INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4, pp. 577-587 APRIL 2013 / 577
© KSPE and Springer 2013
Development of a Surface-based Virtual DentalSculpting Simulator with Multimodal Feedback
Furqan Ullah1 and Kang Park2,#
1 Graduate School of Mechanical Engineering, Myongji University, San 38-2 Namdong, Cheoin-Gu, Yongin, Gyeonggi-Do, South Korea, 449-728 2 Dept. of Mechanical Engineering, Myongji University, San 38-2 Namdong, Cheoin-Gu, Yongin, Gyeonggi-Do, South Korea, 449-728
# Corresponding Author / E-mail: [email protected], TEL: +82-31-330-6344, FAX: +82-31-321-4959
KEYWORDS: Dental sculpting, Haptic rendering, Mesh subdivision and refinement, Multimodal realities, Virtual reality
This paper presents a surface-based virtual dental sculpting simulator based on sensory modalities like visual, auditory and
haptic sensation. The simulator can be used to perform different dental procedures such as grinding, drilling, or surface
scrubbing, and gain experience of using various virtual dental tools of different shapes. The surface-based dental model, which
is extracted from a commercial 3D dental laser scanner, is used for simulating sculpting processes at less memory cost. Large
amount of triangular mesh data is contained in scanned models; therefore, a model reduction algorithm is proposed for large
triangular mesh data. For the computation of repulsive force feedback, a spring-damper force model with a force filter is used.
Vertex deformation method is implemented along with an enhanced bi-tri subdivision method of triangles to perform precision
sculpting simulation. In order to make the mesh regular, a number of mesh refinement algorithms are performed. Finally,
considering the fidelity, stability, computer efficiency, and update rate of the haptic display, it can be concluded that these
multimodal realities based virtual system can generate stable simulation of material removal from a human tooth model with
realistic auditory, visual, and force sensations.
Manuscript received: October 4, 2012 / Accepted: December 23, 2012
1. Introduction
Medical simulation systems are increasingly being developed and
deployed to teach surgery planning and diagnostic procedures as well
as medical concepts and decision making to personnel in the health
professions. The important aim of the virtual dental simulator is to
provide the pre-operation planning that reduces errors and makes the
dentist feel safer, when entering to perform the real operation task on
the real patient. It also creates new dental training opportunities for
dental treatment procedures. It is believed that during dental surgery
procedures, a dentist must learn to operate dental instruments very
precisely and develop a realistic sense of touch for the interaction of
dental instrument with patient anatomy. Currently, in the dental area,
virtual reality simulators are used for research and development of
dental tools for new therapies, treatments, and early diagnoses.
Different methods were presented to sculpt the triangular models, and
these methods could be applied in medical surgery simulations.1-4
Many researchers have also used haptic display to enhance the
performance of virtual reality environments because simulating touch
in the virtual world can lead to the improvement in the performance of
training simulators. Therefore, many haptic force feedback simulators
based on volumetric and surface data have been proposed for virtual
simulation and visualization of dental training.5-27 Some existing
volumetric and surface based systems for dental simulation are
PerioSim (Haptics-based dental simulator for periodontics),10
Augmented reality haptics system for dental surgical skills training,11
VirDenT (virtual and augmented reality technologies in therapeutic
interventions simulation in fixed prosthodontics),12 HAP-DENT (A VR
Haptic Dental system),13 iDental (A haptics-based dental simulator),14
HapTEL (Haptic Technology Enhanced Learning for dental students).37
In addition, Furqan et al. proposed a surface-based virtual dental surgical
simulator based on axis-aligned dental tools.25 Kim et al. proposed a
haptic dental simulation system based on a hybrid representation of
geometric surface and volumetric representation containing implicit
surface and material properties.15,16 Kiminet al. contributed a new method
to incorporate the distance field of the tool into the dental drilling for
stable feedback force generation as well as accurate volume cutting.17
Wu, J. et al. proposed a volume-based tooth modeling and virtual cutting.
However, most of the systems are still in an exploratory stage, some
important functionalities are still missing in current virtual reality
DOI: 10.1007/s12541-013-0078-y
578 / APRIL 2013 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4
dental simulators.28
Furthermore, in most voxel-based dental training systems, we cannot
use any real tooth model and cannot add a new surgical operation.
Therefore, in this surface-based system, the simulation of material
removal can be carried out on any scanned surface model from a
commercial three-dimensional (3D) dental scanner. Furthermore,
compared with the volumetric-based approach, the surface-based
approach is more complex and difficult to handle but costs lesser
amount of computer memory. On account of the low memory cost, real-
time demands, computational efficiency, accurate visualization, and
smooth haptic feedback associated with surface-based simulators, this
paper focuses on a surface-based virtual dental sculpting simulation with
sensory modalities like visual, auditory and haptic sensation. In this
paper, we propose surface-based haptic dental simulation techniques to
address limitations in previous surface-based systems as follows:
- Triangular subdivision techniques in order to make regular mesh.
- Efficient collision detection with oriented bounding box (OBB)
- Force computation based on spring-damping approach for simulating
realistic tool interaction with the virtual tooth
- Smooth force rendering by using 2ndorder bi-quadratic digital filter.
- Twelve possible cases of bi-tri subdivision method for triangles to
efficiently handle abnormalities associated with the triangle problem
- Simulation of physical contact sound
Sound is an important feature of computer based virtual training
systems and should be included in a realistic simulation. It provides a
strong effect on the emotional perception of a scene. During different
dental operations, realism in simulation can be enhanced significantly
by adding sound effects. The further details of the proposed system are
described below.
