Panoramas with different viewpoints

Discussion forum for techniques and issues relating to the creation of panoramic and/or "mosaic" images

Moderator: spamkiller

luigi
Posts: 21
Joined: Mon Apr 06, 2009 6:34 pm

Panoramas with different viewpoints

Post by luigi » Mon Apr 06, 2009 6:44 pm

Lets say I want to stitch a large train. Instead of moving the camera from a fixed position I move the tripod paralell to the train taking a picture with the same focus, WB, camera height etc etc.


(text example)

--***** TRAIN *******---
me1 me2 me3

After I take 3 pics from different points of view of the train is there a good way to stitch them together?

I hope my question is understandable ;)

Luis

elf
Posts: 221
Joined: Wed Dec 01, 2004 1:04 am
Contact:

Post by elf » Tue Apr 07, 2009 2:48 am

Do a search on orthographic stitching in this forum. You'll find quite a few discussions about this.

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Post by maxlyons » Tue Apr 07, 2009 9:31 am

elf wrote:Do a search on orthographic stitching in this forum. You'll find quite a few discussions about this.
Here is a current thread about the same subject.

Max

Karsten33
Posts: 9
Joined: Sat Mar 22, 2008 9:33 am

Post by Karsten33 » Tue Apr 07, 2009 2:47 pm

luigi,

Harf Zimmermann made a very special 80 m long mosaic of a subway train in Berlin with 190 people and with a scale of 1:1. You can see it on a fence of the building site of the new subway station in the near of the Brandenburg Gate ( very impressive) or here: http://www.xdegrees.de/special.html.


Karsten

Jim Z
Posts: 774
Joined: Tue Oct 11, 2005 9:16 pm
Location: Tahoma CA

Post by Jim Z » Thu Apr 09, 2009 12:49 pm

Luis,

Image

Here is one part of a 'orthographic' or 'flat' pano of three locomotives together.

There are very many pitfalls (difficulties) to doing this sort of pano. Objects or surfaces that are not in the same depth plane, the same distance from the camera, will have big paralax errors. Also, there are strange looking perspective errors of some things, like the chimmney and the horns on the roof of the locomative. (this problem is worse the closer you are to the subject, when yoiu take the photos. Taking the pictures from farther away, with a long lens, is better...)

To do the mosaic above, I took the photos in the same way as you intend to do. I first processed the photos to get them all 'flat' and level, and the same size. In PTA, I processed them with horizontal and vertical control points/lines in each picture, not shared or between pictures, for lens correction, correction of the camera alignment or squareness to the subject, and for getting the pics level. I used one set of horizontal lines, the rail at the bottom and the 'belt line' of the locomotives, to to optimize the HFOV of each image to get them scaled or sized the same. (The reason or necessity for this 'pre-processing' of the component images is; it is prob impossible to carry/move the camera, by hand, between pictures and keep the image plane (film plane) of the camera exactly square or normal to the object plane, and keep the sucessive camera positions exactly in a line, and equi-distant from the object plane. Using the roll, pitch and yaw, and FOV parameters adjusts or 'corrects' or these 'mis-alignments'. After the optimizing of each image, the seperate correctly flattened and sized images can be stitched together, only by positioning them in the 'correct positions'...)

I saved the images as a layered PS file, and converted each layer into a image file. I opened the image files in PTA and manually selected control points (this is tricky; paralax is a big problem). I kept the lens parameters at zero, and optomized the 'd' and 'e' shift parameters (and a little shear correction) to postition the images. I output a layered PS file and manually blended the images. (The Enblend blender made a mess of the paralax problems, this was before Smartblend, I don't know how well Smartblend would do.)

Not very easy, all things considered. I also tried preprocessing the images for basic lens correction in PTLens, and stitching the pictures all in one go in PTA with the HFOV set to a very small number (the lens focal lenth set to a very large number). That works well for stitching pictures of a 'flat' subject, one that is in a more-or-less single plane. It didn't work at all for this project...

Where you take each image, relative to the subject (the train), is important too. You need an image at each place where there is an object that receeds dirrectly away from you; at the ends of cars, at square bulkheads, at the axels, etc. (for example, the handrail at the left hand end of the locomotive is not square to the camera. When the pano continues to the next locomotive, this looks strange. I didn't get photos of each stair handrail... the back ground scenery behind the locomotives is a problem, too, between the locomotives; the paralax errors with the distant back ground behind the locomotives are very large, a single clean back ground picture needs to be blended into that part of the mosaic).

The best train pano that I've seen was an old one, taken with a rotating type pamorama camera (a Cirkut camera) from the center of a 'ballon track' siding (That is a semi-circular siding track; a track that is, in the plan view, in the form of a large circular arc. It is most of a full circle, on the ground, about a 200m diameter circle). The camera probably was in the center of the circle and all of the train was equi-distant from the camera, and it made a perfect, orthographic looking, image of the train, when printed (viewed) in the 'circular projection' that is intrinsic to that kind of camera.

