LIP6 2008/001

• Reports
  Fusion de données spatiales (3D) et temporelles (2D+t) pour la reconstruction de scènes 3D+t : deux approches
• W. Rekik, D. Béréziat, S. Dubuisson
• 23 pages - 06/12/2008- document en - http://www.lip6.fr/lip6/reports/2008/lip6-2008-001.pdf - 187 Ko
• Contact : Jean-Marie.CHESNEAUX (at) nulllip6.fr
• Team : PEQUAN
The main focus of this paper is 3D+t shape recovery from 3D spatial data framing 2D+t temporal sequences. Merging these two kind of data yields a temporal sequence describing 3D dynamic of structures of interest by reconstructing the original volume from each 2D frame. 2D to 3D inverse reconstruction is a complex and ill posed problem due to the great deal of distortion and in-depth information loss generated by the projective view in each 2D frame. To handle this critical lack of information, we model relevantly the 3D-to-2D transformation that reduces each 3D image into a 2D one and we match each frame with the 3D image known at a previous time. Assuming that structures evolve scarcely between a couple successive moments, these images are matched by a displacement vector field measuring voxel position variation meanwhile. We embed the luminosty conservation constraint and we propose two approaches to estimate voxel movement over time. We reconstruct gradually the 3D+t sequence by motion compensation. Some applications observing in different modalities 2D+t and 3D data can benefit from this study such as microscopical cell biology or medical imaging.


• Keywords : 3D reconstruction, motion compensation, optical flow constraint
• Publisher : Jean-Marie.Chesneaux (at) nulllip6.fr