Dear Janez

Stereo-vision is the fundamental problem in the science of photogrammetry, which is classically concerned with 3D mapping from images. Although photogrammetry and computer vision are overlapping in many shared goals, the mathematics of 3D computation is well researched in photogrammetry and it can be done in many ways. For example, the computation of 3D point can be approached through the relative orientation in 3 steps by computing the relative orientation parameters between a pair of two images, then using theses parameters to do space intersection in the model space, then followed by 3D similarity transformation to compute the 3D coordinated in the object space. On the other hand, the 3D point computation can be approached by the bundle adjustment in one step. In general, the mathematical models of Photogrammetry are non-linear.

Please check the standard texts in photogrammetry and the available literature in the internet.

With best regards,

Gamal

Thank you very much for your answer. I read a lot of literature for stereo vision based on two cameras, and I understand the basic principles. I was unable to find a clear explanation of how to use multiple cameras for better accuracy of calculated disparity maps. Can you maybe suggest me some and point me to the correct direction? Thank you.

Check [1] which is the standard reference and has a chapter on N-view geometry. It covers from the basic single geometry up to n-geometry in a very thorough way.

Regards

[1] Multiple View Geometry in Computer Vision2nd ed. by Richard Hartley and Andrew Zisserman, Cambridge University Press, March 2004