Find world 3D coordinates from a pair of image correspondence from left and right views
Use depth_finder.py for this purpose. There are also some other example scripts (depth_tester*.py) that demonstrate the usage. Needs camera pose and camera calibration data. Input is left and right (x,y) image co-ordinates of the object for which real-world (X,Y,Z) co-ordinates are to be computed.
When you compute depth from stereo left and right images, the co-ordinates that you get are in the camera co-ordinates system. This may work for most applications where output is used on a computer monitor (which would match camera co-ordinate system). However, to compute real world co-ordinates -- for example, to treat the ground floor as y=0 plane, and a specific point on floor as the real-world origin, and compute all co-ordinates accordingly -- we need to compute camera pose and use that to compute real world co-ordinates from camera co-ordinate system.
In order to compute the real world co-ordinates, we need to know the ground plane (or whichever plane you want to use as ground plane). This requires a special calibration known as pose estimation. Use pose_finder.py utility to compute the pose data required by depth_finder. It computes accurate camera pose using 3D points for a given pair of left and right images of a calibration target.
Use stereo_calibration.py to compute intrinsics and extrinsics needed by the main depth_finder script. This is