Virtual Try-On

Document Sample
Virtual Try-On Powered By Docstoc
					                                            Virtual Try-On
      Topics in Realistic, Individualized Dressing in Virtual Reality
         A. Divivier, Dr. R. Trieb1, A. Ebert, Prof. Dr. H. Hagen2, C. Gross, A. Fuhrmann,
            Dr. V. Luckas, Prof. Dr.-Ing. J.L. Encarnação3, E. Kirchdörfer, M. Rupp, S.
         Vieth4, S. Kimmerle, M. Keckeisen, Dr. M. Wacker, Prof. Dr. W. Strasser5, Mirko
                              Sattler, Ralf Sarlette, Prof. Dr. R. Klein6


         In the course of the project Virtual Try-On new VR technologies have been developed, which
         form the basis for a realistic, three dimensional, (real-time) simulation and visualization of
         individualized garments put on by virtual counterparts of real customers. To provide this cloning
         and dressing of people in VR, a complete process chain is being build up starting with the touch-
         less 3-dimensional scanning of the human body up to a photo-realistic 3-dimensional presentation
         of the virtual customer dressed in the chosen pieces of clothing. The emerging platform for
         interactive selection and configuration of virtual garments, the „virtual shop“, will be accessible in
         real fashion boutiques as well as over the internet, thereby supplementing the conventional
         distribution channels.

      1.      Introduction
Nowadays, consumers with their increasing desire for individuality make high demands
within the services sector. Especially, people want to get a good value easily almost at any
time and at any place while claiming a wide range of goods, a high degree of individuality as
well as quality and service at the highest level. Handling such immense demands will only be
possible, if new fundamental technologies for the presentation, selection and “try out” of
products will be developed in order to supplement the classical selling process.
In particular, in the garment field, the virtualization of familiar paradigms (including a virtual
try-on) leads to the creation of virtual shop environments (at the point-of-sales or in the
internet) which, for the first time, allow offering a wide range of individualized clothing while
additionally enhancing shopping experience and customer support.
Nevertheless, previous approaches within this area have not been very successful. But why ?
One of the fundamental reasons for the missing success and acceptance of such systems is
caused by the lack of identification of the customer with his / her virtual counterpart.

    Human Solutions GmbH, Europaallee 10, D-67657 Kaiserslautern, Germany
    IVS, DFKI GmbH, Erwin-Schrödinger-Str. 57, D-67663 Kaiserslautern, Germany
    FHG/IGD, Fraunhoferstraße 5, D-64283 Darmstadt, Germany
    Hohenstein Institute for Clothing Physiology, Schloss Hohenstein, D-74357 Bönnigheim, Germany
    WSI/GRIS, University of Tübingen, Germany, Sand 14, D-72076 Tübingen, Germany
    Institute of Computer Science II, University of Bonn, Römerstr. 164, D-53117 Bonn, Germany
Currently, most implementations either incorporate pure, two dimensional silhouettes or
oversimplified 3d computer-based mannequins, so that the customer will hardly be able to
recognize himself / herself. Furthermore, at present, the typical visualization and simulation
of garments is not giving any meaningful feedback of the “look and feel” of the selected cloth.
Particularly, garments are rendered independent of the customer’s size, i.e. they always seem
to fit. Therefore, no real decision support is given to the customer. Questions like “How does
the garment really look like?”, “Does it look good, when I’m wearing the cloth ?“, can not be
answered as well as concerns related to fit and sizing can not be resolved.
In particular, in the context of online shopping, after receiving the ordered garments
customers often are disappointed or unsatisfied. This, in turn, leads to high product return
rates as well as future indecision to purchase garments over the internet.
Therefore, the goals defined within the Virtual Try-On project aim at an optimal support of
the customer in decision making and thus to minimize time and costs for manufacturers and

   2.     Virtual Try-On
In the course of the project Virtual Try-On innovative VR technologies have been developed,
which form the basis for a realistic, three-dimensional, (real-time) simulation and
visualization of individual customers and garments. Utilizing these VR techniques an
integrated virtual shop infrastructure is provided, which facilitates the presentation and trade
of individualized garments at the point-of-sales and soon over the internet. Instead of
replacing the current shopping experience (e.g. really touching garments and materials),
Virtual Try-On, in fact, aims at enhancing customer support and decision making through
extending corresponding customer services.
The following scenario describes a typical way a customer will experience the VTO shopping
environment at the point of sales.

