I’m doing something about binocular vision, but I don’t want to use it for ranging. I need to match the two images of the binocular camera and select one of their respective channels to synthesize a new image. Now that I have completed the stereo correction, what should I do next?
It can be clearly seen from the chessboard diagram that their rows have been well spliced together, but the columns are not aligned.
why should they? the cameras are like eyes. they’re beside each other. if both look at infinity (i.e. exactly parallel), anything “nearby” will simply not overlap.
if you want the cameras to be cross-eyed, and the images to coincide for a plane at one fixed distance only, you’ll have to either modify the camera/pose matrices (manually, i.e. by moving principal points) or just perform feature matching and findHomography once to get the transformation.
Thank you for your answer. According to your answer, I think I can improve the fitting effect by finding feature points and then finding transformation matrix.