Provably Correct Algorithms for Multiframe Structure from Motion: the Case of Constant Translation Direction

We have recently demonstrated a new approach to multiframe structure from motion from point features which, in the appropriate domain, provably reconstructs structure and motion correctly. The domain is one well suited to outdoor robot navigation scenarios where perspective eeects are large. In this paper, we describe a version of our approach adapted to an important special case of motion: translational motion approximately in a constant direction (but not necessarily of constant magnitude) with arbitrary rotations. Experimental results are presented for real and synthetic image sequences.tion, low level vision. Abstract We have recently demonstrated a new approach to multiframe structure from motion from point features which, in the appropriate domain, provably reconstructs structure and motion correctly. The domain is one well suited to outdoor robot navigation scenarios where perspective eeects are large. In this paper, we describe a version of our approach adapted to an important special case of motion: translational motion approximately in a constant direction (but not necessarily of constant magnitude) with arbitrary rotations. Experimental results are presented for real and synthetic image sequences.