Shooting Panoramas in Tight Spaces

[Alexander Drecun]

Hey all,

This is a question partially about Hugin but also about how to shoot a specific type of site/location in order to produce a quality panorama. If any of you know of resources that provide solutions to this problem, then please send them along.

In brief, I'm trying to photograph an entire exterior wall of a very large and long property somewhere in the area of 300 ft. The problem is, this property sits on a pretty narrow street with another walled property across from it, putting me no further than thirty feet from the wall. This means there's no way for me to shoot from a single nodal point that would produce a clean panorama with no stretching of pixels at the edges or anything like that. In fact, I'm not even sure I could shoot the entire property from a single point if I tried.

So, I was wondering if anyone has experience with or thoughts on photographing something like this. 

My current plan for how to tackle this is:

- posit a hypothetical nodal point some distance removed from the property that is based up on final panorama rendered from the perspective of, say, a full frame 50mm lens.

- from this hypothetical nodal point, figure out how many component images I would need to make my final panorama.

- then figure out where the midpoint of each component image is on the actual street I'll be photographing from.

- make a panorama at each of these midpoints.

- stitch those panoramas into the final panorama

All that said, I'm relatively new to both Hugin and panoramas in general and so have no idea whether this will work. Does anyone have any thoughts or advice to offer?

[John Fine]

A lot depends on how much relative depth there is to the surface.  If there is significant relative depth, then there will be parallax problems that hugin has no good way of managing.  Otherwise, it should not be terribly difficult.

I'm failing to see the point of computing the hypothetical component images or their midpoints.  The hypothetical nodal point is put into effect by parameters of the final projection.

The usual approach would be to take pictures directly toward the wall that are however wide fits (from the 30 feet away you can get) and about 30% overlapped one to the next, moving the camera sideways along the wall to get all the photos.  You might want to also make a seperate project to compute lens parameters, rather than needing to include lens parameter optimization in the main project.

If you are pretty careful about always pointing straight on, there will be very little yaw, pitch or roll.  The best work around to conflicts of yaw vs Tx etc. is to first optimize with just the parameter that should dominate, then reoptimize including other parameters (expert mode and custom selection of parameters is easier than you might expect).  For the project you describe, Tx should be the dominant parameter (it represents moving the camera sideways without turning it).

--
A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ
---
You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hugin-ptx+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/d8c7a07c-4d5a-41e7-8acd-542265455131n%40googlegroups.com.

[dgjohnston]

Here is an image of something I did a few years ago that sounds similar to what you may be trying to do.

If’s made up of 12 images. I moved along across the street from these buildings trying to be at 90 degrees to each “section” of the buildings; hoping to reduce any parallax problems later in hugin. I stitched the images together then used masking to cut out anything of the images that wasn’t part of the “section” straight across from where the picture was taken.

If you want to download and see the details of how I did this here is the link to the folder with the images and hugin files.

https://ln5.sync.com/dl/380fc1290/x63ceeax-pnft3mpd-fhub63bd-gcasikxn

Don Johnston

To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/CALe0Q_keF5Qa9HQSjEWTWTREp8RGg%2B_1kCPDNU%3DkdOKesWw%2Bpg%40mail.gmail.com.

[Alexander Drecun]

Thanks for the response. Relative depth may be the main issue I run into here because there are parts of a house as well as trees visible behind the wall at various distances. Do you have a sense for how much relative depth the approach you outlined can handle? 

Lol the hypothetical component images is a workaround I thought up because I'm pretty new to Hugin and my initial stabs at making images along the lines of what I described have not gone well. Clearly I just need to spend more time in the software and with some tutorials.

[Alexander Drecun]

Thanks, Donald. Yes, this is very much what I'm trying to do and I appreciate you making the folder available to me. I'll definitely take a look.

[johnfi...@gmail.com]

On Thursday, September 8, 2022 at 11:41:00 PM UTC-4 alexande...@gmail.com wrote:

. Relative depth may be the main issue I run into here because there are parts of a house as well as trees visible behind the wall at various distances.

In that case, your idea of working from a hypothetical nodal point might be good (but the split into component images based on that point still doesn't make sense).

Imagine you had drawn lines from the nodal point to places on the image, and wherever  you stand about 30 feet from the wall, you are on those lines.  If you adjust the yaw of the camera to point down that line from each point, you would eliminate a lot of the parallax.

Compare to a normal panorama in which you change only yaw with no sideways motion, and compare to normal stitching of a long mural etc. in which you have only sideways motion and not a change in yaw.  If you could exactingly change both together to create the point of view of the hypothetical nodal point, you could greatly reduce parallax.

Having never tried that, I don't then know the best path through hugin optimize.  You would have a panorama that could stitch very well once the correct yaw and Tx values are computed, but the optimizer will still see Tx and Yaw as locally ambiguous.  It might take some manual tweaking to get the values close to what you know you did with the camera, then optimization could fine tune.

In stitching from that, the parallax would be largest on the seams where stitching is choosing between two different images.  Hopefully, either automatically or with some masking, the actual seams could be along contours that don't show the parallax.

All that might be too much work.  It is just an idea.

[Bruno Postle]

Some notes I wrote on this eleven!! years ago: https://www.flickr.com/photos/brunopostle/5830006193/

Shooting from the same distance is useful, as is shooting perpendicular to the surface/wall. Though shooting from an oblique angle where necessary is ok, probably this would be better if this is consistent, i.e. turn to the left at the left end of a row, straight forward towards the middle and turn to the right towards the right end.

Most important is to catch openings in one shot, or if the opening/road/alley is too wide then the multiple shots need to be from a single location (like a 'normal' panorama). The same for any features that are out of plane such as roofscapes above or trees in the foreground.

Control points need to be entirely in the same physical plane, you will need to manually delete any that are not. This whole technique only works where the buildings closely follow the road edge.

You need to assign the centre photo as the 'anchor', Hugin doesn't automatically do this for the mosaic XYZ parameters, so you will need to use the Optimiser tab and select parameters to optimise rather than using the Assistant tab.

Use Rectilinear format for the output.

Optimise XYZ for all photos except the anchor, optimise rpy for all photos. Though unless you have vertical/horizontal control points you will need to fix roll for the anchor.

Optimisation of XYZ mosaic parameters is not as stable as a 'normal' panorama, if you have problems, work your way out from the middle anchor photo, adding photos one at a time to the optimisation and adding/deleting control points as necessary.

[Alexander Drecun]

The idea of working of the hypothetical nodal point in order to create component images was because I wasn't sure I could achieve the final image I want just by changing the camera's yaw, primarily out of a concern for excessive parallax issues. I'm trying to avoid obvious indicators of photo stitching as much as possible. But maybe Bruno's suggestion is the right way to approach this - adjusting the camera's yaw at the same time rotating it left incrementally the further left I go and incrementally further right the further right I am.