Matching: Split x and y disparity

edit

Thanks for shedding some light. During debugging and testing though I prefer to have the option available to check vertical disparities, too, to facilitate finding rectification errors. An auto-calibration feature might need something like that as well, but maybe that needs a different algorithm entirely, like surf.

Som calibration algorithms possible using planar targets with easily recognized points... e.g. checkerboard or overlapping circles, on a movable plane. The targets need to move the points through most of the in-focus volume of the cameras. Several hundred data points are necessary for sub-pixel accurate models of the camera distortion and relative poise. Other schemes are possible. I recently saw a video where individually addressable christmas tree lights were sequentially illuminated and the resulting points correlated, well enough to generate point clouds and to program the light string to make rough spatial moving light patterns without conventional calibration targets.

I see. If people go that far, I can see why they're able to eradicate vertical disparity. Thanks for explaining!