For smooth force rendering and cutting simulations in the haptic
environment, the amount of model mesh data must be small. However,
the surface-based model of a human jaw has millions of triangles. Due
to the large amount of data, haptic and visual rendering have some
issues in the real-time updating of the model data. Therefore, in order
to realize stable force feedback, it is better to reduce the amount of data
by efficiently separating of one tooth from the human jaw model. In
this paper, a model reduction algorithm is proposed; by using this
algorithm the desired tooth tissues can be separated from a human jaw
model to achieve a higher visual update frequency of 30-60 Hz and a
stable haptic update rate of 1 kHz. It is also observed that the scanned
model may have irregular triangles that can cause the instability in the
system. Therefore, the mesh triangles must be regular. In order to make
them regular, a number of algorithms have been proposed based on
triangular subdivision techniques.31-33 In real dental treatments, tissues
will be removed according to the interaction between moving tools and
the tissues. This feature involves haptic sensation by computation of
force feedback. For the force feedback, the dental system must have the
ability to efficiently compute the repulsive force and realistic tool
interaction with the virtual tooth, which are all based on efficient
collision detection. In this system, efficient collision is realized using a
bounding box (Bbox) detection method i.e., axis aligned bounding box
(AABB) and OBB. The purpose of the Bbox is to minimize the
computation time required for fast rendering. In addition, many dental
tools with different shapes are used according to different dental
requirements for material removal processes.
In order to perform haptic renderings, a spring-based method is
commonly used for force computations.34-36 The calculation of the
repulsive force depends on the efficient interaction between the virtual
tool and the surface of the tooth. For this purpose, Hooke’s law is used
to calculate the force feedback from deformed mesh. In Hooke’s law
when force is applied to a spring, the spring will be compressed from
one position to another and the displacement during this compression
Fig. 1 Architecture of the virtual system
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4 APRIL 2013 / 579
can be computed. So, the spring force can be computed using this
method with different factors that are involved in the compression.
However, during realizing the repulsive spring force from the deformed
mesh, there are some issues that still need to be resolved such as
instability of the system, low fidelity, and unsmooth haptic sensation.
To handle these problems, a force filter is proposed.
In the surface-based model, the subdivision of triangles during the
sculpting process is an essential factor for simulating smooth material
removal operation. Many researchers introduce novel algorithms to
subdivide the surface and generate interior structures to simulate of the
virtual cutting operation. When the interaction between the virtual
cutter and the tooth surface occurs during the material removal
operation, the size of the mesh triangles must be smaller than the
cutter’s diameter. In order to fulfill this condition, a local-based triangle
subdivision algorithm should be applied to large triangles, and it is also
needed to eliminate the abnormalities of some triangles. In this paper,
we propose twelve possible cases of bi-tri subdivision method for
triangles to efficiently handle abnormalities associated with the triangle
problem. After the deformation of vertices (sculpting process), the
surface tooth model needs to be updated in the real time to enable
correct collision detection in the next computation cycle. In order to
reduce the update time needed for smooth haptic sensation, local
updating of the surface model is performed. Finally, in order to realize
the actual sound during material removal operation, the sound rendering
is performed. Fig. 1 dictates the system architecture of the virtual dental
sculpting system that consists of model reduction, mesh refinement,
collision detection, force computation, haptic and auditory rendering,
local-based subdivision of triangles, and cutting simulation processes.
The rest of this paper is structured as follows. Section 2 describes
the 3D representation of the surface model and different virtual dental
cutters. Section 3 gives the system details i.e., model reduction, mesh
refinement, material removal process, force computation, and
subdivision of triangles, etc. Section 4 explains the experimentsand
results of the proposed system. Section 5 discusses some critical issues
of haptic rendering and Section 6 gives the conclusion.
2. Three-Dimensional Representation
Surface-based modeling costs less computer memory than
volumetric representations. However, it is more complex and difficult
to handle. In this paper, a realistic surface-based 3D human jaw model
is used based on standard geometric references and parameters. The
original jaw model data is extracted from a commercial 3D dental laser
scanner, which defines the boundary surface of the human jaw as
illustrated in Fig. 2(a)-(b). The scanned surface-based data is composed
of a large number of vertices and triangles to construct a realistic virtual
model with the optimum degree of accuracy. The number of 60001
vertices (119995 triangles) is used to visualize the virtual jaw model. In
addition, this simulator is not designed for one specific surface model
but rather the simulation of material removal can be carried out on any
scanned model from a commercial 3D dental scanner.
In the real world, dentists use different kinds of cutter shapes to
sculpture teeth according to several requirements. In this system,
various dental tools with different shapes are included as shown in Fig.
2(c). All dental tools are surface-based modeled with good visual
quality. Dentists can change the size and the shape of a cutter any time
during the real-time sculpting simulation. The ability to change the size
and shape of the virtual dental tool makes the simulator a functional
sculpting system.
3. System Details
3.1 Model Reduction
During the sculpting process of a tooth with haptic interference, a
higher update rate is required for smooth rendering of the haptic
display. However, it is difficult to render a large amount of data with
containing millions of triangles very smoothly because the model must
be updated before the next sculpting action. Due to this problem, an
algorithm is proposed in which the dentist can separate the desired area
of tissues from the original jaw model by creating two cutting planes.
