Iterative cage-based registration from multi-view silhouettes

  • Authors:
  • Yann Savoye

  • Affiliations:
  • -

  • Venue:
  • Proceedings of the 10th European Conference on Visual Media Production
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

Dynamic shape capture from casual videos is a fundamental task at the cross-fertilization of Computer Vision and Computer Graphics. Notwithstanding, recent advances in low-cost dynamic scanners turn the cross parametrization of non-rigid animatable surface into an ill-posed and vision-oriented problem. In this paper, we propose a cage-based technique to register non-rigid observed shapes using a meaningful, reduced and animator-friendly embedding. This subspace offers natural silhouette-awareness to encode the deformation complexity already encapsulated in the targets. The estimated time-varying parameters associated the underlying flexible structure allows potential reuse. In particular, we leverage the problem of highly non-rigid spacetime registration by employing an elastoplastic coarse cage. Thus, we perform scalable handle-aware biharmonic shape registration, relying on the high-level of shape abstraction offered by this space-based paradigm. Finally, we tested the effectiveness of our proposed solution on real-world datasets capturing time-varying multi-view silhouettes.