the question about binocular camera

[zhouq123]

hi, I can understand your camera_calibration procedure for one camera. And I want to use the result for binocular camera three-dimensional reconstruction base on triangulation measurement. Firstly, I want to calibrate to get RT between two cameras, it's easy if I use the central generic model, because the optical central position of camera is only one. But for the noncentral generic model, I have no idea how to calibrate the pose of two cameras, and how to use constraint of polar line? Could you do me a favor?

[puzzlepaint]

Would it be an option to calibrate the two views as separate cameras within a camera rig, using the central-generic model for each view, instead of using the noncentral model? Then the relative pose between the views would get calibrated as well.

With the noncentral model, doing stereo is a bit more complicated, but still possible. But it is not possible to directly use the concept of an "optical center" or "point depth" then. What is still possible is for example:

• For a depth estimation algorithm like PatchMatch Stereo, consider a point somewhere on a viewing line of one camera, and project this point into the image of the other camera. Then a matching cost can be computed for both pixel locations.
• Given a pixel in one image that was matched to a pixel in another image, obtain the viewing lines for both pixels and compute the 3D point that is closest to both lines to obtain a triangulated point.