For example, the dentist can separate one tooth from the human jaw by
creating two cutting planes, which define the boundary (i.e., depth and
width) of a separate tooth. The idea is that when two cutting planes
appear in 3D space during the model reduction process, those planes
are set at desired positions that are defined by the boundary of the
tooth, as shown in Fig. 3(a). In the first step, we check which vertices
Fig. 2 Model representation: (a) Surface-based 3D human jaw model
extracted from a commercial 3D dental laser scanner, (b) Wireframe
model of (a), (c) Virtual dental tools of different shapes
Fig. 3 Model reduction: (a) 3D human jaw model with two cutting
planes, (b) After applying the model reduction algorithm. The dark
portion shows the removed tooth area
Fig. 4 Mesh refinement: (a) Original tooth mesh, (b) After applying
two rounds of subdivision, (c) After applying two rounds of Loop
subdivision
3
580 / APRIL 2013 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4
of the 3D model are inside the two cutting planes by computing the
normal vectors (np1 and np2) of the planes. In the second step, that
vertices are stored into a new data structure and outside vertices are
discarded. With the help of new data structures, a separate tooth requiring
less computer memory is rendered. Fig. 3(b) demonstrates the separation
state of a tooth from the human jaw model with two cutting planes.
3.2 Mesh Refinement
Mostly, the scanned surface-based dental model contains irregular
triangles that may cause the instability in the system during real-time
material removal operation.29,30 In order to realize the smooth tactile
force, it is preferable that the model mesh must contain regular
triangles. A number of algorithms are implemented to make model
mesh regular i.e., , Loop, 3-Split, 4-Split, and 9-Split subdivisions.
The implementation of these algorithms on the virtual tooth model
would be based on the mesh type. However, from our experiments,
the subdivision is recommended for the most dental models
because it produces all triangles after one round of subdivision and
creates C2 (smoothing function in which first and second derivatives
are continuous) surface almost everywhere with the lesser number of
triangles than the Loop subdivision. Note that the mesh refinement
process is implemented before enabling the real-time haptic rendering.
Therefore, there is not any influence of the refinement process on the
real-time material removal rendering. Fig. 5 demonstrates the
implementation of subdivision algorithms on a tooth model that contains
2,888 triangles. After one round of Loop and subdivisions, the
number of triangles was 11,552 and 8,858, respectively. Fig. 4(a) shows
the original tooth mesh and Fig. 4(b) and 4(c) give the results of
and Loop subdivisions, respectively.
3.3 Collision Detection and Sculpting Process
A realistic dental treatment system with haptic interaction requires
natural and real-time interaction between the cutter and the surface of
the tooth. In order to consider the haptic computation frequency of
1 kHz, a Bbox method is used to accelerate the collision detection
between the virtual tool and the surface of the tooth. Bbox method is
adopted to minimize the vertices data computation during the real-time
collision. For the removal of tooth material, a vertex deformation
method is used because of its simplicity and reduced computation
costs. During the collision, the bounding box intersects the surface
before the virtual tool, and when the collision occurs between the tooth
surface and the Bbox, we check how many vertices penetrate the Bbox.
We only consider those penetrated vertices for the second check
between the tool and the penetrated vertex for the deformation
calculation. After that, a new look of the surface can be realized based
on the penetrated vertices. Here, only local triangle properties such as
area, circumcircle radius, and incircle radius of the triangle are
computed to check for abnormalities (too thin/too big: regular/irregular
behaviour) of the triangle. The intersection between the tooth surface
and the Bbox is illustrated in Fig. 5(a), Fig. 5(b) demonstrates the
collision between the spherical tool and the tooth surface, Fig. 5(c)
gives the triangle subdivision, and Fig. 5(d) depicts thedeformed
vertices at the tool’s surface.
3.4 Spherical Collision
In order to detect the collision between a spherical virtual tool and
a mesh vertex, a Bbox is required. In the spherical case, the simplest
and efficient approach is to compute the global space AABB.
Therefore, in order to compute the updating position of the penetrated
vertex by the spherical tool, we simply extend that penetrated vertex to
the sphere’s surface by using following equations.
(1)
(2)
where r is the radius of the sphere, is the distance from the
penetrated vertex to the center of the sphere, Pglobal (Pglobal_x, Pglobal_y,
Pglobal_z) is the position of the penetrated vertex in the global (model)
space, P’global (P’global_x, P’global_y, P’global_z) is the updated position of
the penetrated vertex, Tglobal (Tglobal_x, Tglobal_y, Tglobal_z) is the center of
the sphere.
3.5 Cylindrical Collision
Fig. 6(a) presents the collision between the cylindrical virtual tool
and a mesh vertex. In order to calculate surface coordinates of a
cylindrical virtual tool, the tool can be subdivided into two parts. The
first part is the lower half disk of the sphere and the second part is the
vertical cylinder. The first part is similar to the spherical collision
section. In order to check the collision between the cylindrical part and
the mesh vertex, an OBB is computed in local (cylindrical) coordinate
space. The OBB is a tight fitted dynamic bounding box containing the
cylindrical part of the virtual tool. We can simplify the collision
computation by converting the eight corners of the OBB into the global
3
3
3
3
P'global r dΔ dΔ⁄( ) Tglobal+= 0 dΔ r< <
dΔ Pglobal Tglobal–=
dΔ
Fig. 5 Collision detection: (a) Intersection between the virtual Bbox and the tooth surface, (b) Intersection between the cutting tool and the tooth
surface, (c) Occurrence of subdivision, (d) Updated position of penetrated vertices
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4 APRIL 2013 / 581
(model) coordinate space AABB. After that, the collision is checked by
computing the maximum and minimum of the AABB. In order to
update the position of the penetrated vertex to the surface of the
cylinder, the inverse transformation (rotation and translation) is applied
to transform the vertex position from global space to the local space.
For this purpose, following operation is used to transform the global
vertex into the local space.