All in all, probably the best thing is to take conventional pano images, from very far away, with a very long telephoto lens. Then use a rectilinear projection output for a conventionally stitched pano... What I mean is; the 'walking next to a train or a building' type of orthographic panos were a learning experience, for me. For the train panos, I took the pictures from very close to the train, on the rail of the adjacent track, close to the train for 'better detail' in the pictures... and I had to use a wide angle lens to take a picture of the full height of the subject from that close distance... Both of these things made the pano stitching process much more difficult. Knowing what I know now, I realize that standing farther away, to get the better perspective of an object, makes the pano process more easy. Conventional panos from a single place or point-of-view, are much more easy! (FWIW, I am still, constantly, making the same mistake; I take panos from the place where the 'view' looks best, to my eyes. I always seem to forget that the perspective, for the camera and the stitched pano image, is almost always better looking, from farther away.) ;- )

I hope that helps...
Jim Z

PSHRutPark
Posts: 168
Joined: Sun Jan 21, 2007 3:28 pm
Location: New Zealand
Contact:

Post by PSHRutPark » Fri Apr 10, 2009 2:25 am

Thanks for the detailed description Jim. Interesting.

Regarding SmartBlend: some time ago I compared the output of Enblend and SmartBlend on "real' panoramas with negligible parallax issues and found that Enblend produced the better output.

However for the "parallel" mosaics such as your train, Smartblend's output is definitely better. I guess because of it's superior parallax handling.

I drew this conclusion while undertaking a project using the same technique: but in my case a facade of a long building. I quickly photographed the building one day when the prevailing light conditions caught my eye. I went back later, used a tripod and measured distances and intervals. The resultant mosaic was technically well blended but because I couldn't replicate the light, the resultant image is quite unremarkable!

So I was determined to persevere with blending the original series. My best PTA blending attempt (focal length 10000mm and using SmartBlend) of a section of the building is currently shown at the end of the "Landscape I" series on my photography site.

PETER
http://www.pastoralsystems.co.nz/photography/

cooperin
Posts: 230
Joined: Thu Aug 28, 2003 8:41 pm
Location: Rochester, NY
Contact:

linear pano

Post by cooperin » Fri Apr 10, 2009 8:52 am

Jim, thanks for the information and workflow.

Great results and thanks for sharing.

BTW, it looks to me that this is a 1X6 pano. I counted the similar looking cloud formation.

Richard

Jim Z
Posts: 774
Joined: Tue Oct 11, 2005 9:16 pm
Location: Tahoma CA

Post by Jim Z » Fri Apr 10, 2009 11:13 am

Peter, that is an interesting building photo. The photos on your web site are excellent!

Richard, there are a few more component pics than that, for the lower parts... but not enough to avoid all of the perspective errors. I've done a few others like that, but I don't think that it is worth it: You make a large mosaic for the detail but the closer that you look at it, the more errors that you notice!
Jim Z

cooperin
Posts: 230
Joined: Thu Aug 28, 2003 8:41 pm
Location: Rochester, NY
Contact:

linear vs 1 view

Post by cooperin » Sun Apr 12, 2009 10:25 am

There is quite a bit of extra work to produce a linear panorama vs a 1 view panorama.

What would be the main advantage of a linear vs a 1 view panorama?

Thanks
Richard

PSHRutPark
Posts: 168
Joined: Sun Jan 21, 2007 3:28 pm
Location: New Zealand
Contact:

Re: linear vs 1 view

Post by PSHRutPark » Mon Apr 13, 2009 12:25 am

cooperin wrote:There is quite a bit of extra work to produce a linear panorama vs a 1 view panorama.

What would be the main advantage of a linear vs a 1 view panorama?

Thanks
Richard
For me, the "clear" foreground of the building facade was too shallow for a single point panorama to be taken without too much distortion compromising the detail at the extreme ends.

In looking at the results, the level of detail from end to end confirmed this approach.

Plus it was an interesting and challenging exercise both in the field and in using PTA !

PETER
http://www.pastoralsystems.co.nz/photography/

Jim Z
Posts: 774
Joined: Tue Oct 11, 2005 9:16 pm
Location: Tahoma CA

Re: linear vs 1 view

Post by Jim Z » Mon Apr 13, 2009 11:47 am

PSHRutPark wrote: ...Plus it was an interesting and challenging exercise both in the field and in using PTA !

PETER
That is one very good reason.

Also, there have been a couple of panos posted here that were of subjects that could not be photographed from a single point. One was a 'flat pano' of the Berlin Wall, and another was a pano of a tile mosaic, on the buildings on one side of a street in Nurnburg. Both are very long, crooked, subjects that couldn't other wiae be 'photographed'. (The 'goggle street' thing is sort-of the reductio ad infinitenum of this idea, but it is obvoiusly usefull)

I've done a few, 'flat', locomotive mosaic pictures when there wasn't room for a conventional pano, trains in railroad museums tend to be like supermarket aisles.

