Master thesis proposal 2008-2009
Animated mesh construction from scans of articulated objects


Scan of an articulated model
 The resulting mesh
(Images: Laurent Belcour, Eric Heitz and Agnès Masson-Sibut)

Advisor

Franck Hétroy
LJK/EVASION
E-mail : Franck.Hetroy@imag.fr
Tel. : 04 76 61 55 04

Context

This project follows a ENSIMAG end-of-year project (results are here) done during May and June, 2008. The final goal is to create a virtual animation of any real articulated 3D object. To do this, ENSIMAG provides a laser 3D scanner, which gives a virtual representation of any object, as a mesh. The goal of this Master thesis is to develop a method to get, starting from an articulated object that has been scanned under different poses, a mesh animation between the poses.

Pre-requisites include interest and skills both in animation and in geometry processing (for instance, MoSIG's second year courses "Computer Graphics" and "Computational Geometry").

Objectives

Several issues need to be tackled during the thesis. First, when scanning several times the same object, there is no coherence between constructed meshes: vertex positions and neighborhoods differ, and even their number can vary. Second, simple linear interpolation between two shapes can generate several kinds of well-known inconsistencies, such as volume loss (example: point-to-point interpolation between a forearm in upper position and a forearm in lower position, with the arm staying static: the shape goes through an intermediate position, in the axis of the arm, where the forearm disappears). Finally, a laser scanner hardly constructs a "fine" mesh (that is to say a 2-manifold); usually holes remain in areas that the scanner was not able to reach. A pre-processing step is thus necessary. For this, one of the numerous bibliographical solutions can be applied (for instance [4] or [5]).

Once the pre-processing done, the first stage of the work will be to investigate the matching between the meshes of the object's different poses; the most automatic possible method will have to be set up. To do so, the fact that the object is articulated can be used; good starting points can be the techniques recently introduced by de Goes et al. [2] and Mateus et al. [3]. Then, an interpolation method between the different scanned poses need to be defined. This method should preserve the volume and deform the model in a realistic way. An adaptation to meshes of techniques recently proposed by Adams, Wand et al. [1,6] could be studied.

Development will be done in C++; the Graphite library, developed at INRIA Nancy - Grand Est, can be used.

Keywords : reconstruction, mesh animation.

References

  1. B. Adams, M. Ovsjanikov, M. Wand, H.P. Seidel, L. Guibas. Meshless modeling of deformable shapes and their motion. Symposium on Computer Animation, 2008.
  2. F. de Goes, S. Goldenstein, L. Velho. A Hierarchical segmentation of articulated bodies. Symposium on Geometry Processing, 2008.
  3. D. Mateus, R. Horaud, D. Knossow, F. Cuzzolin, E. Boyer. Articulated shape matching using Laplacian eigenfunctions and unsupervised point registration. IEEE Conference on Computer Vision and Pattern Recognition, 2008.
  4. J. Podolak, S. Rusinkiewicz. Atomic volumes for mesh completion. Symposium on Geometry Processing, 2005.
  5. A. Sharf, M. Alexa, D. Cohen-Or. Context-based surface completion. ACM Transactions on Graphics (SIGGRAPH Proceedings), 2004.
  6. M. Wand, B. Adams, M. Ovsjanikov, A. Berner, M. Bokeloh, P. Jenke, L. Guibas, H.P. Seidel, A. Schilling. Efficient reconstruction of non-rigid shape and motion from real-time 3d scanner data. Submitted to ACM Transactions on Graphics, 2008.