(3)
where Plocal (Plocal_x, Plocal_y, Plocal_z) is the vertex position in the local
space and M-1 is the inverse of the 4 × 4 homogenous transformation
matrix. Thus, the surface coordinates P’local of the cylinder can be
calculated as follows:
(4)
(5)
(6)
where r is the radius of the cylinder, θ is the angle between the center
axis and the outer edge of the cylinder and Δd’ is the distant from the
penetrated vertex position to the center axis of the cylinder. Finally, the
vertex is updated at the surface of the cylinder, and then the
transformation is performed again from the local space to the global
space by using the following relation:
(7)
3.6 Cone Collision
A cone can also be subdivided into two parts: a vertical cylinder and
a cone. The collision between the cone and the mesh vertex can also
be checked by subsequently applying the same procedure of the
cylindrical collision. After performing the inverse transformation by
using Eq. 3, the surface coordinate P’global can be calculated as follows:
(8)
where r is the radius of the bottom circle of the cone, hc is the height
of the cone, and hp is the distance from the origin of the cone to a
penetrated vertex along the z-axis. After computing surface coordinates
of the cone, the transformation is needed to be performed again from
the local coordinate system to the global coordinate system by using
Eq. 7. The collision between the virtual cone tool and a mesh vertex is
demonstrated in Fig. 6(b).
3.7 Haptic Rendering
Generally, graphic applications in real-time virtual environment
have display update requirements of 30-60 frames per seconds. The
update rate of haptic renderings must be higher than 1 kHz in order to
maintain a stable force feedback system.25-30 In order to realize the
force feedback in haptic simulations, the spring-damping force model
is commonly used to transform the motion of the haptic tip to the
virtual force. Therefore, the spring-damping approach is adopted to
compute the force feedback. The general force vector equation can be
written as:
(9)
where Ks and Kd are the spring stiffness and damping coefficient
respectively, is the displacement of the spring, and is the velocity
of the virtual tool after collision. For smooth and stabilized force
feedback, different methods have been implemented in surface-based
dental simulation systems.25 All of those methods tried to develop a
stable haptic interface system with higher frequencies. In this paper, a
2nd order bi-quadratic digital filter is adopted to reduce vibration and
obtain the smooth force feedback signal. The details are described as
follows:
(10)
where is the last position and is the updated position of
the penetrated vertex. Thus, the resultant force is the summation of all
forces:
(11)
(12)
(13)
where is the average normal vector, Nj is the vertex normal vector,
and m is the total number of vertices penetrated by the virtual cutter.
The mapping of phantom motion to the average normal direction of the
tooth surface is implemented by using Eq. 12 in which is the value
of projected onto . However, even after computing the resultant
force, low fidelity and vibration of the haptic stylus are issues that still
need to be resolved. Therefore, to resolve these issues, 2nd order bi-
quadratic digital filter is proposed. The discrete transfer function for the
z-transform is defined by:
Plocal M1– Pglobal=
P'local_x
P'local_y
P'local_z
r cosθ
r sinθ
Plocal_z
=0 d'Δ r< <
π θ π≤ ≤
θ 2atan Plocal_y
Plocal_x
,( )=
d'Δ Plocal_x
( )2 Plocal_y
( )2+=
Pglobal M P'local=
P'local_x
P'local_y
P'local_z
r hc⁄( )hp cosθ
r hc⁄( )hp sinθ
Plocal_z
= 0 d'Δ r hc⁄( )hp< <
F Ks xΔ⋅ Kd v⋅–=
xΔ v
xΔ Plast Pupdated–=
Plast Pupdated
Fr Ks( )i
xiΔ⋅i 0=
m
∑ Kd( )i
vp⋅–=
vp v n⋅( )n=
n Nj
j 0=
m
∑⎝ ⎠⎜ ⎟⎛ ⎞
Nj
j 0=
m
∑⎝ ⎠⎜ ⎟⎛ ⎞
⁄=
n
vp
v n
Fig. 6 Collision detection: (a) Cylindrical collision, (b) Cone collision
582 / APRIL 2013 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4
(14)
Hence the discrete transfer function is implemented with the
corresponding difference equation:
(15)
where the filter consists of n cascaded bi-quad sections, g[n] is the
overall gain of the filter, bk is the feedforward filter coefficients, aj is
the feedback filter coefficients, Fin[n] is the input force signal, and
Fout[n] is the output force signal. The parameters used with the filter
equation can be easily adjusted to the dynamic properties of the
phantom device, without changes in the algorithm. Finally, the filtered
force signal is sent to the haptic device controller for haptic rendering
at 1 kHz. The complete sculpting process with force exerted by
deformed vertices is shown in Fig. 7 in which Fig. 7(a) illustrates
separation state between the cutter and the tooth surface, Fig. 7(b)
depicts the intersection between the cutter and the tooth surface, Fig.
7(c) demonstrates the deformation of vertices from the original tooth
surface to the surface of the cutter after an intersection, Fig. 7(d) shows
the exerted force by deformed vertices, and Fig. 7(e) presents the final
tooth shape after cutting.
3.8 Bi-Tri Subdivision
During the real-time sculpting, it is observed that triangle
abnormalities occur along with deformations of tooth vertices, which
affect the tooth shape and vibration in the haptic device. In order to
handle this issue, a reconstruction method of triangles is used within
the local cutting area, in which a bi-tri subdivision of triangles is
utilized to correct abnormalities of triangles. According to the tri-
subdivision method, a midpoint of two longer edges is selected for the
division of the triangle from one to three triangles. At the same time,
the two adjacent triangles also need to be reconstructed for the stability
of the local mesh by a bi-subdivision method in which one triangle is
converted into two triangles. However, in a surface model, both
vertices and edges are shared by several adjacent triangles. Therefore,
six triangles are used to describe the twelve possible subdivision cases,
as illustrated in Fig. 8. In the first case of Fig. 7, suppose that 1- 2- 3
are the vertices and S1, S2, S3 are the edges of an irregular triangle.
