Increasing FOV for cramped architecture

[sharperstill]

Hi all,
A client wants an elevation view of a new 7 story apartment building with a double lane street out the front. That's as far away as I can get directly away from the building. Oblique images across the face of the building understandably have a lot of perspective distortion which the client doesn't like. Getting further away in an oblique direction is made more difficult by the presence of power poles.
From directly across centre of the building it is just too wide for my Canon 17TS and I need significant upward shift to include the upper levels.
I haven't come across the situation too often. Last year I did a 'shift panorama' (is this what is referred to in these forums as a 'flat stitch'?) for an indoor pool I photographed (17mmTS). I wasn't entirely happy with the result as there was some distortion noticeable down in the lower area of the image. Unfortunately I didn't do a panned & stitched composite for comparison. Image attached
With the image of the apartment block though, I need to use most of the rise in order to include the upper levels, so I can't do a 'shift panorama'.
My options for increasing the field of view to include the whole building therefore seem to be;
1) Hire the Canon 11-24, no movements but at 11mm I will probably get the upper level of the building in. May have some room for cropping once alignment/distortion is dealt with.
2) Do a pan & stitch panorama.
So, based on the second option this morning I found a similar size building with a similar shooting distance in a nearby suburb and tried a pan & stitch as a practise run (the actual building is still under landscaping/paving works). The results aren't pretty. In fact, I can't even get Photoshop's Photomerge function to include all the images and produce something even vaguely resembling the desired outcome.. It is seemingly getting confused by the perspective differences in the source images. The second attachment shows my desired outcome, just to be clear. A pure elevation view.

I'd appreciate any advice on how best to tackle this, especially whether it is possible to get the desired result using the pan & stitch mode ( there is no lens rental in this regional city).

Thanks,

Jon

[Bart_van_der_Wolf]

I'd appreciate any advice on how best to tackle this, especially whether it is possible to get the desired result using the pan & stitch mode ( there is no lens rental in this regional city).

Hi Jon,

Pan and stitch, and use a dedicated pano-stitcher, like PTGUI-Pro.

One thing you will never be able to solve is that the perspective of the resulting stitch will always give an impression of being distorted, but that's just caused by viewing the resulting output from too far away. It's simple geometry, and there's not much that can be done about it.

If the facade is very flat, with few recesses and protrusions, one could attempt a different kind of stitch, for which you also need an application such as PTGUI-Pro. It uses a technique that was pioneered by Max Lyons, as explained on his Webpages:
http://www.tawbaware.com/pta_help/ptasmblr_help_camera_position.htm

That would involve using different camera positions, the images of which can be reassembled because they are in a flat plane shot from different angles. And since a flat plane has no depth, there will be no parallax error on the facade. The street in front will require a different approach, because that street is sensitive to parallax errors if the entrance pupil is not stationary. A montage in post between the two will be in order.

The current Beta version of PTGUI can also mix shots with different focal lengths, which might be useful for a pan and stitch of the street in front, and or of the facade if it is not flat. That would allow shooting the more distant parts with more detailed tile taken with a longer focal length.

Lots of planning and postprocessing is inevitable to achieve the physically impossible. It might take some testing to get the hang of the procedures.

Cheers,
Bart

[alan_b]

No matter what lens / perspective correction / stitching method you use, at a certain building height and shooting distance, things are going to look too stretched at the top.  At that point, you could raise your shooting point with a ladder/lift or get access to the building across the street.  You'll lose the ground-level viewpoint into the first floor - everything is a tradeoff!

When I'm scouting such a situation, I take a tilted up shot and post-correct the perspective.  I then ask myself how many of the top floors look wrong - that tells me how high to raise the camera position.

[elliot_n]

Pan and stitch, and use a dedicated pano-stitcher, like PTGUI-Pro.

This is the best solution. If you go this route, you will also need a pano-head. A pano-head isn't cheap and it needs carefully setting-up — i.e. finding the no-parallax point for your chosen lens (a 35mm or 50mm would be good for a job like this).

[kirkt]

How about using a drone to acquire a flat stitch set of images?  I have performed the previous technique of elevating the camera (I used a 14 ft long pole with a Canon 15mm fisheye) and sprayed the facade (approx 120 ft high church) and stitched.  As you would expect, there was still significant distortion at the top level of the facade and spire.

If you were able to "scan" the facade with a drone, perhaps the flat stitch would work.  You could shoot the surrounding street level, etc. with standard techniques and composite the result into something that achieves the look the client wants.

 I do not profess to know how to use a drone, but just thinking out loud.

kirk

[sharperstill]

Thanks or the replies. I'll attach some phone snaps from this morning to give context to the shoot and the distances and angles. I also borrowed a 14mm for a quick recon on the FoV of that lens straight on. You'll see it just fits the width without much wriggle room.

I'm surprised to read that pan & stitch is the preferred technique when yesterday's attempt was such an unmitigated disaster.

