In my previous project, image stitching project, I found the size of keypoints will have an impact on the execution time and the quality of the stitching. No matter what feature detection methods, SIFT or SURF, we find that if we have too many keypoints, it also will spend much time to generate the descriptors of features as well as the matching of them. Moreover, the outliers, some bad ones of the keypoints will exacerbate the perspective transform matrix. The task becomes much more complicated. Ultimately, I found a useful tool inspired by the findHomography method of OpenCV. When I first met this method, I was totally fascinated by it and that's why I want to write about it.
This is the first post of the year 2017.