 3000 pixels.


David

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Roger D Williams
On Fri, 09 Apr 2010 18:13:46 +0900, microfunguy <[hidden email]>  
wrote:

>
> Peter wrote:
>
>> More thoughts on this stuff:
>
> Many thanks Peter, very useful stuff.
>
> As you know, I have a single-camera rotation rig that I will be using in  
> continuous-rotation mode with the Panasonic Lumix FZ3 (in its unique  
> manual video, MJPEG mode) and a stepper mode for use with a portrait  
> format Canon A620.
>
> I am modifying my SDM software so that raw image strips (about 100  
> pixels wide) are saved rather than the entire frame.
>
> Resulting panorama will be about 20,000 x 3000 pixels.

Do keep us posted on these developments. Look forward to seeing some
results!

Roger W.

--
Business: www.adex-japan.com
Pleasure: www.usefilm.com/member/roger
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
In reply to this post by panovrx-2
Well I did more tests. I used a Canon TC-80N3 release as the intervalometer and had the cameras turning on an extremely solid AC turntable(too heavy for regular use) with a 2 min rotation time and shot off shots at what was nominally 1 fps. After I worked out the strip width the mosaiced panorama was pretty perfect (except for the usual slight stepping with close up areas and brightness differences between the slices in the nadir and zenith areas (top and bottom of slice)). So this means that intervalometer is consistent to at least 0.001 sec I think, and the turntable is consistent, and the camera latencies are consistent. So this approach is viable. But 120 or so frames is not enough for  very large scene depth ranges like I was shooting using just strip assembly.

Things looked a lot better using the same source images and Smartblend blending  derived exact  10 degree wide equi slices. But it was a pain entering the incrementing yaw values for each equi frame. I calculated it with Excel using the degrees per frame PTGUI result (say X degrees)  of the first and final frame yaw calibration but then had to copy and paste away for ages putting the values in PTGui -- ie frame 1 yaw 0, frame 2 yaw X, frame 3 yaw 2X, frame 4 yaw 3X, etc
--- PTGui needs the ability to automatically increment particular yaw values --
.. but it must be possible to script the generation of the PTStitcher scripts anyway

Peter M



>
> More thoughts on this stuff:
> -- if you have a really solid setup, and can level everything exactly and a very constant motor, then if your frame timing is very constant it should be possible to shoot continuous rotation strip assembly stereo panoramas without calibration, except for working out the number of frames for 360 + a bit degrees rotation.
>
> Suppose you have a 1 frame per second approx intervalometer timer and it is very consistent. And your turntable turns at a very steady rate. So you look at your frames and you notice that frames 1 and frames 120 are pointing at approximately the same area, but frame 120 is slightly advanced relative to frame 1.  You set the yaw of frame 1 to 0 degrees and calibrate the yaw of frame 120 -- say it is -0.57 degrees. So 120 frames = 360.57 degrees in terms of strips So each equi strip must be 360/360.57 of the normal pixel width for 120 frames in 360 degree rotations.
>
> With video shooting the camera provides the intervalometer firing off shots at a very steady 30 fps. The same principle can apply I think to slower intervalometer regular stills timed sequences. The video would be a lot more forgiving though as the strips are so narrow.
>
> If you want to use Smartblend then you could use the same idea to generate equi strips say exactly 6 degrees wide. Then in PTgui you set the yaw manually to 0, 3, 6, etc degrees etc and stitch those strips.
>
> Must do some more tests.
>
> Peter M
>


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

crane



On Fri, April 9, 2010 11:41 am, panovrx wrote:

> --- PTGui needs the ability to automatically increment particular yaw
> values

image parameters fill yaw  or have I misunderstood ?


mick


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
In reply to this post by microfunguy



I tried making a strip-mosaiced video stereo panorama with a quicker turntable -- 10 seconds for 360 = 300 frames approximately. This is with a tripod and the strips are not blended, just mosaiced. Here is one of the views unretouched --
http://www.mediavr.com/lunapark.htm

or as a Shockwave stereo movie
http://www.mediavr.com/lunapark.htm

I used a hood with two slits for the cameras here so there is little brightness banding but you can see the slit artefacts on the nearest girls (from them moving as the camera spins past them).

It is great not to have to do any optimising with the mosaicing method.With regular still frames instead of video frames it would look much better .. it just would take longer to shoot enough frames (100+). You need a steady turntable and very level camera rig though.

I look forward to seeing your FZ panoramas David.