Virtual Try-On scenario at the point-of-sales
A customer decides to buy new garments in a fashion boutique, supplying the Virtual Try-On
cloning and dressing service. If not done already, his / her body surface will have to be
scanned (by 3d laser scanner) in order to create the customer’s digital twin. Henceforth, this
virtual avatar will be representating the customer within the virtual try-on application, also
serving as the basis for capturing necessary information (i.e. body dimensions and feature
points) with respect to the production and / or simulation of garment in relation to the virtual
human body. Therefore, additional scans will only be necessary in case the customers body
shape has changed noticeable.
By browsing through an interactive, virtual catalog the customer is able to shortlist interesting
pieces of clothing. To support this preselection the desired garments are presented in a (fast)
preview mode, already showing his / her digital twin wearing the clothes in the chosen colors
and materials. Although, thereby, the customer gets a first impression of how the garments
would look like when he / she is wearing it, this preview does not provide enough information
regarding sizing and fitting.
Therefore, in case the customer wants to have a detailed look at a favoured combination of
clothes, a virtual construction process is invoked, which based on the cloth data (model type,
color, material, configuration) and the given body dimensions will create individual, adapted
three-dimensional models of the desired garments. Further processing of the results will dress
the avatar by utilizing extended, physically-based simulation techniques. For presentation
purposes a new output device, the virtual mirror, is used, which provides a life-size display of
the customers virtual counterpart dressed in the chosen pieces of clothing from arbitrary
viewing angles.
Through all this, modifications with respect to size, configuration and color can be tried out
virtually within a short amount of time. After finally deciding to purchase the selected clothes
an appropriate order will be created automatically and sent to the corresponding manufacturer.
The Virtual Try-On scenario regarding the internet will be implemented similar to the
scenario described above. As a matter of course, the customer will have to be scanned in one
of the "Virtual Try-On fashion boutiques”, before being able to access the Virtual Try-On
services at home. Instead of utilizing a “virtual mirror” output device, simulation results will
be displayed on the customer’s local computer screen.
The following sections give a detailed description of essential topics and implementations
required to build up the process chain for handling the Virtual Try-On scenarios.

Creating the virtual customer (Human Solutions GmbH)
The starting point of the cloning and dressing process within Virtual Try-On is the creation of
a three-dimensional virtual counterpart of the customer, the so-called customer avatar.
For this purpose, a 3D laser scanner catches touch-less the customers body surface within a
few seconds and produces a three-dimensional point cloud consisting of round about 450.000
to 600.000 points. After post - processing (purifying, smoothing, ...), the point cloud is
transferred into a smooth, closed polygonal surface. This is achieved by separating the raw
scan into approximately cylindrical parts, reconstructing each part into a B-spline surface
(NURBS) individually and merging the results into a single coherent mesh again. To further
ensure identification with his / her virtual twin, besides the shape of the human body, color
images of the body surface are captured during the scanning process. Mapping this texture
data onto the triangle mesh created so far leads to the final version of the static customer
avatar (see figure 1, middle).

 Figure 1: Different stages creating an individual customer avatar : Scanning the customer’s
             body surface (left). Static avatar (middle). Dynamic avatar (right).
Based on the exact digital representation of the customer, all measurements and characteristic
feature points (e.g. ellbow, shoulder, wrist, …), which are necessary for actually producing
and simulating garment in relation to the virtual human body, are captured automatically.
Such characteristic feature points are also taken into account while computing an inner model
and a corresponding segmentation of the up to now static customer mesh. By applying skin /
mesh deformation methods – based on the combination of vertex-blending and bone-blending
– the avatar is enabled to execute simple, but typical movements (Walk, Turn around, ..) of a
real customer standing in front of a mirror, controlling fit, look and feel (see figure 1, right).
Hereby simulation and presentation of virtual garment worn by the customer avatar can be
improved leading to a maximum realistic impression.

Interactive individual clothing catalog (DFKI)
After the virtual customers has been created, he / she can select and combine different
garments as well as various colors and patterns in an individual 3D catalog. Instead of
applying a time-consuming physically-based cloth simulation, clothing models have been
generated before by draping real clothing over a dressmaker’s dummy and catching the model
geometry through a 3D laser scanning process. This leads to a model of the cloth with
absolute realistic wrinkles. In our approach, the desired garments must be scanned in only one
basic size – all other sizes will be calculated in the morphing process.

