multi-perspective panoramas

[sesshin]

I have heard this mentioned before, but was wondering if anyone had any experience stitching together images taken from multiple perspectives to form a panorama.

Specifically, I've been tasked with photographing a mural on the side of a building and it is very wide but not that tall, about 250 feet wide by 20 feet tall. The client has requested the image be as high resolution as possible while also forgoing perspective distortion as much as possible.

My first instinct is to line the camera up with the wall so that the perspective is straight on and encompasses the height, and then move down the side of the building taking photos to fit the whole thing in. From my initial tests though, while the mural lines up nicely, the parallax of objects outside of the mural quickly becomes an issue.

Does anyone else have any examples of ways this has been accomplished? I imagine retouching would have to be a big part of making it work

[Bart_van_der_Wolf]

I have heard this mentioned before, but was wondering if anyone had any experience stitching together images taken from multiple perspectives to form a panorama.

Specifically, I've been tasked with photographing a mural on the side of a building and it is very wide but not that tall, about 250 feet wide by 20 feet tall. The client has requested the image be as high resolution as possible while also forgoing perspective distortion as much as possible.

Hi,

This is a relatively straightforward function of competent pano stitching software like e.g. PTGUI (requires Pro version!) or Hugin, as long as the subject is perfectly flat (no significant protrusions or recessions, or things sticking out in front). PTGUI calls it Viewpoint correction, and assumes that the subject is in a flat plane, but viewed from different positions.

This allows to shoot multiple sub-segments of the mural with a longer focal length for higher detail, and later stitch them together, without the need for rotation calibration or exact shooting position information. All you need to make sure of is that each image has enough DOF for good focus when shooting the surface at an angle, and sufficient overlap with features between the tiles to allow the automatic creation of control points that correspond to both of each overlapping source image(s).  Using a larger distance and longer focal length allows more perpendicular surface orientation, but may also risk the occlusion by lamp posts/powerlines/parked cars/street furniture/etc. that need to be masked out for the stitch).

Here is a link to an example from the first pano stitcher that featured this type of shooting, others soon followed suit. Here is the tutorial for PTGUI, with the example mainly focused at creating a Nadir image component in a 360 degree VR scene, so a much more complex scenario than you have.

Based on the required output size/resolution, you need to calculate how many tiles+overlap are required for the total stitch job. It can also be turned into a zoomable project for e.g. a web page, but that requires other software to generate the HTML content and tiles for responsive internet access.

Microsoft's Image Composite Editor (ICE) can also handle such tasks automatically, but offers less intervention possibilities that may be needed for a specific usage of the image.

Cheers,
Bart

[sesshin]

Thanks for the response! Very useful information.

[Ajoy Roy]

I have heard this mentioned before, but was wondering if anyone had any experience stitching together images taken from multiple perspectives to form a panorama.

Specifically, I've been tasked with photographing a mural on the side of a building and it is very wide but not that tall, about 250 feet wide by 20 feet tall. The client has requested the image be as high resolution as possible while also forgoing perspective distortion as much as possible.

My first instinct is to line the camera up with the wall so that the perspective is straight on and encompasses the height, and then move down the side of the building taking photos to fit the whole thing in. From my initial tests though, while the mural lines up nicely, the parallax of objects outside of the mural quickly becomes an issue.

Does anyone else have any examples of ways this has been accomplished? I imagine retouching would have to be a big part of making it work

I have used Microsoft ICE and it does an excellent job of flat object, as long as the horizontal angle is contained within 90 degrees. The lesser the better. Here are some tips
. Find out how much distance you can have between the camera and the mural. The longer (to some extent) the better, as you will have lesser distortion to correct.
. Calculate the image resolution you will need. I guess that for 250 feet (76200mm) if you use 10mm, then you need 7,600 pixels, if 1mm then 76,200 pixels. Note that stitching and image correction will require some down sampling. Similarly for 20 feet you will need 6100 pixels at 1mm resolution
Now the calculations
. Assume 1mm resolution
. Assume 24MP sensor - 6000x4000
. Then number of ROWS without overlap = 1.5 in landscape or 2 with overlap
. Number of horizontal shots = 76200/6000 = 12.7 without overlap. With 50% overlap ~ 25.
. Each shot will capture 19.7' (6m) of the wall.
. If you are using an FX sensor - 24mmx36mm, then the magnification required is around 0.006x (1/167), so you can calculate the distance required for the lens focal length you choose.

Once you have the images, you just use Microsoft ICE to line them up. Then choose different perspective and camera motion options to get the best possible linearity