3D Motion Estimation from Three Orthographic Views without Matching Constraints or Brightness Gradients

An important task in computer vision is to estimate the 3D motion of rigid objects from multiple views. Conventional methods typically rely on point correspondences or image brightness gradients. The underlying matching and brightness constraints are generally not met in practice which limits the accuracy of these techniques. In this paper, we present a new approach that performs 3D motion estimation directly from trinocular point clouds under orthography. The method requires no correspondences or brightness gradients which constitutes a major advantage over existing approaches. Results are presented which highlight the estimation accuracy of the proposed method with respect to conventional correspondence-based techniques.