Don’t wanna say this is impossible, I’m not a programmer, but that sounds like it’s really difficult to do and would turn out unreliable, even with good AI.
Imagine a photo of a big, open empty space (this for example) with very few landmarks and all, how would you measure anything in it without knowing the distance between two objects at least? Also the field of view can be stretched for example, like when it’s a wide angle camera or dashcam that the pictures were taken with. This was taken with a wide angle dashcam for example, you can see how the railing stretches on the right side.
Again, not saying its impossible to recognize for a computer program, but I think very much of development power has to go into it to be at least somewhat accurate and even then I personally wouldn’t trust it to be 100% accurate. And if its just accurate by ±1m for example I don’t really see the use for it, as you can probably approximate such a rough estimate yourself by looking at the photo details yourself.