Rule-based Morphing
In contrast to existing morphing techniques (e.g., [LV94]) here an absolute control over the
intermediate shapes is required. I.e., it must be assured that the intermediate shape has the
exact associated measures of the needed cloth-size and not something that just looks similar.
Cloth sizes are usually defined by individual measures like collar size, sleeve length, back
width or sleeve circumference. Therefore we have derived a set of rules which describe the
individual changes that have to be made by our morphing agent when transforming one size
into another. This process is much more complicated than just zooming in and out, because
the changes of the single measures are not uniform. As mentioned above only one piece of
real garment is needed for the scan process. Therefore, the developed algorithms apply
deformation techniques to the shape in order to produce a new size with corresponding
measurements. So our approach [EGH03] deforms a shape but morphs between the sizes. In
contrast to the well-known interpolation between two shapes we do not produce additional
wrinkles for intermediate shapes. Therefore, the computed new size of the garment looks
much more realistic than one which would be produced with ordinary morphing techniques.
Furthermore, our approach is very flexible and extensible: by adding additional scans of the
same garment in other sizes, during the import process the representation which has the most
minimal distance regarding the individual measurements can be chosen. Hereby the accuracy
is improved going along with higher but tolerable storage requirements.
After computing the needed size of a garment, the virtual dressing of the figurine is done. In
the first step, the garments are positioned around the body – a jacket, for example, is fixed at
the shoulder area. Due to the fact that people will always differ a little in their posture during
the scan process, the algorithm has to correct the position of the arms in a second step. From
observing real life configurations it is clear that there are body parts which are totally hidden
by the garment. Consequently, these parts can be blended out by just making them invisible
during the rendering process. Only the segments that are partly visible (like the lower arm)
must be handled during the following collision detection. This process is shown in figure 2.
                          Figure 2: Virtual Dressing in the clothing catalog.

Retexturing using cooperative patterns
Our concept [ESD03] is based on the generation and analysis of a colour code (comparable to
a 2D barcode) that is printed on the fabric before the cloth is tailored. Here, the colour code
defines a discrete coordinate system, which also represents the fabric direction. The wrinkles
– or better: the corresponding hidden areas – can be located by detecting missing portions of
code in the texture images produced during the scanning process. We are calling this colour-
coded pattern a cooperative pattern.
As a coding unit we’ve chosen a square, because a right-angled coordinate system is very
suitable for parameterisation. A character of the coding alphabet is formed from one square
and its eight neighbours. Thus, each character has a size of 3×3 squares. We are making use
of a hierarchical two-level pattern in an applicable size for the wrinkle recognition, which is
composed of a lower and an upper level. The lower level consists of a coded sample as
described above. Three colours (red, green, and blue) are used for the coding and a fourth
colour (white) is used for the separation of the coded parts. The composition of elements of
the lower level forms the upper level. With the two-level pattern a rapport is directly given, so
a fabric of arbitrary size can be manufactured.

       pattern of cloth                     matrix representation
                                character      101020101202101
                                 1 0 2         210201010121210
                                 0 2 0         210212021201210
                                 2 0 2         101020212012102

                                 0 2 1
                                 2 0 2
                                 0 2 0

    Figure 3: Composition of the coded                              Figure 4: Individual 3D clothing catalog
                 pattern.                                                         visualization.
In case the customer wants to have a detailed look at a favoured combination of clothes, a
virtual construction and simulation process is invoked, consisting of three basic steps : pre-
positioning, pysically-based simulation and high quality visualization.

Geometric Pre-Positioning (FHG/IGD)

Virtual Clothing, that comes from CAD Systems used in the apparel industry, is represented
by its two-dimensional cloth patterns. There is also information available, how these patterns
must be stitched together. Our dressing method consists of two steps: geometric pre-
positioning and physically-based cloth simulation. In the pre-positioning step the cloth
patterns are positioned automatically around the body segments. These pattern positions serve
as initial values for the cloth simulation, where the patterns are sewed together directly onto
the human figure and where the final fitting is computed. The main idea of our pre-
positioning algorithm (see also [FGLW03] and [GFL03]) is to use developable bounding
surfaces for the human body segments, onto which the cloth patterns are positioned. We
assume, that the virtual human body is segmented into several body segments. Additionally,
we need the positions of some feature points, which mark special positions of the human

Figure 5: Bounding cylinders for arms and legs (left). Cloth patterns of a male shirt positioned
                       according to their sewing information (right).

Bounding Surfaces
Figure 5 (left) shows the minimal bounding surfaces for the arms and the legs. We established
a mapping between the minimal bounding surfaces and their flattened counterparts. That
means, cylinders are mapped into 2D rectangles and cones into 2D circle parts. This mapping
allows us to position the cloth patterns in two-dimensional space and to map them back into
their final 3D positions around the human body.

