Collision Detection for Cloth Simulation Using

Bounding Sphere Hierarchy

Abdullah Bade, Ching Sue Ping, Siti Hasnah Tanalol
School of Science & Technology,
University Malaysia Sabah,
Kota Kinabalu Sabah,


.
Abstract For the past 2-decades, the challenges of collision
detection on deformable objects have attracted numerous
researchers. One of the deformable objects which have been
widely researched is cloth model. Simple mass spring model is
used to model the cloth where the movement of the particles
within the cloth was controlled by applying the Newtons second
law. After the modeling stage, implementation of the collision
detection algorithm took place on cloth has been done. The
collision detection technique used is bounding sphere hierarchy.
Then, quad tree is being used to partitioning the bounding sphere
and the collision search was based on the top-down approach. A
prototype of the collision detection system is developed on cloth
simulation and several experiments were conducted. Time taken
for this system to be executed is around 235.258 milliseconds.
Then the frame rate is at the average of 22 frames per second
which is close to the real time system. Times taken for the
collision detection system travels from root to nodes were 23
seconds. As a conclusion, the computational cost for bounding
sphere hierarchy is much higher because the bounding sphere
required more vertices for generation process, however the
execution time for bounding sphere hierarchy is faster than the
AABB hierarchy.
Keywords-component; Collision detection, bounding sphere
hierarchy, quad tree, cloth
I. INTRODUCTION
In this modern era, modeling and simulation using
computer has been commonly used since the computational
speed and the realism of the model are improving as the time
passes. Deformable object are object in which their shape will
change whenever there are external forces applying on the
object, but the shape of the object will return to its original
state when there are no other forces applied [1]. By using
mathematical and computational techniques such as Bezier
Curve, Splines, Free-form deformation and mass spring
model, deformable objects can be formed efficiently [2].
Mass spring model has been widely used for modeling
deformable object even though it is not as accurate as the
finite element models. The computational cost of this model
is not expensive, thus making it more favorable. The simple
mass spring model is created using a mesh of points
connecting each other with a spring in a regular-shaped
structure [3].
Collision detection is one of the most frequently used
techniques in computer graphics to detect the intersection
between two or more objects. Collision occurs when there is at
least one points of the object were in contact with the other
object [4]. Most of the collision detection occurs in animation,
modeling and simulation. There are many algorithms
developed to detect the collision within objects, most of the
algorithms which can apply on the deformable object were able
to perform on the rigid body [5]. The algorithms purposed for
collision detection are Sweep and Prune algorithms, Lin-Canny
Bounding Volume algorithm, Gilbert-Johnson-Keerthi (GJK)
Distance algorithm and Bounding Volume Hierarchy (BVHs).
Among the algorithms, BVHs has been proven to be one of the
most efficient data structures for collision detection [6]. There
are many types of bounding volume been studied from the past,
the examples are spheres, axis-aligned bounding boxes
(AABBs), object-oriented bounding boxes (OBBs) and
discrete-oriented polytopes (DOPs) [5].
II. METHODOLOGY
A moving ball was used to simulate the collision on the
cloth. Whenever the bounding sphere collides with the moving
ball, the bounding sphere hierarchy will be partitioned based
on quad tree concept. Quad tree is used instead of binary tree
in order to set up a faster formation. Once the distance
between the objects is smaller than the radius of the bounding
sphere, then the bounding sphere will subdivide until the
subdivision reaches the forth level which contain the smallest
bounding sphere, then the subdivision will stops and the
collisions between the objects will be detected. The nodes will
continue the division process until each of the child nodes
contains only one particle within the cloth. Then the search of
the collision was done by top-down approach meaning that the
search travelled from the root to the node. The colour of the
bounding spheres will change to red when there are collisions
detected and remained green if there are no collisions
detected.
A. Simple Mass SpringModel
Cloth is a highly deformable object which is complex to
model and require high demand on computational resources
[7]. In this paper, the cloth was created using simple Mass
Spring Model where the distance between the particles inside
the cloth was control using the spring constraint. The structure
of the cloth was formed by arranging the number of particles
along the width and height of the cloth from the point (0, 0, 0)
to (width, -height, 0). Then the position of each of the particles
was calculated based on formula (1).



Pos
Particle
= y x number of particles within width x x (1)

Where x is the column of the particles and y is the row of
the particles. Then structure of the cloth was done by forming
triangles using 3 particles was shown in Figure 1.