Bart, the facade has enough depth that the use of different camera positions would not work, also because of the limited positions for the cameras in the first place.

Elliot, I have a rail, and I have the measurements for the magic spinning point for my lenses. What advantage would a pano-head have over this?

Jon
Attachments;
Phone pic of building and tripod position
Phone panorama of lower part of building
Street level view with 14mm

[Martin Kristiansen]

The problem with shooting a building that height from so close and doing such a strong perspective correction to fix the keystoning is it looks odd. You end up seeing the ceilings of the upper floors while the building is squared up.  That is an impossible perspective in reality.

I think you need to get some height. I can’t see properly but it looks like there is a building opposite. Try to get access. Or get a cherry picker. Any height at all is going to help. If the client wants it shot from ground level and corrected for converging verticals then you are going to need to manage his expectations.

[elliot_n]

Elliot, I have a rail, and I have the measurements for the magic spinning point for my lenses. What advantage would a pano-head have over this?

A pano-head (at least by my definition) allows you to shoot rows as well as columns. But looking at your test shots, that might not be necessary. Your 14mm shot, when corrected for perspective distortion, gets close to what you're looking for. But it is clearly not wide enough. The 11mm lens you mentioned in your first post would do it (but the perspective distortion correction would lose a lot of resolution). Better, use the 14mm lens, mounted in portrait format on your rail, and shoot 3 frames, left to right, then stitch with PTGui. You'll need to step forward a couple of feet to clear the power cables, road sign, and nearside cars.

A difficult shoot for sure. I agree that an elevated position would be preferable, but the power cables might prevent that.

[Bart_van_der_Wolf]

Thanks or the replies. I'll attach some phone snaps from this morning to give context to the shoot and the distances and angles. I also borrowed a 14mm for a quick recon on the FoV of that lens straight on. You'll see it just fits the width without much wriggle room.

I'm surprised to read that pan & stitch is the preferred technique when yesterday's attempt was such an unmitigated disaster.

Bart, the facade has enough depth that the use of different camera positions would not work, also because of the limited positions for the cameras in the first place.

Hi Jon,

Thanks for the snaps that give more context.

Pan and Stitch it is. And do make sure to use PTGUI Pro. The Pro edition can use masking which will come in handy when removing the powerlines, road signs, etc., when healing them out won't work. Just shoot from 2 slightly different positions, so that one shot can look behind the obstruction. It is then possible to align the two shots in the flat plane, and take away the obstruction by revealing the area behind with a mask.

I'd also use a longer focal length and more tiles, to make sure you get the required level of detail after software warping. PTGUI also allows compressing the Horizontal and Vertical dimension, to achieve a slightly more natural look.

But the short distance perspective will remain a difficult thing to tame in small output sizes. The larger the output size, the easier it will become to look at it from the correct (distortion free) viewing position. My estimate is that anything output smaller than, say, 135 cm (53 inches) wide and at the same time viewed at further than 10-inch reading distance, will 'look' distorted. So it will remain a compromise that even PTGUI can only partially improve on.

Cheers,
Bart

[mcbroomf]

Pan and shift should have worked out fine and it's the easiest to do which is why it has been recommended so I wonder why your test failed. 
Were you careful to make sure there was adequate overlap, especially along the top which would need a lot more images to cover (the width)? 
What lens did you use? 
If you used your 17mm TSE did you leave it centered or did you apply some shift to some or all images?

[sharperstill]

Pan and shift should have worked out fine and it's the easiest to do which is why it has been recommended so I wonder why your test failed. 
Were you careful to make sure there was adequate overlap, especially along the top which would need a lot more images to cover (the width)? 
What lens did you use? 
If you used your 17mm TSE did you leave it centered or did you apply some shift to some or all images?

In my test I used significant vertical shift, to include the top of the building. I wondered whether repetitious building elements were confusing the software ( windows and other elements that are the same on multiple levels of the building etc. I though I had left April 1/3 frame width (in portrait) as overlap.
my test was done on street level but I have access to the first floor balcony of the house opposite, which will raise viewpoint by about3.5 metres.  Some works have completed and I might be able to run a test in situ next week.

[mcbroomf]

I've had problems combining images with vertical shift and without as (I suppose) the stitching software sees them as different images, ie effectively one set is distorted compared to the other.  Repeated elements should not screw up the software and PS/LR are much better than they used to be.  Try repeating it with no shift at all.  It should work fine.

[Bart_van_der_Wolf]

In my test I used significant vertical shift, to include the top of the building. I wondered whether repetitious building elements were confusing the software ( windows and other elements that are the same on multiple levels of the building etc. I though I had left April 1/3 frame width (in portrait) as overlap.

Hi,

Non-dedicated stitchers cannot handle shifted images, but PTGUI can (by using an offset).