Arranging The Patterns
Cloth patterns, which belong to the same body segment, are positioned by using their sewing
information: If two cloth patterns must be stitched together, then they are positioned side by
side (see Figure 5, right). Then we use the feature points of the human body to move the cloth
patterns onto the flattened bounding surface, before we map them into 3D space around the
corresponding body segments.
Pre-positioning of Many Pieces of Clothing
When dressing people with many pieces of clothing, a dressing order is required. This
dressing order determines for example, whether you like to wear you shirt inside the trousers
or outside. It also identifies the sequence under which the cloth patterns are processed in our
pre-positioning algorithm. After a cloth pattern has been pre-positioned the corresponding
bounding surface is broadened, so that a cloth pattern, which is processed afterwards on the
same body segment, encloses the first pattern. We obtain a sequence of bounding surfaces
lying one upon the other for every body segment, onto which the several pieces of clothing
can be positioned. (See Figure 6).

              Figure 6: Simultaneous pre-positioning of trousers, shirt and jacket

Interactive Java-based garment simulation (FHG/IGD)
For the second step of our dressing method we developed a system for interactive animation
of cloth. It is implemented in Java and thus can be easily integrated into the envisioned
internet scenario. The triangulated cloth patterns serve as a basis for a mass spring system.
Since cloth is a very rigid material when stretched, extremely large forces occur in such a
system. Several methods have been described in the recent years to solve the underlying
differential equations efficiently [BW98], [CK02], [HE01]. We have developed an algorithm
which replaces the internal cloth forces by several constraints and therefore can easily take
large time steps without much computational overhead [FGL03].
The simulator also has to handle self collisions and collisions between the human body and
the cloth. In order to solve these problems efficiently we are testing only particles against the
surface of the body and each other. Distances between particles and the human body are
rapidly computed with a signed distance field [FSG03].
  Figure 7: Simulation of shirt and trouser (left). Interactive changing of sleeve length. The
               length can be changed with a slider at interactive rates (right).

Technical requirements of textile and clothing (Hohenstein)
One of the main focuses of this research is defining the influencing variables of the relevant
material parameters on the clothing simulation or virtualisation. For this, the individual textile
material is not merely regarded in isolation; for the first time, the influences arising from the
combinations of materials and the different methods of processing used are specifically
analysed and described. At the visualisation stage it is important to be able to accurately
represent clothing made of the same pattern but which has been processed in different ways.
Both individual materials and combinations of materials for the outer fabric, lining and
interlining, as well as materials which have been processed differently, are therefore also
analysed. To clearly define the material properties, special measuring systems are employed
which are used in the textile and clothing sector to assess the processability of materials and
to provide a (comparative) assessment of the handle of textile products. On the basis of the
following individual tests the calculation for the material simulation are made: bending
(flexing resistance), shearing (deformation of the warp and weft threads from the standard 90
degree angle) and the tensile stress-strain value for the material in question.
On the basis of the results of these tests, methods are derived which enable the relevant
material parameters to be determined more simply. This creates the basis on which to consider
new textile materials and possible ways of processing these for optimal visualisation with
minimal additional effort and to make these new materials available for the virtualisation of

Figure 8: Defining the relevant material parameters (left) and simplified method to group new
                                       materials (right)

Another important aspect of the work is defining the influencing variables which result from
the correlation between body geometry and pattern section geometry. 2D pattern sections
currently available, which show different types of stitching and different areas of material are
generated in the form of DXF files, edited and input in a product database. In addition to the
2D CAD files, information on the sewing of the individual pattern sections, the areas of
material and related material parameters, the body reference points, types of stitching,
variations in drape and types of fastening are supplied. This additional information is
integrated in XML files, which form the basis for the automatic positioning and sewing of the
pattern sections on the virtual bodies. It should be borne in mind here that clothing is not
intended to replicate the figure exactly, but to flatter the body shape. In order to make it
possible to reproduce the aspect of fit on the computer and to place the clothing optimally on
the body, the contact points for appropriate clothing are defined in different sizes, graduations
in width and cuts as well as for different postures.

 Figure 9: Processing the 2D CAD data as a DXF file (left) and integration of the additional
                             information in an XML file (right)

Physically-Based Cloth Simulation in Virtual Reality (WSI/GRIS)
The physically based cloth simulation is responsible for sewing the pre-positioned garments
along the seam lines and for computing the drape of the clothes on the avatar.

