Building Computer Vision Projects with OpenCV 4 and C++

Image feature matching

SfM, as presented in the last section, relies on the understanding of the geometric relationship between images as it pertains to the visible objects in them. We saw that we can calculate the exact motion between two images with sufficient information on how the objects in the images move. The essential or fundamental matrices, which can be estimated linearly from image features, can be decomposed to the rotation and translation elements that define a 3D rigid transform. Thereafter, this transform can help us triangulate the 3D position of the objects, from the 3D-2D projection equations or from a dense stereo matching over the rectified epilines. It all begins with image feature matching, so we will see how to obtain ...