First tri-subdivision is performed and two vertices 8 and 9 are inserted
into the edges S1 and S3, respectively. After that both adjacent triangles
1-3-4 and 1-7-2 are reconstructed through bi-subdivision. The vertices
8 and 9 are inserted when S1greater than S3 and S3 should be greater
than S2. Similarly, all the other cases are checked for the subdivision
of an abnormal triangle. In order to check which triangle lost its
normality behaviour within the local deformation area after the
collision and deformation of vertices, some basic triangle information
such as the area, incircle radius, and circumcircle radius of the only
local triangle is computed. The abnormality of the local triangle is
checked using predefined criteria, which is the ratio of incircle and
circumcircle of the local triangle. Regular and irregular behaviours of
the triangle can be seen in Fig. 9, where Rc is the circumcircle radius,
Rin is the incircle radius, and S1, S2, S3 are the lengths of edges of the
local triangle. In addition, if the case S1 = S2 = S3 occurs and if the
triangle does not meet the predefined criteria then 3-split subdivision is
performed on that triangle. In 3-split subdivision, one triangle is
converted into three triangles and only one centroid vertex is inserted.
After reconstructing the mesh in real-time, the model needs to be
updated due to changes in the position of vertices and the normal
information, therefore, the local updating algorithm is utilized because
of its efficiency and simplicity.
H z( ) g z( )b0 b1z
1–b2z
2–+ +
1 a1z1–
a2z2–
+ +---------------------------------------⋅=
Fout n[ ] g n[ ] bkFin n k–[ ]k 0=
2
∑ ajFout n j–[ ]j 1=
2
∑–⋅=
Fig. 7 Sculting process with force feedback: (a) No collision between
the cutter and the tooth surface, (b) Intersection between the cutter and
the tooth surface, (c) Deformation of verties from the original tooth
surface to the cutter’s surface after and intersection, (d) Force exerted
by deformed vertices, (e) The final tooth shape after cuttingFig. 8 Bi-tri subdivision: Twelve possible cases of subdivision
Fig. 9 Bi-tri Subdivision criteria: (a) Regular triangle, (b) Irregular triangle
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4 APRIL 2013 / 583
3.9 Local Updating
After each sculpting stroke, the surface model must be updated in
real-time to enable correct collision detection in the next computation
cycle. In order to reduce the update time required for smooth haptic
sensation, local updating of the surface model is carried out because the
surface point data does not change very much during the sculpting
process. We only consider the deformed area for updating and
reconstructing.
3.10 Sound Rendering
In order to realize different sound effects during different virtual
dental operations, a simple approach is used in which various sound
effects were recorded from the real dental operations and stored in an
audio format. In order to play that audio file in the virtual environment,
an audio library is utilized. To realize the actual sound during material
removal operation, pitch, gain of the sound, and source and listener
positions are adjusted. The approximate result is achieved by using this
approach. However, the realistic physical contact sound can also be
improved in the future.
4. Experiments and Results
4.1 Hardware Platform
The adopted hardware setup of the proposed system is illustrated in
Fig. 10. The specifications include a 2.0 GHz Intel Core™2 CPU, 2
Giga Bytes of RAM, simulation rendering based on Nvidia GT220
GPU with 512 MB memory, a 22” monitor, two stereo speakers, and
windows based operating system Microsoft Windows Vista. The
Phantom Omni is used as a haptic device which offers three degree-of-
freedom output and six degree-of-freedom positional sensing
capabilities and was provided by SensAble Technologies.
4.2 System Architecture
The proposed virtual dental simulator has been implemented on a PC
using OpenGL graphics library to visualize 3-D models, OpenAL audio
library to realize drilling sound, and OpenHaptics toolkit to render the
force signal. The programming environment is Microsoft Visual Studio
C++ 2005. All predefined coefficients are selected according to the
maximum stiffness and force capability by Phantom OmniTM.
In this system, the simulation of material removal can be carried out
on any scanned surface model from a commercial 3-D dental scanner.
From our experiments, the computational calculation time was reduced
for real-time sculpting when the triangle subdivision algorithm was
used. All components, i.e. haptic rendering, visual rendering,
subdivision, collision detection, simulated sound, and model updating
were tested to ensure all functions work properly as per design
specifications. The testing procedure is summarized as follows:
1. Execute the system by importing tri-mesh file of a 3D human jaw
model in the virtual environment.
2. Perform the model reduction process and separate one tooth from
the model by creating two cutting planes.
3. Perform the mesh refinement process to make the model mesh
regular and initial communication between Phantom and software
was carried out.
4. Select the cutting tool with appropriate configurations. For the
verifying of frame rate, the dentist drives the cutting tool around the
tooth model freely along with the translation of the haptic stylus.
5. Perform the material removal operation by driving the haptic stylus
and increase the driving force until the material is removed from the
surface of the tooth. The depth of penetration of the cutter and the
force were then computed.
6. At the same time, the abnormality of triangles was checked by
finding neighbor triangles around the local triangle that need
enhancement and then the subdivision of the only local triangle was
performed.
7. After that, the model updating was carried out using local-based
updating method.
If there was something wrong in system stability or haptic
rendering, steps 3-7 were repeated.
4.3 Experiments and Results
First a model reduction algorithm was implemented by creating two
cutting planes on the jaw model. The original number of triangles was
5,992, the number of vertices was 3,103, and the percentage of
remaining data after data reduction was 4.99%. Next of mesh
refinement and real-time sculpting simulation using spherical and
cylindrical virtual tools were performed and the results are
demonstrated in Fig. 11. The three-dimensional material removal
results show that the visual and haptic update rates can be maintained
over standard criteria (30-60 Hz and 1 kHz). The sound of drilling in
material removal and in idle mode was also realized. Furthermore, two
modes were defined for the model manipulation. The first mode was
Fig. 10 Hardware setup of the system
Fig. 11 Material removed from teeth by implementing real-time drilling
and surface scrubbing using spherical and cylindrical virtual tools
584 / APRIL 2013 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4
the non-sculpted mode in which dentists can touch and feel the tissues
of the tooth, but cannot operate the sculpting process. The second was
a sculpted mode in which the dentist can cut, drill, or scrub the tooth.
