Stereo vision involves extraction of depth information from two different views of a scene obtained by two different cameras.
After calibration (find cameras relative position), and then rectification (project images to a common plane) corresponding image points can be found by minimizing cost function, which in simplest case summarises differences in pixels intensities in neighbourhood of two points.
So by comparing these two images, disparity map can be computed which, for this stereo pair cointains differences in horizontal coordinates of corresponding points. Values in disparity map are inversely proportional to the scene depth.
Stereo vision is similar to human binocular vision. Objets which are closer to our eyes have larger relative shift, then the ones further away.
