Visual Shapes of Silhouette Sets

1
Visual Shapes of Silhouette Sets
J. S. Franco, M. Lapierre, E. Boyer

• Wonwoo Lee
• 2006. 07. 28
• GIST U-VR Lab.

2
Introduction

• Shape from Silhouettes
• Extensively used in modeling dynamic and
  non-rigid objects
• Visual hull
• The maximal solid shape consistent with the
  silhouettes
• Not always a good approximation, when observing
  smooth surface with few cameras

3
In This Paper

• Purpose of the paper
• To find better approximations of an object shape
  given its silhouettes
• Keeping the ability to model in real-time
• Visual Shapes
• Silhouette consistent shapes
• Projected silhouette boundaries match the given
  silhouette contours

4
3D Reconstruction with Visual Hull
Camera calibration
Image Acquisition
Background Subtraction
Silhouettee detection
Visual Hull computation
3D Model Generation
5
Preliminaries

• Viewing cone
• The closure of viewing rays
• Visual hull
• The intersection of viewing cones
• Viewing cone strip
• Contributions of a silhouette contour to the
  boundary surface of the visual hull
• Viewing edges
• Contributions of viewing rays to the visual hull
  surface

6
Visual Shapes

• Let S be a set of scene silhouettes associated to
  a set of viewpoints C.
• The Visual shapes V(S, C) of S and C are space
  regions V such that
• All rim points on the surface of V belong to
  viewing cone strips of S
• All viewing cone strips of S are tangent to V

7
Computation of Visual Shapes

• Procedure
• Identify the contributions of the visual shapes
  inside the viewing edges
• Estimate the surface connecting the contributions
  
• Constraints
• Thining viewing cone strips
• Single contact point
• Local surface with order 2

8
Contribution along viewing edges

• Thining viewing edges
• Calculate viewing edges by projection
• Reconstruct visual shapes by thining the edges
• Single contact point
• Choose a point inside a viewing edges
• Assumption Photo-consistency, etc.
• Unpredictable local behavior
• Local surface with order 2
• Predictable local behavior

9
Contribution along viewing edges
Local order 2
Single contact point
Thining viewing cone strips
Visual hull
10
Surface Reconstruction

• Interpolate 3D points obtained from previous step
  to reconstruct shape of an object
• Delaunay tetrahedrization
• Generate 3D mesh from a 3D point set by using
  tetrahedrons
• Carving to preserve consistency with silhouettes
• Visual shapes are the union of the tetrahedrons
  consistent with all the input silhouettes

11
Local order 2
Single contact point
Thining viewing cone strips
Visual hull
12
Conclusion

• Visual shapes
• A class of silhouette consistent shapes
• Useful to characterize shapes that project onto a
  set of silhouettes
• Robust to reconstruct smooth surfaces
• Future work
• Visual shapes over time sequence
• Photometric constraint