Physical Model and Numerical Solution
To compute realistic animations of clothes, we developed an efficient model based on finite
elements for viscoelastic, highly flexible surfaces. It is particularly designed for numerically
stiff materials such as textiles because it yields linear equations in each time step and allows
fast time stepping in an implicit integration method. This is achieved by reducing the
nonlinear elasticity problem to the planar, linear case in each step. With this model, we are
able to assemble garments from CAD cloth patterns [WKK+02], seam these together, and
animate the cloth in dynamic scenes with any chosen material properties (see figure 10). This
results in a physically accurate but also fast simulation. The basic idea in our approach is to
use a linear strain formulation and to construct a rotated rest state for each element [EKS03].
The arising ordinary differential equations time are solved by an implicit Euler method.
Textiles show very different physical behavior in weft and warp directions, so we model
elastic and viscous material parameters for the two directions independently. Material
measurements are carried out with the Kawabata evaluation system for the two Young
moduli, the shear modulus and the Poisson number, which controls the transverse contraction.
Additionally, the bending moduli describe the curvature elasticity in the weft and warp
directions. In order to model the exact hysteresis effects of the corresponding tissue, dynamic
material parameters are measured with the Kawabata and Zwick systems.

       Figure 10: Starting with the pre-positioned cloth patterns, the draping is simulated.

Collision Detection and Response
Interactions of the textile with itself and other objects play an important role in physically
based animation in order to model collisions and friction and to produce realistic behaviour.
We use hierarchies of discrete oriented polytopes (k-DOPs, [EKK+01, MKE03]) to
approximate the objects of the simulation. As the meshes in cloth simulations deform almost
arbitrarily, efficient update mechanisms for the hierarchies are essential. The hierarchies can
be built by a top-down splitting method. Figure 11 shows such an 18-DOP-hierarchy for an
avatar. The k-DOP hierarchies can be updated efficiently by merging the bounding volumes
from bottom to top.
Since self-collisions are crucial for realistic cloth simulation, they must not be neglected by
the collision detection and response. We combine the idea of normal cones with the k-DOP
hierarchy to estimate the surface curvature for the region covered by a hierarchy node. Thus,
parts of the textiles, where self-intersections are impossible due to their low curvature, are
identified and skipped during the self-collision test.
The collisions of the detected particles have to be resolved using a collision response scheme.
In our system we therefore implemented three different collision response schemes
   •    Constraint based collision response
   •    Force based collision response
   •    Iterative impulse based collision response
For most collision cases the constraint based method is used, because it turned out to be
exceedingly valuable in order to avoid collisions before they occur and to achieve large time

Figure 11: 18-DOP hierarchy for an avatar.             Figure 12: Interactive manipulation of

Interactive Manipulation of Clothes in Virtual Reality (WSI/GRIS)
When we try-on real clothes, we frequently adjust the garments on our body manually. To
provide an equivalent in a virtual try-on scenario, we developed interaction techniques which
allow to select and drag parts of the garments during the physically based simulation
[KSF+03, KSW+03, WKS+03]. In our system, this can be accomplished by utilizing Virtual
Reality input devices that allow 6 degrees of freedom to select parts of the clothes, or more
precisely, vertices of the underlying mesh. The selected points are visualized by small cubes,
which can be moved in the scene (see figure 12). The transformations of the selected vertices
are then integrated into the simulation as constraints. When the constraints are released, the
cloth relaxes due to internal forces and gravity. This technique allows moving the simulated
garments into shape, just like a real person does after putting on real clothes. Moreover, it is a
basic tool for virtual garment design. In the future, we want to enable a tailor to experiment
with different cloth shapes, creases, and seams in Virtual Reality.

High-Quality Visualization (University of Bonn)
Realistic and high quality visualization is essential to provide the „look & feel“ of cloth,
which depends on the material properties of the cloth surface.
The whole visualization is based on the usage of bidirectional texture functions (BTF) as
introduced by Dana et al. [DANA99]. A BTF dataset can be interpreted as a set of images of a
flat material surface viewed and lit under a discrete set of direction. Therefore, anisotropic
reflection properties of the material, subsurface light transport, interreflections, self-
shadowing, occlusion and foreshortening are captured and can be reproduced during

