There are many reads regarding stereo (computer) vision. The most notable message to take away from these reads is that you are pretty much stuck with OpenCV... which is definitely not a bad thing.
Most articles are about calibrating through the OpenCV functionalities to compensate for poor/cheap lenses.
Right now I am more concerned about the mechanical calibration of the camera positioning on its backplane. There is only so much you can compensate through calibrations, which is why I want my positioning of the camera's to be near perfect.
I am screwing 2 Logitech 9000's on an aluminum backplate about 10cm apart. My goal is to do some depth perception in the range of 2-40 meters.
How would I go about an check if the camera's are positioned correctly in relation to eachother? I was thinking of pointing the camera at a large distance (> 500m) and making sure that the center (pixel) of the camera is looking at the same (part of) the scene. Is this enough? Should I also pay some attention to what the is displayed in the corners, or is that done later through lens calibration?
Also, on an somewhat related note: Eventually I need to be drawing a disparity map. There are countless guides out there on OpenCV using 'the fast' and 'the slow' algorithms. I will be using 'the fast' one in realtime, which needs to be finetuned using a lot of parameters. Is this a one time tune for the camera type / lenses, or should I do this for every (new) scene? In case of the latter it would make my stereo camera a lot less usefull in practice.