yeah I think you are right, I have some ideas I'll try to play around with. If I use an homography matrix to compensate for the keystoning then I can compare the overlap of the two flatten images
what I think they mean is, since they're trying to find a nice stereo pair, is that there has to be some overlap between the two images, more than 40% actually
the paper I'm trying to implement doesn't say anything more apart from "The fractional overlap, in the range [0, 1], of the image areas for the framesi and j, computed based on their viewing frustums"