BTF Measurement - Laboratory
A complete measurement laboratory was set up to allow the automatic measurement of
reflection properties of flat material samples, as shown in figure 13. It consists of a robot, a
rail system with a moveable 14 Megapixel high-end
digital camera and a HMI light source, which simulates
the sun emission spectrum. The whole lab is controlled
via self-written computer programs.
The VTO textiles were measured out of 81 different
viewing directions. For every viewing direction 81
different lighting directions were used resulting in a
dataset of 6561 images per sample. Details are described
in Sattler et. al [Sattler03].
                                                                 Figure 13. Laboratory setup.
BTF Measurement - Postprocessing
The captured images represent a data amount of 90GB. This amount is reduced by registering
only a representative part of the surface and cut out of the corresponding images. To allow the
texturing of large objects, the images are made repeatable using blending methods. For further
reduction of the data a principal component analysis (PCA) as well as clustered PCA
[Müller03] are applied resulting in up to 24MB for a 256x256 Pixel BTF, where the chosen
texture size depends on the material structure and quality requirements. This compression
allows for fast decompression, advanced multi-texturing algorithms as explained in the
rendering part and is therefore especially suitable for rendering.
Besides the measurement of the material samples for the VTO project, several other samples
were measured and made publicly available through an internet site ( ).

Rendering (University of Bonn)
High-quality visualization of the simulated cloth with the measured material surface reflection
properties on a “virtual-mirror” is a main goal of the Virtual Try-On project. The rendering is
the last part of the process chain; the simulated cloth geometry and the desired material
selection besides the measured BTF data serve as input for this part.

Macroscopic self-shadowing
Macroscopic self-shadowing is an important visual clue to recognize the draping of cloth.
Therefore, we compute a local shadow value for each vertex of the geometry [Ganster02].
This is done using an hemi-cube approximation of the hemisphere of each vertex, defined by
its normal. Rendering this “view” of the vertex, the incoming radiance out of the discretized
directions is determined and stored. Interpolating between values of neighboring vertices
yields in a smooth result. By now, the geometry could be rendered with simple texturing and
correct self-shadowing. To add the material reflection properties the principal components of
the BTF data are incorporated.

BTF reconstruction and rendering
Using high-end graphics accelerators and shading language programming the reconstruction
of the BTF out of the principal components is possible at interactive frame rates for the whole
geometry [Sattler03]. This is done by evaluating the current viewing and lighting direction for
each vertex and using precomputed weights. Using multitexturing the appropriate texture for
each triangle is computed. To achieve smooth results and avoid edge artifacts blending is
used. Comparisons with other rendering methods show the superior quality of PCA based
BTF rendering [Meseth03a]. But using more advanced LPCA compression the rendering
quality can still be improved [Klein03,Meseth03b].
Comparison between “real” and “virtual” cloth
The comparison between “real” and “virtual” cloth shows, that using only point or directional
light sources results in a non natural illumination condition. Therefore, we integrated the
possibility to use high dynamic range images (HDRI) to allow for a real-world illumination.
As explained in the section “Macroscopic Self-Shadowing”, at each vertex the radiance of the
incident light from the measured BTF directions is computed and integrated during the BTF
reconstruction. To acquire the HDRIs from real world locations, e.g. shops, a portable
measurement system was build. The interactive change of viewing positions and HDRIs were
integrated in the BTF renderer to allow the customer a high degree of freedom for her/his
judgment (see figure 14).

  Figure 14. BTF Rendering of different cloth materials in a measured HDRI environment.

The comparison reveals also, that the geometry silhouette is an important visual clue. To
incorporate a correct silhouette representation into the rendering the laboratory was enhanced
to also acquire the material silhouette and a new rendering technique for experimentally
recorded real material silhouettes was developed.

Dynamic geometry
If the avatar pose is changed, e.g. the customer turns around in front of the “virtual mirror”,
the cloth geometry changes. Therefore, geometry simulation and lighting computations have
to be carried out in real-time. Due to the computational complexity of the draping simulation
as well as the lighting simulation an algorithm was developed, that is capable to estimate the
cloth geometry together with vertex based shadow information. Based on a statistical
evaluation of geometry and shadow information which is precomputed for most postures and
typical movements of humans trying on a certain cloth.

Virtual Shop
The VR techniques that have been developed in the course of the project Virtual Try-On have
been incorporated within an virtual shop infrastructure which will be installed at the point-of-
sales (cove&co) for evaluation. By providing an additional realisation of an internet related
infrastructure, the E-Commerce Shop, we will supplement conventional distribution channels.
Both applications will be based on a common user paradigm, which provides a uniform way
of interaction independent from the customers location and access point. Figure 15 shows
screenshot of both the virtual shop and E-Commerce Shop, which is currently under

        Figure 15: Virtual shop at the point-of-sales (left). E-Commerce-Shop (right).

