1. NCS 2009: Workshop on Image Processing, Computer Graphics,
and Multimedia Technologies(ICM)
Adaptive Continuous Collision Detection for Cloth Models using a Skipping Frame Session
黃 世強 ( Sai-Keung Wong )
Department of Computer Science
National Chiao Tung University
Paper ID: 16

2. Contents Introduction Background
The pipeline of continuous collision detection
Self-collision detection
Analysis and discussion
Experiments and results
Conclusions

3. Introduction Continuous collision detection?
- Perform linear (or higher order)
interpolation for two discrete frames
- Compute contact time of primitives,
such as triangles
- Triangular meshes: linear interpolation
- solving cubic equations
- computing shortest distances
- six point-triangle and nine edge-edge pairs
Frame 1
Frame 2

4. Introduction Applications:
Deformable objects
Cloth simulation
A problem: A large set of potentially colliding pairs
Slow performance
An observation: Local coherence
A large portion of colliding pairs remains the same

5. Introduction:
Local coherence

6. Introduction BVH traversal contributes a significant amount of time
Elementary test processing
BVH traversal
BVH update

7. Contributions
A novel adaptive pipeline for continuous collision detection:
BVH update
BVH traversal
A skipping frame session
Elementary tests
A partial traversal scheme
Handling triangles with large movement
Robustness: keep track of colliding pairs

8. Background: Cloth simulation
D.E. Baraff and A. Witkin, “Large steps in cloth simulation”, SIGGRAPH, 1998.
R. Bridson, R. Fedkiw, and J. Anderson, “Robust treatment of collisions, contact and friction for cloth simulation”, ACM ToG 2002.
K.J. Choi and H.S. Ko, “Stable but responsive cloth”, SIGGRAPH, 2004
A. Selle, J. Su, G. Irving, and R. Fedkiw, “Robust high-resolution cloth using parallelism history-based collisions and accurate friction”, TVCG 2008.
P. Volino and N. Magnenat-Thalmann, “Efficient self-collision detection on smoothly discretised surface and animation using geometrical shape regularity”, Computer graphics forum, 1994.

9. Background: Continuous collision detection
M. Hutter and Fuhrmann, “Optimized continuous collision detection for deformable triangle meshes”, WSCG, 2007.
T. Larsson and T. Akenine-Moller, “Efficient collision detection deformed by morphing”, Visual computer, 2003.
M.Tang, S.E. Yoon and D. Manocha, “Adjacency-based culling for continuous collision detection ”, Visual Computer, 2008.
M.Tang, S. Curtis, S.E. Yoon and D. Manocha, “ICCD: Interactive continuous collision detection between deformable models using connectivity-based culling”, TVCG, 2009.
S.K. Wong and G. Baciu, “Dynamic interaction between deformable surfaces and nonsmooth objects”, TVCG, 2005.
S.K. Wong and G. Baciu, “A randomized marking scheme for continuous collision detection in simulation of deformable surfaces”, ACM International Conference on Virtual Reality Continuum and Its Applications, 2006.

10. The Pipeline of CCD Preprocessing stage Runtime stage
Construction of bounding volume hierarchies (BVHs)
The primitive assignment
Runtime stage
BVH update with inflation, refitting each node
BVH traversal, collecting potentially colliding pairs
Adaptively apply a skipping frame session
Front-end and back-end filtering stages

11. Preprocessing stage Construction of bounding volume hierarchies (BVHs)
Build a hierarchy structure to bound each object

12. Preprocessing stage Primitive assignment q0 eB1 (a) (b)
Assign each triangle to itself
Assign each edge and each vertex to one of its incident triangles
eA0
eB0 q0
Vertex assignment A B
Edge assignment
eB1
(a)
(b)

13. Runtime stage: BVH update
Purpose: refitting each node with inflation
Estimated
movement
distance
d(p) = ( a v (p) + b v ) D t+ g l
Larger bounding volume

14. Runtime stage: BVH traversal
Purpose: Collect potentially colliding pairs
Bounding volumes overlap
Problems:
A large number of such pairs
Too far away to collide within
the current interval
Further treatment
A potentially colliding pair

15. Runtime stage: Front-end and back-end filtering stages
Front-end: eliminate redundant triangle pairs
Check estimated shortest distance
de i (T0, T1) <= dei+1(T0) + dei+1(T1) + d
dei+1(T0)
Back-end:
Perform continuous collision detection
Solve coplanar times for each primitive pair
Compute shortest distances for verification

16. Runtime stage: Skipping frame session
Purpose: Adaptively apply a skipping frame session
Advantages:
Local movement of a triangle
Lying inside its inflated bounding volume
> no need BVH update
> no need traversal

17. Handling dangling triangles
Triangles with large movement
Passing through their bounding volumes
Two solutions:
1. Traversal individually
2. Partial traversal scheme
Large movement

18. Self-collision detection
Low curvatured surface partitioning
Compute continuous normal cone for each triangle
Perform bottom-up merging to obtain low curvature surfaces
Collision detection between low curvatured surfaces

19. Analysis Consider a small box b lying inside another box B
Free movement distance: d = d(b, B)
Time step = D t
Relative speed = v
Number of skipping frames
= d / ( v D t )

20. Experiments and Results
Intel (R) Core (TM2) Quadcore CPU machine with 2.4GHz of 2GB memory
One thread for computation
Comparison with two methods and others:
NoDup :
S.K. Wong and G. Baciu, “A randomized marking scheme for continuous collision detection in simulation of deformable surfaces”, ACM International Conference on Virtual Reality Continuum and Its Applications, 2006.
R-RTI :
S. Curtis, R. Tamstorf, and D. Manocha, “Fast collision detection for deformable models using representative-triangles”, Proceedings of the 2008 symposium on Interactive 3D graphics and games, 2008

21. Experiment Set One:
< 100k triangles

22. Average Collision Detection Time Per Time Step ( sec )
Experiment Set One:
Average Collision Detection Time Per Time Step ( sec )
(including self-collision detection)
NoDup
R-TRI
nSwD
SwD
( % R-TRI )
Ani. One
Spinning ball
0.23
0.18
0.15
0.13
( 72%)
Ani. Two
Ball-cloth
0.12
0.10
0.082
0.074
( 74%)
Ani. Three
Four cones
0.31
0.27
0.16
( 59%)
Ani. Four
Garment
0.092
0.076
0.062
0.052
(68%)

23. Experiment Set Two:
> 300k triangles

24. Experiment Set Two Labels: a x f a
x : length of skipping frame session

25. Experiments and Results
Experiment Set One: Model Complexities
Rigid Objects ( #Tri )
Cloth Models ( #Tri )
Ani. One
5.2 k
97 k
Ani. Two
10 k
45 k
Ani. Three
0.5 k
Ani. Four
34 k
20 k
Experiment Set Two: Model Complexities
11 k
320 k
7 k
500 k
1 k
502 k
40 k

26. Comparison: Traditional approach vs our approach
Traditonal
Our
Bounding volume size
smaller
larger
BVH update
yes
sometimes
BVH traversal
Number of potentially colliding pairs
fewer
Memory storage
less
higher
Robustness ok
better
Speed
faster

27. Conclusions & future work
Propose a novel adaptive framework for continuous collision detection using a skipping frame session
Fast performance
High memory requirement but increased robustness
Apply to multilayered garments
Reduce storage size

28. Thank you.
Q & A.