Sequential Updating of Projective and Affine Structure from Motion

  • Authors:

  • P. A. Beardsley;A. Zisserman;D. W. Murray

  • Affiliations:

  • Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, UK. E-mail: pab@merl.com, az@robots.oxford.ac.uk, dwm@robots.oxford.ac.uk;Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, UK. E-mail: pab@merl.com, az@robots.oxford.ac.uk, dwm@robots.oxford.ac.uk;Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, UK. E-mail: pab@merl.com, az@robots.oxford.ac.uk, dwm@robots.oxford.ac.uk

  • Venue:
  • International Journal of Computer Vision
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

A structure from motion algorithm is described which recoversstructure and camera position, modulo a projective ambiguity. Cameracalibration is not required, and camera parameters such as focallength can be altered freely during motion. The structure is updatedsequentially over an image sequence, in contrast to schemes whichemploy a batch process. A specialisation of the algorithm to recoverstructure and camera position modulo an affine transformation isdescribed, together with a method to periodically update the affinecoordinate frame to prevent drift over time. We describe theconstraint used to obtain this specialisation.Structure is recovered from image corners detected and matchedautomatically and reliably in real image sequences. Results are shownfor reference objects and indoor environments, and accuracy ofrecovered structure is fully evaluated and compared for a number of reconstruction schemes.A specific applicationof the work is demonstrated—affine structure is used to computefree space maps enabling navigation through unstructured environmentsand avoidance of obstacles. The path planning involves only affineconstructions.