Building Computer Vision Projects with OpenCV 4 and C++

Feature extraction and robust matching for panoramas

We create panoramas from overlapping images. In the overlapping region, we look for common visual features that register (align) the two images together. In SfM or SLAM, we do this on a frame-by-frame basis, looking for matching features in a real-time video sequence where the overlap between frames is extremely high. However, in panoramas we get frames with a big motion component between them, where the overlap might be as low as just 10%-20% of the image. At first, we extract image features, such as the scale invariant feature transform (SIFT), speeded up robust features (SURF), oriented BRIEF (ORB), or another kind of feature, and then match them between the images in the panorama. Note ...