Peter M


--- In [hidden email], "microfunguy" <killspammers2@...> wrote:

>
>
> Peter wrote:
>
> > More thoughts on this stuff:
>
> Many thanks Peter, very useful stuff.
>
> As you know, I have a single-camera rotation rig that I will be using in continuous-rotation mode with the Panasonic Lumix FZ3 (in its unique manual video, MJPEG mode) and a stepper mode for use with a portrait format Canon A620.
>
> I am modifying my SDM software so that raw image strips (about 100 pixels wide) are saved rather than the entire frame.
>
> Resulting panorama will be about 20,000 x 3000 pixels.
>
>
> David
>


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
http://www.mediavr.com/luna1.jpg   this is the JPG url

--- In [hidden email], "panovrx" <panovrx@...> wrote:

>
>
>
>
> I tried making a strip-mosaiced video stereo panorama with a quicker turntable -- 10 seconds for 360 = 300 frames approximately. This is with a tripod and the strips are not blended, just mosaiced. Here is one of the views unretouched --
> http://www.mediavr.com/lunapark.htm
>
> or as a Shockwave stereo movie
> http://www.mediavr.com/lunapark.htm
>
> I used a hood with two slits for the cameras here so there is little brightness banding but you can see the slit artefacts on the nearest girls (from them moving as the camera spins past them).
>
> It is great not to have to do any optimising with the mosaicing method.With regular still frames instead of video frames it would look much better .. it just would take longer to shoot enough frames (100+). You need a steady turntable and very level camera rig though.
>
> I look forward to seeing your FZ panoramas David.
>
> Peter M
>
>
> --- In [hidden email], "microfunguy" <killspammers2@> wrote:
> >
> >
> > Peter wrote:
> >
> > > More thoughts on this stuff:
> >
> > Many thanks Peter, very useful stuff.
> >
> > As you know, I have a single-camera rotation rig that I will be using in continuous-rotation mode with the Panasonic Lumix FZ3 (in its unique manual video, MJPEG mode) and a stepper mode for use with a portrait format Canon A620.
> >
> > I am modifying my SDM software so that raw image strips (about 100 pixels wide) are saved rather than the entire frame.
> >
> > Resulting panorama will be about 20,000 x 3000 pixels.
> >
> >
> > David
> >
>


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

panovrx-2
In reply to this post by panovrx-2
To recap, one way of avoiding the horrors (because of the extreme non-NPP nature of the image sequences) of calibration for stereo pano sequences is to use a very level tripod and a steadily rotating and shooting camera(s). Then you only need to know the number of frames for 360 (approx) and the yaw difference of last and first frames of the first repeating sequence to work out the constant yaw increment for each frame.

So how to do this without a tripod, ie. hold a spinning pair of heavy cameras, more or less separate from each other, very constantly level high on a hand held pole. Ah, gyroscopes I thought, the faster the better, maybe there will be a spinning top effect keeping it pretty level automatically -- provided of course that you have the cameras and motor perfectly balanced on top of the pole. Shooting video stereo panoramas would not be a problem I think like this except for the start and end of the 360 section you have chosen (of the unwinding >360 pano strip your stitching will produce). These will be unlikely to have identical roll and pitch values.

Peter M




--- In [hidden email], "panovrx" <panovrx@...> wrote:

>
>
>
>
> I tried making a strip-mosaiced video stereo panorama with a quicker turntable -- 10 seconds for 360 = 300 frames approximately. This is with a tripod and the strips are not blended, just mosaiced. Here is one of the views unretouched --
> http://www.mediavr.com/lunapark.htm
>
> or as a Shockwave stereo movie
> http://www.mediavr.com/lunapark.htm
>
> I used a hood with two slits for the cameras here so there is little brightness banding but you can see the slit artefacts on the nearest girls (from them moving as the camera spins past them).
>
> It is great not to have to do any optimising with the mosaicing method.With regular still frames instead of video frames it would look much better .. it just would take longer to shoot enough frames (100+). You need a steady turntable and very level camera rig though.
>
> I look forward to seeing your FZ panoramas David.
>
> Peter M
>
>
> --- In [hidden email], "microfunguy" <killspammers2@> wrote:
> >
> >
> > Peter wrote:
> >
> > > More thoughts on this stuff:
> >
> > Many thanks Peter, very useful stuff.
> >
> > As you know, I have a single-camera rotation rig that I will be using in continuous-rotation mode with the Panasonic Lumix FZ3 (in its unique manual video, MJPEG mode) and a stepper mode for use with a portrait format Canon A620.
> >
> > I am modifying my SDM software so that raw image strips (about 100 pixels wide) are saved rather than the entire frame.
> >
> > Resulting panorama will be about 20,000 x 3000 pixels.
> >
> >
> > David
> >
>


Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Action spherical stereo panoramas with video DSLRs

Wim Koornneef
panovrx-2 wrote
...hold a spinning pair of heavy cameras, more or less separate from each other, very constantly level high on a hand held pole. Ah, gyroscopes I thought, the faster the better...
I very much doubt that this will the best road to go, as you mentioned, and I agree, you will still get a different roll/pitch between the last and the first image frames of the sequence but what about the weight of the gear? I am pretty sure as it is right now with 2 heavy cameras+motordrive+pole it isn't exactly fun to walk around with and when adding a gyroscopic device it will be even more unpleasant.
I guess it is better to fine tune and automate the optimizing and the stitching of the frames instead..
Anyway success.

Wim
12345
Loading...