In one embodiment, a method comprises determining an initial set of feature correspondences for a pair of images of a three dimensional (3D) space; identifying one or more first regions in one of the pair of images and one or more second regions in the other one of the pair of images. The method comprises identifying additional feature correspondences by matching first regions and second regions in the pair of images and generating a motion of a camera in the 3D space responsive to the initial set of feature correspondences and the additional feature correspondences, wherein the camera captured the images. Each of the first regions and the second regions potentially maps to a plane in the 3D space corresponding to the images. Additionally, each of the first regions and the second regions includes at least one feature in the initial set of feature correspondences.