In the implementation of three-dimensional sculpting, the results of
the virtual force feedback and the material removal operation were
evaluated in two experiments i.e., drilling and surface scrubbing. The
experiments can differ with respect to the surface conditions of the
static tooth model. For demonstrating the force signal, the magnitude of
the virtual force vector was recorded. Fig. 12 gives the interaction force
between the virtual cutter and the tooth surface in the real-time drilling
without force filtering. The force signal illustrates that the vibration of
the haptic stylus can be realized by the dentist’s hand and noise of
servo motors occur. In Fig. 13, the filtered smooth force signal can be
realized during the same drilling operation. However, the force fidelity
is decreased due to the force filtering; it can be optimized by adjusting
the value of the gain of the filter. Fig. 14 depicts the virtual force signal
of the real-time surface scrubbing before and after the force filtering.
The optimum coefficients that used for the force filtering were b0 = 1.0,
b1 = -0.1, b2 = -0.1, a1 = -0.2, a2 = -0.798, and g (gain) = 0.004. The
both successive drilling and surface scrubbing operations were
performed with force filter and are illustrated in Fig. 15. All
illustrations of the computed filtered force signal verify that the
realized force from deformed mesh is as expected for a surface-based
3D mesh model.
In order to test and verify the bi-tri subdivision algorithm, a
rectangle was used, as illustrated in Fig. 16(a). The collision between
the cutter and mesh vertices was essential for the material removal
process, due to this restriction Fig. 16(b) shows the tool interacting with
the rectangle at a corner vertex. The rectangle model was subdivided
from 12 to 212 triangles after sculpting operation. Fig. 17 depicts the
verified bi-tri subdivision algorithm during the real-time drilling of the
original tooth, and the new generated triangles were 404 after
subdivision. Furthermore, concerning the computation time of all
processes, the precision of the model reduction, mesh refinement, and
the exact collision between the virtual cutting tool and the surface of
the tooth was precisely investigated. Table 1 describes the computation
Fig. 12 Virtual force signal in the drilling operation without force filtering
Fig. 13 Virtual force signal in the drilling operation with force filtering
Fig. 14 Signal of the virtual force with and without force filtering
during the surface scrubbing operation
Fig. 15 Signal of the filtered force in drilling and surface scrubbing
operations
Fig. 16 Bi-tri subdivisions on a rectangle: (a) Beforeintersection, (b)
After intersection
Fig. 17 Bi-tri subdivisions on the original tooth mesh: (a) before real-
time drilling, (b) After real-time drilling
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4 APRIL 2013 / 585
time of the model reduction and the mesh refinement algorithm. It is
important to mention here that the mesh refinement process was
implemented before the real-time haptic rendering. Therefore, there is
not any influence of the refinement process on the real-time material
removal rendering; thus, increased the frame rate in the real-time haptic
rendering and reduced the haptic stylus vibration. The table data also
shows the comparison between loop and subdivisions, and it is
concluded that the subdivision creates less number of triangles but
takes more time than the loop subdivision. Because the mesh
refinement process occurs before the real-time haptic rendering and
time was not the issue at that stage, therefore, it is acceptable for
dentists and their candidates. In addition, the Table 2 gives the
computation time of real-time collision between the surface of the tooth
and the virtual tool that shows considering the more number of
triangles in the real-time haptic rendering results more computation
time and less frame rate. Therefore, it was required to reduce the model
data for fast haptic rendering. Finally, the results from the experiment
verify that this system can generate stable three-dimensional sculpting
simulation from a human tooth model with realistic auditory, visual,
and haptic sensations.
5. Discussion
Haptic devices allow dentists to perform different operations like
pulling, pushing, and cutting of soft or hard tissue with realistic force
feedback. During tooth cutting operation, too much applied force will
increase the rate of heat generation and thus damage the tooth tissues,
while too little force may prolong the painful treatment procedure for
the patient. Therefore, haptic sensation is crucial for the dentists to
operate successfully.
Trials were performed with 20 users performing 15 sessions. Each
user was given instructions for performing the drilling and surface
scrubbing operations. This practice was repeated once with 20 users.
From the results, it was also observed that the speed of the virtual
cutting tool should be realistic because fast phantom speed could create
instabilities in the force computation and model mesh. As seen in Fig.
14, the vibration of the haptic stylus occurs. This could be due to the
vibrations in the wrist and the hand instability. Therefore, we tried to
solve this issue by introducing a force filter and two modes that control
the tool’s speed during material removal simulation: the low speed
mode is activated when the tool moves fast and otherwise the normal
speed mode is active.
Since, there is no internal part of the surface-based jaw model;
therefore, the system is restricted to perform sculpting simulation on
the surface. We only consider the material’s surface properties for
sensing the force feedback. In order to consider deep drilling operation
or if the user continues to drill, the system may become unstable due
to overlapping of triangles. To solve this problem, bi-tri subdivision
with predefined criteria for abnormal triangles can have reasonable
sculpting and visual results. However, future work should address this
problem for stabilizing the system.
