Algorithms for Estimation of Three-Dimensional Motion

We derive robust algorithms for estimating parameters of motion of rigid bodies that are observed by a television camera. Motion may be three-dimensional, containing both translational and rotational components, but the observations using the television camera are two-dimensional, i.e., projections on the camera plane. Our algorithms do not require a priori knowledge of any corresponding points in three- and two-dimensional spaces. We give both recursive as well as nonrecursive algorithms that minimize the error in intensity by using the estimated motion parameters. Our theory has applications in interframe coding, computer vision, and computer animation. The efficacy of our methods and the quality of the estimation procedures must await experimental verification.