Integrating Occlusion Culling with ViewDependent Rendering

1
Integrating Occlusion Culling with View-Dependent
Rendering
2
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

3
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

4
Motivation

• Goal
• Realism requires detailed representation
• Interactive rendering of three-dimensional
• Problem
• Visualization of large dataset
• Computation power
• Memory capacity

5
Previous work

• Multi-Resolution Rendering
• Merge trees Xia, El-Sana and Varshney
• Progressive meshes, view-dependent rendering
  Hoppe
• Vertex hierarchy Luebke and Erikson
• Multi-triangulation De Floriani
• View-dependent trees El-Sana and Varshney
• Illumination light dependent refinement Klein
  and Schilling
• Progressive transmission Gueziec et al

6
Previous Work

• Occlusion Culling Algorithms
• Lists of potentially visible geometry Airey
• Volumetric representation Sillion
• Prioritized-layered projection Klosowski and
  Silva
• Z-buffer visibility Green et al
• Occlusion maps Zhang et al
• Viewspace partitioning Cohen-Or et al
• Hardly Visible Sets Andujar et al

7
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

8
View-Dependent Rendering

• The Idea
• A technique that allows several levels of details
  to co-exist across different regions of the same
  object
• Requirements
• Appropriate levels of detail selection
• Levels of detail should merge seamlessly

9
View-Dependence Trees

• Construction
• Encode various level of detail representations
• Based in vertex-pair collapse operation.
• Bottom-up construction
• The collapsed pairs are selected based on a
  simplification metric
• User implicit-dependencies to prevent foldovers

Split
Merge
10
Level-of-Detail Selection

• View parameters
• Illumination parameters
• Back-facing culling
• Emphasize Silhouette
• Screen-space projection

11
Levels of Detail

• The various level of detail are represented by
  breadth cut
• Determination active nodes and triangles
• Regions close to the roots corresponding to low
  resolution
• Regions close to the bottom corresponds to high
  resolution

12
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

13
Occlusion Culling

• Often large fraction of the model polygons do
• not contribute to the final image.
• Determine hidden polygons
• Render an estimation of visible set
• Work on a budget

e
d
b
c
a
14
Prioritized-Layered Projection

• Preprocessing
• Tessellation of space
• Finding list of primitives for each cell
• Computing cells solidity
• Rendering
• Projecting front cells
• Computing solidities for neighbors

15
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

16
Motivation

• View-dependent rendering reduces resolution based
  on view-parameters.
• Occlusion Culling removes hidden polygons
• Still
• View-Dependent Rendering renders occluded
  triangles
• Occlusion Culling renders small triangles

17
Our approach

• The combination of view-dependent rendering
  with occlusion culling

18
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

19
Preprocessing

• Impose 3D grid over a dataset
• Determine the set of polygons in each cell
• Assign a solidity value for each cell 0.0 1.0
  Based on two approaches
• Face projection
• Ray shooting

20
Face Projection

• Projection polygons on faces of the cell
• Ratio between projection areas and total area of
  faces

21
Ray Shooting

• Shoot rays into a cell
• Determine which ray manage to leave the cell
• Ratio between passed rays and total number of rays

22
Run-Time Processing

• View-dependent rendering reduces the resolution
  of far-from-viewer regions.
• Hidden and close polygons are still in high
  resolution.
• Occlusion probability helps to recognize
  invisible regions.
• Reduce the selected resolution for regions with
  respect to their occlusion probability.

23
Occlusion Probability

• Occlusion probability (OP)
• depends on solidity of cells.
• From viewpoint to a vertex
• Initialized OP to 0.0
• Accumulate the visited-cells solidity
• Stop when OP reaches 1.0

24
Spheres

• Four spheres Solidity values
  Occlusion
• dataset by point
  size probability
• by color

25
Adaptive Level-of-Detail

• View-dependent rendering
• Occlusion probability
• Adding the visibility as parameter in selecting
  level-of-detail
• LODfinal (1 - OP) LODview OP LODlowest
• OP occlusion probability

26
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

27
Results

• 650K triangles
• 74K triangles

28
Results

• 140K triangles 86K triangles 24K
  triangles
• Full resolution View-dependent
  View-dependent
• model rendering
  
• occlusion culling

29
Results

• 366K triangles 296K triangles 162K
  triangles
• Full resolution View-dependent
  View-dependent
• model rendering
  
• occlusion culling

30
Results
31
Outline

• Motivation
• View-Dependent Rendering
• Occlusion Culling Algorithms
• Integration
• Implementation
• Results
• Conclusions

32
Conclusion

• Integration view-dependent rendering with
  occlusion culling
• Reduce the number of rendered triangles
• Based on estimated occlusion probability
• Selection of appropriate level
• view-parameters
• occlusion probability

33
Acknowledgements

• The Stanford Computer Graphics Laboratory
• Reviewers

Thank You