Figure 1. Particles arrangement for cloth structure
According to Newtons second law which state that
acceleration is equal to forces divided by mass, therefore this
law was applied on each of the particles within the cloth so
that when there are any collision, the particles will act
accordingly. However, applying just the Newtons second law,
on the particles will not create a cloth moving naturally.
Therefore, the position of the particles has to be updated in all
times. The concept of updating the position is shown in Figure
2.

Figure 2. Concepts of updating the position of the particles
Referring to Figure 2, the position of the ball will be
updated whenever there are any forces applying on the
particles which cause the particles to accelerate. The position
of the particles can be calculated using equation (2).

Pos
new
= Pos
new
+ (Pos
new
Pos
old
) x (1.0 D) + acceleration x T (2)

Where Pos
new
is the current position of the particles, Pos
old
is
the old position of the particles,D is the damping value of the
cloth preset in the program, acceleration is equal to force
divided by mass and is the size of time step to be taken for
each frame.
is the current position of the particles, is the old position of the
B. Collision Detection
The quad tree is used to subdivide the particles within the
cloth until there is only one particles contain in each of the
child nodes of the tree. In order to detect the collisions, the
radius of the bounding sphere has to be calculated and the
distance between these bounding spheres and the moving ball
has to be updated. Figure 3 shows the formation of the quad
tree.

Figure 3. Concepts of updating the position of the particles
Quad tree partition was applied on the cloth in order for
each of the particles was surrounded by a sphere for the
collision detection. This has to be done in order to detect the
collision more accurately.
The bounding sphere was created, where the radius of the
sphere is differing for each level of the quad tree. Therefore,
the algorithm to getting the radius is shown in Figure 4.
1. Figure 4. Algorithm to
get the radius of the
sphere.



Figure 4. Algorithm to get the radius of the sphere
After the bounding sphere created, the checking for
collision was done by calculating the distance between the
bounding sphere and the moving ball. Since the ball used to
collide with the cloth is a moving object, the position of the
ball will be changing at all times. Therefore, the radius of the
moving ball can be calculated using equation (3).

R
ball
= Pos
ball
+ ball radius (3)

Where R
ball
is the radius of ball, Pos
ball
is the position of the
ball and ball radius is the radius of the ball which has been
preset as 1. Then the distance between the cloth and the
moving ball is being calculated using equation (4).

1. Check the longest length of the cloth. (either width or
height)

2. Radius of the first sphere = (longest length + 1) 2.

3. Radius for the next spheres = radius of previous
sphere 2

Distance = R
spbere
+ R
ball
(4)

Where, R
spbere
is the radius of the bounding sphere which
will be updated when the ball is moving and R
ball
is the radius
of the moving ball. The next step will be the checks for
collision. The checking and updates for the bounding sphere
were done. Collisions were detected when the Distance is
smaller than the radius of the moving ball.
III. RESULT
The construction of the collision detection for cloth was
shown in Figure 5 and Figure 6. Several tests were conducted
and the analysis was discussed. The tests conducted are times
execution test, frame test, OpenGL functions call test, and
CPUs Average Utilization test.

Figure 5. The bounding sphere hierarchy.




Figure 6. The Axis Aligned Bounding Box (AABB)
The times execution test was the calculation of time to
execute the system. This test was conducted 20 times in order
to get the average time of execution and the results were
shown in Table 1.
TABLE I. 20 SETS OF TIME EXECUTION TEST FOR COLLISION DETECTION
USING BOUNDING SPHERE HIERARCHY AND AABB HIERARCHY

Experiment
Time Execution (ms)
for Bounding Sphere
Hierarchy
Time Execution
(ms) for AABB
Hierarchy



1 215.877 1233.01

2 188.131 1259.05

3 220.172 1267.02

4 216.726 1203.98

5 236.049 1263.00

6 223.576 1256.01

7 215.311 1177.10

8 233.671 1211.98

9 253.518 1260.03

10 259.710 1242.00

11 249.192 1259.02

12 245.186 1240.99

13 253.523 1259.00

14 232.061 1241.01

15 229.389 1243.03

16 274.904 1249.02

17 237.714 1272.03

18 226.821 1262.04

19 235.760 1271.02

20 257.875 1224.01



Average 235.2583 1244.718




From Table I, the average time execution for AABB
hierarchy (1244.72 ms) is more than of using bounding sphere
hierarchy (235.26 ms) which indicates that computational cost
for AABB is higher than bounding sphere. Figure 7 shows the
comparison of time execution for bounding sphere hierarchy
and AABB.