Virtual Mirror
One of the fundamental ideas within the conception of the Virtual Try-On technology chain is
to provide a life-size display of the customers virtual counterpart dressed in the chosen pieces
of clothing. By incorporating / combining suitable display techniques and hardware, as well
as novel cloth rendering techniques we enable the customer to visualize himself / herself
wearing a variety of combinations of different garments from different views (see figure 16).

  Figure 16 : Virtual Mirror (left). Comparing simulation results with real garments (right)
Thereby, getting an realistic impression of the “look and feel” of the garment, this kind of
presentation will serve as an important decision support, leading to an enhanced shopping
experience for the customer.
   3.      Experiences, Assessments
In the course of the project, up to now, a complete, prototypical integrated shop infrastructure
for customized garment retail at the point-of-sales has been developed. A first small collection
of pieces of clothing is being supplied by Odermark and Hohenstein. Although, certainly,
calculating speed needs to be improved, (latest) evaluations, have proven that (basic)
concepts to be competitive / effective. Especially comparisons of real garments with the
corresponding simulation results show a high degree of correspondence (see figure 16 left).
Regarding the high quality visualization the project definitely demonstrates the great
improvement of the visualization of cloth by using measured reflectance properties of the real
world materials. In addition the greatly improved realism of real reflectance properties of the
cloth under natural illumination conditions provides the user with a much better “look & feel”
of the material than previous rendering techniques. This way the client can already judge
physical material properties based on the visualization. During the project it became clear,
that the collection of reflectance properties of the different cloth materials used in the textile
industry is a critical part in the whole visualization chain. In the context of the mass market
more optimized labs will be necessary to acquire all this data.

   4.      Realization Potential, Outlook
For the first time, Virtual Try-On has developed a complete process chain regarding the photo
realistic visualization and simulation of individual customers and garments. Starting with the
creation of individual customer avatars, up to a realistic, physically-based simulation of cloth
as well as a the life-size presentation of the customers digital twin wearing the selected cloth,
is provided.
All members of the project consortium intend to further transfer the gained experiences and
the acquired knowledge into products for retail and manufacturing. Here the benefits will
mainly include a reduced cost risk (manufacturing on demand), less cost through rapid
prototyping capabilities, less storage as well as a closer customer relationship.
Thereby, it will be expected, that technology and ideas developed in the course of the Virtual
Try-On project will contribute to the increasing market of individualized products within the
area of garments.
Last but not least it should be mentioned that based on the outstanding visualization results
achieved in this project also further applications of this technology in the area of the
automotive industry and architecture were initiated.

[BW98]            Baraff, D. and Witkin, A. (1998). Large steps in cloth simulation. In Cohen, M., editor,
                  SIGGRAPH 98 Conference Proceedings, Annual Conference Series, pages 43–54, Orlando,
                  FL, USA. ACM SIGGRAPH.

[CK02]            Choi, K.-J. and Ko, H.-S. (2002). Stable but responsive cloth. In ACM Transactions on
                  Graphics (SIGGRAPH 2002), volume 21.

[DANA99]          Kristian J. Dana and Bram van Ginneken and Shree K. Nayar and Jan J. Koenderink,
                  Reflectance and Texture of Real-World Surfaces, ACM Transactions on Graphics, 1999.
[EGH03]       Ebert, A., Ginkel, I., and Hagen, H.: Rule-based Morphing Techniques for Interactive
              Clothing Catalogs. Dagstuhl 2003, Scientific Visualization Proceedings, 2003.

[EKK+01]      Olaf Etzmuß and Michael Keckeisen and Stefan Kimmerle and Johannes Mezger and
              Michael Hauth and Markus Wacker, A Cloth Modelling System for Animated Characters,
              in: Proceedings Graphiktag, 2001.

[EKS03]       Olaf Etzmuß, Michael Keckeisen and Wolfgang Straßer, A Fast Finite Element Solution for
              Cloth Modelling, in: Proceedings of Pacific Graphics 2003.

[ESD03]       Ebert, A., Schädlich, J., and Disch, A.: Innovative Retexturing Using Cooperative Patterns.
              IASTED International Conference on Visualization, Imaging and Image Processing (VIIP
              2003), Malaga, Spain, 2003.

[FGL03]       Fuhrmann, A., Gross, C., and Luckas, V. (2003a). Interactive animation of cloth including
              self collision detection. Journal of WSCG, 11(1):141–148.

[FGLW03]      Fuhrmann, A., Gross, C., Luckas, V., and Weber, A. (2003b). Interaction-free dressing of
              virtual humans. Computer & Graphics, 27(1):71–82.