Fidelity is defined as the force feeling during virtual cutting
operation is similar to that in real dental operation, and stability refers
to the stable running of the haptic device with high update rates. It is
observed that there is always a large difference for the fidelity and
stability of haptics simulation compared with real dental operation and
that unstable haptic sensation in dental training can mislead the dentists
in real operation. Therefore, fidelity and stability of haptics simulation
are very important for dental sculpting operation. In this system, in
order to achieve high update rate, model reduction, local based data
updating, and simple force model was adopted. To achieve suitable
fidelity, the digital filter was proposed with the gain factor. User can
manually adjust the gain of the filter for stable and smooth force
feedback.
6. Conclusion
In this paper, a surface-based virtual dental sculpting system has
presented using mixed realities in which dentists can perform different
dental procedures with smooth tactile feeling, auditory and visual
realization. Different dental treatment procedures in this proposed
simulator could help dentists to perform safer operations before
entering to the real operation task. A model reduction algorithm is
proposed in which dentists can separate the desired area of the tooth
from the human jaw model for fast haptic rendering. Different mesh
refinement techniques are utilized to make the model mesh regular. A
fast vertex deformation method is used to simulate the removal of tooth
tissues. To prevention abnormalities of triangles during deformation of
vertices, detailed twelve possible cases of the bi-tri subdivision are
proposed. Enhanced model updating is used to achieve the display
update rate 30-60 Hz using a local-based updating method.
Experiments are carried out using a Phantom OmniTM haptic device;
3
3
Table 1 Computation time of model reduction and mesh refinement
algorithms before real-time haptic rendering
Model
TrianglesExp.
Triangles
after model
reduction
Model
reduction
time (sec)
Triangles
after mesh
refinement
(one round)
Mesh
refinement
time (sec)
119995
11645
half tooth0.075
6748 – L
5209 – S
0.01 – L
0.08 – S
24419
one tooth0.090
17676 – L
13451 – S
0.11 – L
0.16 – S
39698
two teeth0.110
38792 – L
29377 – S
0.92 – L
1.10 – S
L = Loop subdivision, S = 3 subdivision
Table 2 Computation time of real-time collision after 3 subdivision
Exp. Triangles
Collision detection
+ force computation
time (sec)
Results
1 5209
0.009 - sphere
0.010 - cylinder
0.012 - cone
i) Higher frame rate
ii) Smooth haptic rendering
2 13451
0.011 - sphere
0.019 - cylinder
0.022 - cone
i) Higher frame rate
ii) Smooth haptic rendering
3 29377
0.018 - sphere
0.021 - cylinder
0.031 - cone
i) Less frame rate
ii) Haptic vibration occurs
586 / APRIL 2013 INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4
these experiments involve the three-dimensional cutting of a tooth
model using a 2nd order bi-quadratic force filter. The experiment
operations verify that the force stability can be easily maintained under
the specified operation criteria.
In the future, we plan to enhance the simulator by improving the
stability, fidelity, realistic drilling sound further. We also have plan to
add augmented reality (AR), cleaning, bleeding, evaluation of student
performance, and recoding the training procedure performed by dental
students.
ACKNOWLEDGEMENT
This work was supported in part by the Korea Ministry of
Knowledge Economy, under Grant of the Strategic Technology
Development Project on Biomedical Supplier (Development of the
Digital Fusion Artificial Tooth Treatment Supporting System).
REFERENCES
1. Bruyns, C. and Senger, S., “Interactive Cutting of 3-D Surface
Meshes,” Computer and Graphics, Vol. 25, No. 4, pp. 635-642,
2001.
2. Zhang, H. and Payandeh, D. S. J., “Simulation of Progressive
Cutting on Surface Mesh model,” DRAFT 6-08, pp. 1-8, 2002.
3. Brown, J., Sorkin, S., Latombe, J., and Montgomery, K.,
“Algorithmic Tools for Real-Time Microsurgery Simulation,”
Medical Image Analysis, Vol. 6, No. 3, pp. 289-300, 2002.
4. Choi, K. S., “Interactive cutting of deformable objects using force
propagation approach and digital design analogy,” Computer and
Graphics, Vol. 30, pp. 233-243, 2006.
5. Dachille, F., Qin, H., and Kaufman, A. E., “A novel haptic-based
interface and sculpting system for physical based geometric design,”
Compter-Aided Design, Vol. 33, No. 5, pp. 403-420, 2001.
6. Chang, Y. H., Chen, Y. T., Chang, C. W., and Lin, C. L.,
“Development scheme of haptic-based system for interactive
deformable simulation,” Computer-Aided Design, Vol. 42, pp. 414-
424, 2010.
7. Basdogan, C., De, S., Kim, J., Muniyandi, M., Kim, H., and
Srinivasan, M. A., “Haptics in minimally invasive surgical
simulation and training,” IEEE Computer Graphics and
Applications, Vol. 24, No. 2, pp. 56-64, 2004.
8. Rodrigues, M. A. F., Silva, W. B., Neto, M. E. B., Gillies, D. F., and
Ribeiro, I. M. M. P., “An interactive simulation system for training
and treatment planning in orthodontics,” Computer and Graphics,
Vol. 31, pp. 688-697, 2007.
9. Thomas, G., Johnson, L., and Stanford, D., “The Design and Testing
of a Force Feedback Dental Simulator,” Computer Methods and
Programs in Biomedicine, Vol. 11, No. 64, pp. 53-64, 2001.
10. Luciano, C., Banerjee, P., and DeFanti, T., “Haptics-based virtual
reality periodontal training simulator,” Virtual Reality, Vol. 13, No.
2, pp. 69-85, 2009.
11. Rhienmora, P., Gajananan, K., Haddawy, P., Dailey, M. N., and
Suebnukarn, S., “Augmented Reality Haptics System for Dental
Surgical Skills Training,” VRST 2010, pp. 97-98, 2010.
