It would seem that you have a good idea of ββwhat is required. It might be a little worm - so I would recommend getting the famous Hartley and Sisserman book for a canonical explanation. Here is a link to the corresponding chapter.
But in a nutshell ...
I did not use the opencvstereovision shell class directly, but it looks like it took the headache out of calibration (internal and external cameras) and calculated straightening through the homography matrix (H) for planar geometry or the fundamental matrix (F) for more complex epipolar geometry.
Probably similar to this original post .
This means that this correction means that it has established a mathematical comparison between the same point in each image.
In the previous answer (from me) you can do the math using the Fundamental matrix to perform triangulation - i.e. distance calculations.
However, note that this distance is only in the frame of the image coordinates (i.e. in pixels).
What is really needed to take measurements of the βreal worldβ (ie, the actual physical distance) is the calculation of the main matrix (E), which combines the main matrix and the internal properties of the cameras (K) with, if you want, project the distances into the real world.
source share