Invited Talk February 2007
LASMEA laboratory in Clermont-Ferrand, France will be giving an invited talk in the oberseminar room MI 03.13.010
"Modeling and Reconstructing Paper From Multiple Images"
Modeling and Reconstructing Paper From Multiple Images
Recovering the threedimensional scene surface from images is a challenging computer vision problem. We tackle the case of paper-like surfaces, for which we propose a generative, direct model. This model is governed by intuitive, physically meaningful parameters. The generated surface satisfies the physics of a smoothly bent paper, namely inextensibility and vanishing Gaussian curvature.
We demonstrate the model's representational power with a multi-camera threedimensional reconstruction algorithm. This includes model-based bundle adjustment minimizing a point-based reprojection error. The method naturally handles self-occlusion. We show convincing surface texture recovery and editing in various image sequences.
Since september 2005, Mathieu Perriollat is a PhD student at the LASMEA lab, in the team ComSee
of the GRAVIR group, under the supervisation of Adrien Bartoli, Jean-Marc Lavest and Maxime Lhuillier. He is working on `Structure-From-Motion' for non-rigid scenes.
A Single Directrix Quasi-Minimal Model for Paper-like Surfaces