Hi AlexC.,
When I think back, two decades ago, you asked for a staircase for a theater set in the same city. With your vast production experience, I know you have all the parts needed to get this done, but to get the best out of the Calibrator, please allow me list it clear some points, just in case.
The Camera calibration tool has an option to set the size. Like take any door, it is rarely below 200cm, or way above 260cm, of course, exceptions for large doors are given. The counter in the last image might be 80-85cm, or a chair around 45cm. Things like that help me.
The Calibrator takes all the data and tries to make sense out of it. Meaning, sloppy placements, or not providing X, Y, or Z, axis (shift click on the Lines until R, G, or B, appears (see world axis of the project), will not lead to good results.
If one axis has a measurement, it works for the whole model. However, placing axes that have little perspective value might change the precision.
Perspective lines that would cross soon and at a large angle, are more precise. If the angle is small, the possible crossing point (in distance) of the two lines will be more more "fuzzy". Exploration: Draw two lines on a piece of paper with 3º and one pair with 90º, define then where their distant crossing point will be, they often do not cross in the image. Is that always simple? It's not.
Another point is that all perspective lines that are horizontal typically indicate eye or lens level (Tilt-Shift lenses might tell you otherwise, or lenses with strong and/or mixed distortion.) But those are a good orientation to check if the established camera is OK.
As mentioned, crops from images, or "lens-un-distorted images, are less useful, or not useful at all, distorted images with a lens profile are useful.
Not knowing the sensor size and focal length leaves one in the dark, except when being able to read what lens was used, even roughly, might be supportive. When you get no set survey either, then the people providing it have no pro-level delivery in mind, sorry to say that, but that is just not really how it works on a certain level.
Always ask yourself: Is it plausible with what the Calibrator came up with? If not, start over, there is often not a quick fix.
In short, when we turn the world (3D) into an image (2D), we lose information and rectangles are no longer drawn as 90º geometries (Exept we look in a perpendicular way to it, and typically when in the center. If we correctly project those trapezoids to a polygon, then the show up as a rectangle again, as we move them from tehir 2D "frozen" state into a 3D again.
In the end, if you do not define a pin, the model is somewhere in space.
All the best
Edited for clarity 07:21 pm Pacific.
