Determine object world coordinate by two or few cameras

edit

Look at the stereo-vision methods (Google it) or the OpenCV implementation StereoBM or StereoSGBM. You also could look at the samples: cpp/stereo_match.cpp or python2/stereo_match.py or gpu/stereo_match.cpp or gpu/stereo_multi.cpp in the OpenCV trunk.

If you have more than two camera, epipolar geometry can be hard to understand... You could look at the Shape From Silhouettes (the principles) and a lot of details here. It allows you to reconstruct a 3D object from multiple views, which could be a solution for you to find the 3D coordinate.

Otherwise, almost in the same way, you could look at Space Carving which used photometric consistency for objects reconstruction.