Figure 7. Time execution(ms) for BSH and AABB.
The second experiment conducted was the frame test. The
dramatic fall of the frame rate for bounding sphere as shown
in Figure 8 is due to the huge number of OpenGL calls
functions. The average frame per second for bounding sphere
hierarchy during the simulation was 13.41 whilst AABB was
21.17 seconds which is roughly closer to the real time. This
shows that AABB hierarchy is formed faster than the
bounding sphere hierarchy. Figure 8 shows the comparison of
frame test for bounding sphere is and AABB.


Figure 8. Comparison of frame test for BSH and AABB.



The OpenGL (OGL) functions call test indicates the
number of functions called when executing the program for
both bounding sphere hierarchy and AABB hierarchy.
Referring to Figure 9, the graph of OGL calls per frame for
bounding sphere hierarchy shows that the highest number of
OGL calls can reach up to 27,000 calls which is a huge number
of OGL calls. This will highly affect the frame test which
proves that the frame test was affected by the OGL call as
indicated in Figure 9.

Figure 9. Comparison of OpenGL function cal per frame for bounding
sphere hierarchy and AABB hierarchy

CPUs Average Utilization test is to test on the
percentage of the memory usage for CPUs when both
bounding sphere hierarchy and AABB hierarchy programs
are being executed. Figure 10 shows the percentage of
CPUs average utilization for bounding sphere hierarchy is
more than the AABB hierarchy for about 15%. Since the
OGL calls function for bounding sphere hierarchy is higher
than AABB, the CPUs utilization for bounding sphere
hierarchy will also be higher than AABB. Hence, the
CPUs average utilization is directly proportional to OGL
calls


Figure 10. Graph of CPUs Average Utilization test for bounding sphere
hierarchy and AABB hierarchy
IV. CONCLUSION
A series of tests has been conducted on the prototype for
cloth simulation using bounding sphere hierarchy. Meanwhile,
the cloth simulation using AABB hierarchy acted as the control
experiments for a comparison with the bounding sphere
hierarchy.
After running the tests, the processes of building and
updating the bounding sphere hierarchy is affected by the
number of generated vertices. This will cause a huge number of
OGL functions call which will lower down the frame per
second generated and CPUs utilization. Mean while, the
number of OGL functions calls also had a little bit of effect on
the time generation in which the bounding sphere require more
time in detecting the collisions. However, the execution time
test showed that bounding sphere hierarchy required lesser time
to execute comparing with AABB hierarchy.
Besides using bounding sphere hierarchy to detect collision
on cloth, other bounding volume can be used for more
efficient and accurate collision detection, for example the
oriented bounding boxes and k-DOP. Other than collision
detection on the surface, experiments for the volumetric
collision detection can be done too.
REFERENCES
[1] Hamzah Asyrani Sulaiman & Abdullah Bade. 2011. The Construction
of Balanced Bounding-Volume Hierarchies using Spatial Object
Median Splitting Method for Collision Detection. International Journal
on New Computer Architecture and Their Application.
[2] Gibson, S.F., & Mirtich, B. 1997. A Survey of Deformable Modeling in
Computer Graphic. MERL .
[3] Rajiv, P. 2011. Cloth Simulation using mass-spring technique. NCCA
CGIT.
[4] Jimnez, P., Thomas, F. & Torras, C. 2001. 3D Collision Detection: A
Survey. Computers & Graphics 25: 269-285.
[5] Teschner, M.E., Kimmerle, S., Heidelberger, B., Zachmann, G.
Raghupathi, L., Fuhrmann, A., Cani, M.-P., Faure, F., Magnenat-
Thalmann, N., Strasser, W. & Volino, P. 2005. Collision Detection for
Deformable Objects.
[6] Andersen K., Bay C., 2006. A survey of algorithms for construction of
optimal Heterogeneous Bounding Volume Hierarchies. Saatavilla pdf-
muodossa http://image. diku. dk/projects/media/christian. bay. kasper.
andersen. B, 6
[7] T.Simnett, 2012. Real-Time Simulation and Visualisation of Cloth
using Edge-based Adaptive Meshes, University of East Anglia.