Building Computer Vision Projects with OpenCV 4 and C++

Warping images for panorama creation

Given that we know the homographies between images, we can apply their inverse to project all the images on the same plane. However, a direct warping using the homography ends up with a stretched-out look if, for example, all the images are projected on the plane of the first image. In the following image, we can see a stitching of 4 images using concatenated homography (perspective) warping, meaning all the images are registered to the plane of the first image, which illustrates the ungainly stretching:

To cope with this problem, we think of the panorama as looking at the images from inside a cylinder, ...