[FSG03]       Fuhrmann, A., Sobotka, G., and Gross, C. (2003c). Distance fields for rapid collision
              detection in physically based modeling. In Proceedings of GraphiCon 2003.

[Ganster02]   B. Ganster, R. Klein, M. Sattler, R. Sarlette, Realtime Shading of Folded Surfaces, CGI
              2002, 2002.

[GFL03]       Gross, C., Fuhrmann, A., and Luckas, V. (2003). Automatic pre-positioning of virtual
              clothing. In Proceedings of the Spring Conference on Computer Graphics, pages 113–122.

[Hauth02]     M. Hauth, O. Etzmuss, B. Eberhardt, R. Klein, R. Sarlette, M. Sattler, K. Daubert, J. Kautz,
              Cloth Animation and Rendering, Eurographics 2002, 2002.

[HE01]        Hauth, M. and Etzmuß, O. (2001). A high performance solver for the animation of
              deformable objects

[KKM+03]      S. Kimmerle, M. Keckeisen, J. Mezger, M. Wacker, TüTex: A Cloth Modelling System for
              Virtual Humans, in: Proceedings of 3D Modelling 2003.

[Klein03]     R. Klein, J. Meseth, G. Müller, R. Sarlette, M. Guthe, Á. Balázs, RealReflect - Real-time
              Visualization of Complex Reflectance Behaviour in Virtual Protoyping, ACM proceedings
              of Eurographics 2003 Industrial and Project Presentations, 2003.

[KSF+03]      M. Keckeisen, S. L. Stoev, M. Feurer and W. Straßer, Interactive Cloth Simulation in
              Virtual Environments, in: Proceedings of IEEE Virtual Reality 2003.

[KSW+03]      Michael Keckeisen, Stanislav L. Stoev, Markus Wacker, Matthias Feurer and Wolfgang
              Straßer, A User Study On Advanced Interaction Techniques in the Virtual Dressmaker
              Application, in: Proceedings of HCI International 2003.

[LV94]        Lazarus, A. and Verroust, A.: Feature-based shape transformation for polyhedral objects.
              Fifth Eurographics Workshop on Animation and Simulation, Oslo, Norway, 1994.

[Meseth03a]   J. Meseth, G. Müller, M. Sattler, R. Klein, BTF Rendering for Virtual Environments,
              Virtual Concepts 2003, 2003.

[Meseth03b]   J. Meseth, G. Müller, R. Klein, Preserving Realism in Real-Time Rendering of
              Bidirectional Texture Functions, OpenSG Symposium 2003, 2003.

[MKE03]       J. Mezger, S. Kimmerle and O. Etzmuß. Hierarchical Techniques in Collision Detection for
              Cloth Animation. In Journal of WSCG 2003.
[Müller03]    G. Müller, J. Meseth, R. Klein, Compression and Real-Time Rendering of Measured BTFs
              Using Local PCA, Vision, Modeling and Visualisation 2003, 2003.

[Sattler03]   M. Sattler, R. Sarlette, R. Klein, Efficient and Realistic Visualization of Cloth,
              Eurographics Symposium on Rendering 2003, 2003.

[WKK+02]      Markus Wacker, Michael Keckeisen, Stefan Kimmerle and Johannes Mezger, Fortschritte
              in der Textilsimulation für Fashion on Demand, in: Tagungsband 1. Paderborner Workshop
              Augmented Reality Virtual Reality in der Produktentstehung, 2002.

[WKS+03]      Markus Wacker, Michael Keckeisen, Stanislav L. Stoev and Wolfgang Straßer, A
              Comparative Study On User Performance in the Virtual Dressmaker Application, in:
              Proceedings of ACM VRST 2003.

                    HUMAN SOLUTIONS GmbH, Kaiserslautern
                    Project coordination, Customer –Virtualization, Virtual Shop at the POS,
                    Virtual E-Commerce Shop

                    Institute of Computer Science II, University of Bonn
                    Efficient and realistic visualization of Cloth

                    WSI/GRIS, University of Tübingen
                    Physically-Based Cloth Simulation

                    Fraunhofer IGD, Darmstadt
                    Pre-Positioning of Garment, Interactive Java-based Garment Simulation

                    Hohenstein Institute for Clothing Physiology
                    Physical Textile Parameters

                    DFKI GmbH, Kaiserslautern
                    Interactive individual clothing catalog

                    Odermark, Goslar
                    Individualized garment production

                    Cove & Co., Düsseldorf
                    Pilot installation, innovative shop concepts