Relative Camera Rotation from a Single Oriented Correspondence

Steven Mills

Abstract: A classical problem in computer vision is the estimation of geometric information from point correspondences, and from minimal sets of correspondences in particular. The number of correspondences required can be reduced by additional information, and recently the orientation estimates produced by feature detectors such as SIFT, SURF, and ORB, have been shown to be useful in this regard. This paper presents a new minimal solution to estimating purely rotational camera motion, and shows that a single oriented feature correspondence is sufficient for this task. The proposed method is shown to outperform both general four-point homography estimation and two-point rotation estimation when estimating rotational homographies from handheld image capture.