Cheers,
Bart

[sharperstill]

Thanks for all the replies. I’m going to try another test on Tuesday and also try out PTGUI Pro.
I have another question. What can I achieve with a ‘pano head’ that I can’t achieve by just adjusting the tilt angle on my Manfrotto 405 geared head?  Can’t i use my feared head for a multi tow panorama.
Related question:  As i have never done one before would I be stupid to attempt a multi-row in this instance and what focal length would be ideal?

Jon

[elliot_n]

What can I achieve with a ‘pano head’ that I can’t achieve by just adjusting the tilt angle on my Manfrotto 405 geared head?  Can’t i use my geared head for a multi row panorama?

When you tilt with your Manfrotto head, you won't be tilting around the lens's no-parallax point. This will cause stitching errors. A pano-head supplies a second 'upper-rail' which solves this problem.  But as has been already said, you can satisfactorily make this image with a single row, shooting several frames with the 14mm lens in portrait orientation. A multi-row panorama, shot with say a 35mm or 50mm lens, will offer you significantly more resolution (possibly more than you need), but it won't have any effect on the geometry of the image (i.e. it will look the same as a single-row shot with the 14mm lens).

[sharperstill]

When you tilt with your Manfrotto head, you won't be tilting around the lens's no-parallax point. This will cause stitching errors. A pano-head supplies a second 'upper-rail' which solves this problem.  But as has been already said, you can satisfactorily make this image with a single row, shooting several frames with the 14mm lens in portrait orientation. A multi-row panorama, shot with say a 35mm or 50mm lens, will offer you significantly more resolution (possibly more than you need), but it won't have any effect on the geometry of the image (i.e. it will look the same as a single-row shot with the 14mm lens).

Ah, of course.

Thanks

[Bart_van_der_Wolf]

Ah, of course.

Personally, I prefer to avoid the Parallax-Error risk ('Murphy'  tends to strike when you can least use it) and therefore use a full-fledged MultiRow Pano setup.

However, with PTGUI Pro and the specific scenario of a frontal facade shot, it might be worthwhile to test a Multi-Row stitch even with a Single-Row stitching setup. When you tilt the tripod head, you will introduce parallax error because the entrance pupil moved relative to the other Row(s). But that parallax will occur mostly at street-surface level and at closer distances.

Well, PTGUI Pro has Masking functionality that can mask out the street-level Parallax where the Rows might overlap, while minimizing the more distant (and less visible) Facade Parallax errors. Parallax errors are most prominent at close distances and can be minimized for a given (facade) distance.

Parallax will be zero at one specific distance, especially when the control points are all placed at only that distance. So make sure that at the bottom of the facade, only control points are placed on the facade as part of the facade row. And as always, avoid control points on clouds (which usually move between exposure of neighboring tiles). It's a bit of extra work, but it is possible when the software allows manual control. Photoshop and similar don't offer that level of control.

Cheers,
Bart

[sharperstill]

Hi all,
I did a dry-run this morning. It took some finagling to get the tripod into a position where the balcony ny ceiling was not visible. I basically did 3 tests.
The first was a pan -n- stitch with a horrible borrowed 14mm in portrait orientation.
Second was a pan -n- stitch with my 17 TS shifted about 9mm up. I shot 5 frames, although this was possibly too much.
Then, for the heck of it, I did a shift stitch with the 17mm.
On my first go of PTGUI Pro I used the 5 frames shot with the 17. Result had some bad stitching errors and I had to leave it for a while. I gorged on a few YouTube tutorials to get across the finer points of the software. I then had another go but used only 3 of the images and got a result without stitching errors.
One thing I didn’t like, and I haven’t yet figured out why this happened, but PTGUai has done something funky with the tonality and colour of my images. That’s tonight’s homework.
Lastly I merged the shift pano images in Photoshop. I can’t get the 2x3 image ratio that the rest of my work for this client confirms to with this method.
I doubt at this stage I’ll bother with a multi row attempt. Client wants to proceed.
My favoured option at this stage would be to do the 17 shift-pan-n-stitch again with more finesse in better light.
I’m pretty certain I’m going out to buy a Rogeti TSE frame at some stage but it won’t come in time for this shoot, and anyway the coverage of the scene is beyond the image circle in this case.
I’m also not sure I would gain anything from hiring the 11-24.
I’ve attached the results so far.

Jon

 

[elliot_n]

Re. the colours in PTGui, what are you feeding it? Best practice is tiffs.

[elliot_n]

How 'horrible' is the 14mm lens? If you're talking about the distortion (barrel/moustache) that was evident in your earlier shots, this will be totally removed in the PTGui stitching process. (If it's soft, that's a more serious problem.) The 14mm would give you some useful space around the building.

It looks like you're getting there with the PTGui stitch, but there are clear stitching errors in the power cables. Of course, you will clone out these cables, but have you zoomed into the image to check for stitching errors on the brick-work of the building?