12. Noborio, H., Sasaki, D., Kawamoto, Y., Tatsumi, T., and Sohmura,
T., “Mixed reality software for dental simulation system,” IEEE
International Workshop on Haptic Audio Visual Environments and
their applications, pp. 19-24, 2008.
13. Bogdan, C. M., “Domain ontology of the VirDenT system,” J.
Comput. Commun. Control, Vol. 6, pp. 45-52, 2011.
14. Wang, D., Zhang, Y., Hou, J., Wang, Y., Lv, P., Chen, Y., and Zhao,
H., “iDental: a haptic-based dental simulator and its preliminary user
evaluation,” IEEE Trans. Haptics, Vol. 5, No. 4, pp. 332-343, 2011.
15. Laehyun, K., Gaurav, S. S., and Mathieu, D., “A Haptic Rendering
Technique Based on Hybrid Surface Representation,” IEEE
Computer Graphics and Applications, Vol. 24, No. 2, pp. 66-75,
2004.
16. Kim, L. and Park, S., “An efcient teeth modeling for dental training
system,” Int. J. CAD/CAM, Vol. 8, No. 1, pp. 41-44, 2008.
17. Kimin, K. and Park, J., “Virtual bone drilling for dental implant
surgery training,” VRST 2009, 2009.
18. Novint Technologies, “Virtual reality dental training system (vrdts),”
http://www.novint.com/index.php/apg/medicaldental, [Accessed:
Sept. 2012]
19. Marras, I., Papaleontiou, L., Nikolaidis, N., Lyroudia, K., and Pitas,
I., “Virtual Dental Patient: a System for Virtual Teeth Drilling,”
IEEE International Conference on Multimedia and Expo., 2006.
20. Noborio, H., Sasaki, D., Kawamoto, Y., Tatsumi, T., and Sohmura,
T., “Construction of Dental Simulation System with Mixed Visual,
Tactile, and Sound Realities,” 18th International Conference on
Artificial Reality and Telexistence, 2008.
21. Chang, H. C., Jin, Y. L., Yong, K. L., and Mun, T. C., “Determining
the Passive Region of the Multirate Wave Transform on the Practical
Implementation,” Int. J. Precis. Eng. Manuf., Vol. 12, No. 6, pp.
975-981, 2011.
22. Park, J., Kim, K., and Hong, D., “Haptic-based resistance training
machine and its application to biceps exercises,” Int. J. Precis. Eng.
Manuf., Vol. 12, No. 1, pp. 21-30, 2011.
23. Ullah, F. and Park, K., “Virtual Dental Sculpting Simulation using a
Surface-based Tooth Model and a Haptic Device,” Korean Society
of CAD/CAM Engineers, pp. 838-851, 2011.
24. Ullah, F. and Park, K., “Visual, Haptic, and Auditory Realities based
Dental Training Simulator,” 2012 International Conference on
Information Science and Applications, pp. 106-111, 2012.
25. Ullah, F. and Park, K., “Surface-Based Virtual Dental Surgical
INTERNATIONAL JOURNAL OF PRECISION ENGINEERING AND MANUFACTURING Vol. 14, No. 4 APRIL 2013 / 587
Simulator using Haptic Display,” Computer-Aided Design &
Applications, Vol. 8, No. 6, pp. 841-848, 2011.
26. Ullah, F. and Park, K., “Surface-based Virtual Dental Sculpting
Simulation with Multimodal Feedback,” Asian Conference on
Design and Digital Engineering, pp. 548-555, 2011.
27. Ullah, F. and Park, K., “Virtual Dental Treatment Training System
using a Haptic Device,” Korean Society of CAD/CAM Engineers,
pp. 115-121, 2009.
28. Xia, P., Lopes, A., and Restivo, M., “Virtual reality and haptics for
dental surgery: a personal review,” The Visual Computer, pp. 1-15,
2012.
29. Ullah, F., Lee, G. S., and Park, K., “Piezoelectric Transducer based
3D Intraoral Scanner,” 2012 International Conference on Information
Science and Applications, pp. 118-123, 2012.
30. Ullah, F., Lee, G. S., and Park, K., “Development of a Real-time 3D
Intraoral Scanner based on Fringe-Projection Technique,”
Transactions of the Society of CAD/CAM Engineers, Vol. 17, No. 3,
pp. 156-163, 2012.
31. Kobbelt, L., “3 Subdivision,” Proc. of SIGGRAPH, pp. 103-112,
2000.
32. Loop, C. T., “Smooth subdivision surfaces based on triangles,”
Master’s Thesis, Department of Mathematics, University of Utah,
1987.
33. Bischoff, S. and Kobbelt, L., “Teaching meshes, subdivision and
multiresolution techniques,” Computer-Aided Design, Vol. 36, No.
14, pp. 1483-1500, 2004.
34. Foskey, M., Otaduy, M. A., and Lin, M. C., “ArtNova: Touch-
enabled 3D model design,” IEEE Virtual Reality Proceedings, pp.
119-126, 2002.
35. McDonnell, K. T., Qin, H., and Wlodarczyk, R. A., “Virtual clay: a
real-time sculpting system with haptic toolkits,” ACM Symposium
on Interactive 3D Techniques, pp. 179-190, 2001.
36. Tanaka, A., Hirota, K., and Kaneko, T., “Virtual cutting with force
feedback,” Proc. of Virtual Reality Annual International
Symposium, Vol. 199, No. 8, pp. 71-77, 1998.
37. Tse, B., Harwin, W., Barrow, A., Quinn, B., San Diego, J. P., and
Cox, M., “Design and development of a haptic dental training
system - hapTEL,” Euro Haptics Conference, pp. 101-108, 2010.