If the subject is flat, or if enough component photos can be taken, an 'impossible, perspective-less' pano can be made. Good for a "How'd they do that?" reaction.

I've also used a similar idea to take landscape panos where there are things (trees) obstructing the view. That is; I've moved parallel to the pano, taking clear pictures of each part of the pano landscape through a gap between trees, where a single panoramic view was avalable (obscured by the trees.)

Related to that: One of the neatest, most intriguing, photos that I've seen was a reconnaissance photo (Korean war) taken from a low flying airplane, with an oblique angle, moving film camera (slit camera). It was very long, perfectly detailed, diorama view of a valley and small mountain range. It is an unusual, impossible, perspective-less view, like one of those famous long scroll-like diorama paintings.

thanks
Jim Z

PSHRutPark
Posts: 168
Joined: Sun Jan 21, 2007 3:28 pm
Location: New Zealand
Contact:

Re: linear vs 1 view

Post by PSHRutPark » Wed Apr 15, 2009 5:08 am

Max et al

Further to my exercise of a multi-position mosaic of a derelict building mentioned above. I have been blending the left three images of the series because they are of most interest.


When I preview the mosaic the three images are "over plotted" correctly but once the third image is displayed the preview is automatically refreshed and the right-most edge is cut off.

The result is shown as the last image of the "Landscape I" series on my photography site (below). The right end of the image should include all the large doorway.

Bearing in mind that the mosaic has a "forced" total HFOV of 0.3 degree (10000mm focal length 0.1373 degree images): is the right hand trimming because of a limitation of the horizontal degree resolution of PTA routines?

This is the first time I've noticed it happening and it happens regardless of whether PTAStitcher is being used or the older functions (PTMender etc).

Thanks

PETER
http://www.pastoralsystems.co.nz/photography/

maxlyons
Posts: 3649
Joined: Fri Jun 20, 2003 8:55 pm
Location: USA
Contact:

Re: linear vs 1 view

Post by maxlyons » Wed Apr 15, 2009 9:41 am

PSHRutPark wrote:When I preview the mosaic the three images are "over plotted" correctly but once the third image is displayed the preview is automatically refreshed and the right-most edge is cut off.
The images are initially projected using equirectangular projection, and then once they have all been created, the composite is remapped to the user's projection of choice. I think that is what you describe as "automatically refreshed". If the right edge is chopped off, there may be a rounding error in PTAssembler's calculation of the appropriate FOV for the panorama. You can always manually adjust the FOV of the panorama using the preview screen.

Max

PSHRutPark
Posts: 168
Joined: Sun Jan 21, 2007 3:28 pm
Location: New Zealand
Contact:

Re: linear vs 1 view

Post by PSHRutPark » Thu Apr 16, 2009 12:22 am

maxlyons wrote:The images are initially projected using equirectangular projection, and then once they have all been created, the composite is remapped to the user's projection of choice. I think that is what you describe as "automatically refreshed". If the right edge is chopped off, there may be a rounding error in PTAssembler's calculation of the appropriate FOV for the panorama. You can always manually adjust the FOV of the panorama using the preview screen.

Max
Thanks Max. That did the trick.

I had to increase the HFOV from 0.3 to 0.35 before it would work. Had tried that previously but had clicked "Re-Optimse" rather than just "ReDraw".

PETER
http://www.pastoralsystems.co.nz/photography/

nwphoto
Posts: 10
Joined: Thu Aug 13, 2009 10:20 pm
Location: Cochrane, Alberta, Canada

Camera traversing the subject - Traversoramas

Post by nwphoto » Thu Aug 13, 2009 10:48 pm

I had been thinking about wide images of wide subjects for a long time, and had tried shooting Main Street from viewpoints on the other side. After some interesting failures I decided that what I needed was a LOT of very narrow images. But how to shoot those without wearing out the DSLR shutter, and myself?

Solution: video cameras happily shoot LOTS of images. So rotate an HD videocam to 'portrait' mode, mount it on my car window, and shoot while driving slowly along Main Street. Then write a program to extract, select every nth image and join them.

I did that, and the results were bizarre but interesting. Items near the viewpoint were foreshortened, and more distant items were 'shredded' into narrow strips and replicated many times.

A lot more thinking and I realised that the distant replication thing is because the more distant objects traverse across the camera's sensor more slowly, so will appear in more frames than nearer objects.

I basically did all that because I could not imagine what the result would be - now I know, and am trying to determine if the technique could be useful in another context. (It's not much good for Traversoramas/Travoramas/Travos).

The project is on hold for now. I may try providing a better frame selection mechanism than taking every Nth image (and a variable-width strip from each selected image). I feel that showing an image and it neighbours might allow me to select images that can be extracted and then stitched in a regular panorama program. A task for long winter evenings, I reckon.

TravoMaker is written in Python, and should be runnable on Windows, Linux, BSD and most popular computer platforms. If anyone is interested in exploring it I can make it available on request.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest