AUGMENTED REALITY INTERACTION TECHNIQUES

Design and Evaluation of the Flip-Flop Menu

Mickael Naud, Paul Richard and Jean-Louis Ferrier

LISA Laboratory, University of Angers, 62 Avenue Notre-Dame du Lac, Angers, France

Keywords:

Augmented reality, 3D interaction techniques, 3D menus, Human performance.

Abstract:

We present the evaluation of a bimanual augmented reality (AR) interaction technique, and focus on the effect

on viewpoint and image reversal on human performance The interaction technique (called ﬂip-ﬂop) allows the

user to interact with a 3D object model by using a V-shaped AR menu placed on a desk in front of her/him.

The menu is made of two complementary submenus. Both submenus (master and slave) are made of four

Artag ﬁducial markers. The functionalities of the slave submenu are the following : (1) increase/decrease the

size or rotate/stop the 3D object, (2) apply a color (one over four) or (3) a 2D texture (one over four) on the 3D

object and (4) apply predeﬁned material parameters. Each event is triggered by a masking of a Artag marker

by the the user’s right or left hand. 40 participants were instructed to perform actions such as rotate the object,

apply a texture or a color on it, etc. The results revealed some difﬁculties due to the inversion of the image on

the screen. Finally, although the proposed interaction technique is currently used for product design, it may

also be applied to other ﬁelds such as edutainment, cognitive/motor rehabilitation, etc. Moreover, other tasks

than the ones tested in the experiment may be archived using the menu.

1 INTRODUCTION

Virtual Reality (VR) allows a user to experience both

immersion and real-time interactions that may involve

visual feedback, 3D sound, haptics, smell and taste.

Instead, Augmented Reality (AR) proposes to im-

merse virtual objects in the real world. In this sense,

AR seems to be more interesting for product design

because it allows the designer or the end-user to both

visualize and interact with the virtual mockup of the

designed product in the real world (its working envi-

ronment). There is therefore no need to build a whole

3D world (kitchen, bedroom, etc.) to estimate the es-

thetic properties of the product in a given context.

Although several compelling AR systems have

been demonstrated (see (Azuma, 1997) for compre-

hensive survey), many serve merely as information

browsers, allowing users to see or hear virtual data

embedded in the physical world (Ishii and Ullmer,

1997).

In order to make AR fully effective, simple and

efﬁcient interaction techniques need to be developed

and validated through user studies. Moreover, inher-

ent drawbacks of AR techniques such as real/virtual

image discrepancy, system calibration (free marker

AR), or ergonomic problems related to head-mounted

displays have to be overcome.

In most of AR techniques designed for control ap-

plication, each ﬁducial marker is used to trigger a sin-

gle event or action. This may become a drawback

when a large number of event / action are required.

In this paper, we present an evaluation of a bi-

manual augmented reality (AR) interaction technique

based on a V-shaped menu that uses 8 ﬁducial mark-

ers. The menu is made of two complementary sub-

menus. Both submenus (master and slave) are made

of four ARtag ﬁducial markers (Fiala, 2004).

The paper is organized as follows. In section 2,

we give an overview of the related work. In sec-

tion 3 we describe and provide ﬁnite state modeling

of the ”Flip-ﬂop” interaction technique. In section 4

we present an experimental study that was carried out

in order to evaluate the ”Flip-ﬂop” interaction tech-

nique in different experimental desk-top conﬁgura-

tions. The paper ends by a conclusion.

345

Naud M., Richard P. and Ferrier J.

AUGMENTED REALITY INTERACTION TECHNIQUES - Design and Evaluation of the Flip-Flop Menu.

DOI: 10.5220/0001807903450352

In Proceedings of the Fourth International Conference on Computer Graphics Theory and Applications (VISIGRAPP 2009), page

ISBN: 978-989-8111-67-8

2 RELATED WORK

Most AR application examples of design work devel-

oped up to present are limited to large-scale objects

that the designers are not able to grasp and move. In

these examples, the information provided was lim-

ited to visual information, lacking physical interac-

tions between the observer and the object. For ex-

ample Fata Morgana project (Klinker et al., 2002),

designers were able to walk around a newly designed

virtual car for inspection and comparisons with oth-

ers.

In this context, Lee et al. proposed to bring the users

hand into the VE using a Mixed Reality platform (Lee

et al., 2004). The hand region was separated and in-

serted into Virtual Environment (VE) to enhance real-

ity.

Regenbrecht et al. developed a Magic Pen that

uses ray-casting techniques for selection and ma-

nipulation of virtual objects (Regenbrecht et al.,

2001a). Camera-based tracking was performed us-

ing ARToolkit, a software library that supports tan-

gible user interaction with ﬁducial markers (Kato

and Billinghurst, 1999). Two or more light emitting

diodes (LEDs) were mounted on a real pen barrel. A

camera was used to track the position of the pen, using

the LEDs as position markers. Direct manipulation

was performed with the end of the pen, or a virtual

ray can be cast in the direction that the pen is point-

ing. This provides a cable-less interaction device for

AR environments.

Slay et al. have developed an AR applica-

tion called ARVIS (Augmented Reality Visualization)

that allows users to view and interact with three-

dimensional models (Slay et al., 2002). The key AR

technology incorporated in the ARVIS is also AR-

Toolkit. Based on ARVIS, three different interaction

techniques which extend the use of ARToolkit ﬁducial

markers were proposed.

In the ﬁrst one Array of Markers ﬁducial markers

are simply arranged on a surface. To change one of

the attributes of the virtual object the associated ﬁdu-

cial marker was covered so it was no longer recog-

nized by ARVIS. To turn the attribute back, another

marker associated with the opposite of the task that

was performed ﬁrst was covered.

In a second interaction technique called ”Menu Sys-

tem” the ﬁducial markers were used as a menuing sys-

tem (Slay et al., 2002). Five small markers were cre-

ated to represent ﬁve menu options. Each layer of the

menuing system had a limit of ﬁve options. To select

one of the options, a card has to be ﬂipped over. The

virtual representations of the cards would then change

to show the next layer of the menuing system. This

was very restrictive on the design of the menu system.

It was also found very cumbersome to reach tasks in

the lower layers of the menu hierarchy, and because of

the limited number of tasks to choose from, the tech-

nique seemed very heavy-handed.

In a third interaction technique called Toggle Switches

the ﬁducial markers were used as toggle switches.

This technique can be seen to stem from the Array of

Markers interaction technique. The pattern had a ﬁdu-

cial marker on each side of the card. The two ﬁducial

markers were associated with tasks that would pro-

vide opposite functions. This technique was used to

connect the nodes of a directed graph together. The

nodes were connected when the ﬁrst side of a card was

shown, and disconnected when it was ﬂipped over.

The system remembers the state that each option is

in. ARVIS continually checks to see if any cards are

present in the ﬁeld of view. If any changes have been

made it generates the appropriate VRML model and

displays this as the virtual object.

The toggle switch system was found easy to use. Out

of the three proposed interaction techniques, it was

found to be the most easy and self-explanatory to use.

Turning the cards over, in effect toggling the behavior

exhibited by the virtual models seemed very natural.

Dias et al. have developed a series of tools based

on ARToolkit : the Paddle and the Magic Ring (Dias

et al., 2004). They both have a speciﬁc marker at-

tached. As a visual aid, when a marker is recognized

by the system, a virtual paddle or virtual blue circle

is displayed on top of it. The Magic Ring (a small

marker attached to a ﬁnger, by means of a ring) is

used in several interaction tasks, such as object pick-

ing, moving and dropping, object scaling and rotation,

menu manipulation, menu items browsing and selec-

tion, and for all various types of commands given to

the system.

A similar AR/VR uniﬁed user interface was pro-

posed by Piekarski and Thomas . This interface called

the Tinmith-Hand is based on the ﬂexible Tinmith-

evo5 software system (Piekarski and Thomas, 2002).

Using some modeling techniques, based on construc-

tive solid geometry, tracked input gloves, image plane

techniques, and a menu control system, it is possi-

ble to build applications that can be used to construct

complex 3D models of objects in both indoor and out-

door settings.

In the context of collaborative design, Szalavri and

Gervautz have introduced a multi-functional 3D user

interface, the Personal Interaction Panel (PIP), which

consists of a magnetic tracked simple clipboard and

pen, containing augmented information presented to

the user by see-through HMDs. The low technical

level the panel and pen itself allows ﬂexible design of

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

346

the interface and rapid prototyping. The natural two-

handed interaction supported by this device makes the

device ﬁt to a rich variety of applications (Szalavri

and Gervautz, 1997).

Buchmann et al have proposed a technique for nat-

ural, ﬁngertip-based interaction with virtual objects in

Augmented Reality (AR) environments (Buchmann

et al., 2004). They use image processing software and

ﬁnger- and hand-based ﬁducial markers to track ges-

tures from the user, stencil buffering to enable the user

to see their ﬁngers at all times, and ﬁngertip-based

haptic feedback devices to enable the user to feel vir-

tual objects. This approach allows users to interact

with virtual content using natural hand gestures. It

was successfully applied in an urban planning AR ap-

plication.

Besides more traditional AR interaction tech-

niques, like mouse raycast, MagicBook, and

models-on-marker e.g. (Regenbrecht et al., 2001b)

some techniques were proposed by Regenbrecht and

Wagner (Regenbrecht and Wagner, 2002). The main

one, called cake platter uses a turnable, plate-shaped

device functions as the central location for shared

3D objects. The objects or models can be placed on

the platter using different interaction techniques, e.g.

by triggering the transfer from a 2D application or

by using transfer devices brought close to the cake

platter.

Another technique uses a Personal Digital Assistant

(PalmPilot IIIc) as a catalogue of virtual models; the

main form of interaction within the system being

model selection and transfer to and from the cake

platter. Still in the context of the cake platter, a

clipping plane and lighting technique is used to

see what is inside of a virtual object. The user

holds a (transparent or opaque) real plane in his or

her hand to clip through the model on the cake platter.

Others more recent applications, based on the

same approach (use of Artag ﬁducial markers), are

propose by Xin (Xin et al., 2008) and Hender-

son (Henderson and Feiner, 2008).

3 FLIP-FLOP INTERACTION

TECHNIQUE

In most of AR techniques designed for control appli-

cation, each ﬁducial marker is used to trigger a single

event or action. This may become a drawback when

a large number of event / action are required. The

approach we propose here is based on a bimanual in-

teraction and a V-shaped menu that allows to trigger

many event/action with only 8 ﬁducial markers.

3.1 Description

The ﬂip-ﬂop interaction technique uses 9 ARtag ﬁdu-

cial markers arranged on a table, as illustrated in Fig-

ure 1. This speciﬁc number of markers was partially

determined based on technical and ergonomic consid-

erations. Firstly, the size of the markers has to be large

enough to facilitate the interaction technique. Sec-

ondly, all the markers have to be always viewed by

the camera and recognized. The number of markers

was also determined by the number of menu options

we needed to implement.

The main (central) ARtag ﬁducial marker is used

to display the 3D model(a virtual mannequin). Side

parts of the menu are made-up of 4 ARtag ﬁducial

markers. The left part of the menu (the master sub-

menu) allows to activate different functionalities. The

right part of the menu (the slave sub-menu) is used to

interact with the 3D model.

We called this interaction technique ”ﬂip-ﬂop” be-

cause of the multiples back-and-forth movements that

the user must do between the master sub-menu and

the slave sub-menu.

Figure 1: Top view of the V-shaped ﬁducial markers ar-

rangement.

Figure 2: Illustration of the functionalities of the slave sub-

menu activated from the master sub-menu.

AUGMENTED REALITY INTERACTION TECHNIQUES - Design and Evaluation of the Flip-Flop Menu

347

The ﬁgure 2 illustrates the functionalities of the

developed application, activated by the covering of

the ﬁducial markers of the master sub-menu by the

user’s hand. These functionalities are the following :

• Colors exploration: Exploration of the colors

palettes (ﬁgure 3). The change of palette is done

each 800ms. This value was tuned using prelimi-

nary testings involving few participant.

• Model animation: Activation of functionalities

allowing to (1) reduce or (2) increase the size of

the mannequin and (3) to make the mannequin ro-

tate or (4) to stop it in a speciﬁc position.

• Texture database exploration: Exploration of

the different preset texture sets. The display of

a new texture sets is done automatically each

800ms.

• Materials exploration: Activation of functional-

ities allowing to change the material that simulate

the fabric visual aspect.

Figure 3: Exploration of the colors palettes displayed on the

slave sub-menu ﬁducial markers from the master sub-menu.

3.2 State Automata Modelling

The automaton relative to the activation of the slave

sub-menu functionalities from the master sub-menu is

presented in ﬁgure 4. This automaton is composed of

the following states:

• C : ”color” (application of a color among 4)

• I : ”image” (application of an image among 4)

• T : ”transformation” (change of size and rotation)

• M : ”materials” (application of materials)

The changes of state are activated by the covering

of the master sub-menu ﬁducial markers by the user’s

hand. The following variables are involved:

• T

: activation of the ”color” state,

Figure 4: Finite state automaton of the slave sub-menu.

• T

: activation of the ”image” state,

• T

: activation of the ”transformation” state,

• T

: activation of the ”materials” state.

Application of the colors, displayed on the slave

sub-menu ﬁducial markers, on virtual clothes, is based

on the same type of automaton as the precedent, in

which states C, I, T and M are replaced by the colors

, C

, and C

, and the buttons T

, T

, and T

by the buttons T

, T

, and T

Animation of the virtual mannequin is currently

limited to a change of size (enlarging or reduction)

and to a rotation. The automaton presented in ﬁgures

5 and 6 represent a modeling of the associated inter-

action.

The automaton relating to size modiﬁcation of the

model is composed of the following states:

• A : model size increases (by steps)

• I : same size

• D : model size decreases (by steps)

Figure 5: Finite state automaton controlling the size of 3D

model.

As before, the changes of state of this automaton

are activated by the covering of the master sub-menu

ﬁducial markers by the user’s hand. The following

variables are involved:

• T

: increase of the size

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

348

• T

−

: decrease of the size

Thus, as the user keeps covering the ﬁducial

marker associated with the ”+” symbol (ﬁg. 2), the

size of the model increases incrementally.

Similarly, when the user keeps covering the ﬁdu-

cial marker associated with the ”-” symbol, the size of

the model decreases incrementally.

Figure 6: Finite state automaton controlling the rotation of

the model.

The automaton handling the rotation of the 3D

model is composed of the following states (ﬁg. 6):

• A : static model

• R : model in rotation

Changes of state are snapped by the covering

of the ﬁducial markers by the user’s hand. Slave

sub-menu functionalities relative to model animation

are modeled by the following variables:

• T

: activation of the rotation

• T

: stopping the rotation

Thus, as the user keeps covering the ﬁducial

marker associated with the > symbol(ﬁg. 2), the 3D

model is rotated. Similarly, as the user keeps covering

the ﬁducial marker associated with the k symbol, the

3D model stops.

4 EVALUATION

Augmented reality techniques may involve differ-

ent conﬁgurations : (1) immersive conﬁgurations in

which the users are equipped with either an optical

see-through or a video see-through head-mounted dis-

play, (2) desk-top conﬁgurations that may be based

on a large screen placed in front of the users, and (3)

embedded conﬁgurations that involve mobile devices

such as PDAs.

In our context, desk-top conﬁgurations may be in-

teresting because the user could see himself/herself on

the screen, manipulating a virtual product. This as-

pect is crucial in the context of fashion design where

the product (clothe) are worn buy the user. In such

a conﬁguration, the video camera has to be placed in

front of the user. Then the images on the screen have

to be reversed.

4.1 Aim

The objective of this experiment is to study the usabil-

ity, and the effectiveness of this technique for differ-

ent conﬁgurations in which the video camera is placed

in front or behind the user (ﬁgure 7).

4.2 Task

The task was split into 9 sub-tasks that the sub-

jects have to perform in a sequential predeﬁned order.

These sub-tasks are the following:

1. Scroll all textures (5 sets of 4 textures)

2. Scroll colors and select the red set

3. Apply the clearest red color on the 3D model

4. Scroll colors and select the blue set

5. Apply the darkest blue color on the 3D model

6. Select transformations

7. Shrink the character (ﬁve steps)

8. Enlarge the character (three step)

9. Rotate the character and stop it (one turn)

4.3 Experimental Protocol

A total of 40 volunteers subjects (20 males and 20 fe-

males) right-handed, aged 15 to 35 years, participated

in the experiment. They never used augmented real-

ity techniques before. The subjects were divided in

4 groups (G1,G2,G3,G4). Each group has to perform

the task in the following conditions :

• C1 : the webcam was placed in front of the user,

and the image was not reversed

• C2 : the webcam was placed in front of the user,

and the image was reversed

• C3 : the webcam was placed behind the user, and

the image was reversed

• C4 : the webcam was placed behind the user, and

the image was not reversed

Task has to be repeated 5 times by each subject.

A 2 minutes rest period was imposed between each

trial. Task completion time was measured for each

single trial.

AUGMENTED REALITY INTERACTION TECHNIQUES - Design and Evaluation of the Flip-Flop Menu

349

(C1) (C2)

(C3) (C4)

Figure 7: Experimental conﬁgurations for the evaluation of

the ﬂip-ﬂop menu.

4.4 Results

The results illustrated in ﬁgure 8 are presented in the

next three sub-sections. They were analyzed through

a two-way ANOVA. We ﬁrstly examine the effect of

camera position on subjects performance. Data from

groups G1 and G2 were compared to the data from

groups G3 et G4.

Then, the effect of image reversal on subjects per-

formance was examined. Therefore, data from groups

G1 and G3 were therefore compared to the data from

groups G2 et G4. Finally we look at the joint effect of

the studied parameters.

Figure 8: Mean task completion time vs. experimental con-

ditions.

4.4.1 Effect of Camera Position

We observed that the effect of camera position on task

completion time is statistically signiﬁcant (F(1, 19) =

17.7, P < 0.0005). Subjects that have the camera in

front of them performed the task, on average in 71.5

seconds (std = 5.2) while those whose camera was

placed behind achieved the task in 67.7 seconds (std

= 4.6).

4.4.2 Effect of Image Reversal

We observed that the effect of image reversal on

task completion time is very signiﬁcant (F(1, 19) =

790.26, P < 0.0001). Subjects who performed with

the reversed image achieved the task, on average, in

84.55 seconds (std = 6.4) while those who performed

with none reversed image achieved the task in 54.7

seconds (std = 3.4). We observe that image reversal

has more signiﬁcant effect on user performance than

camera position.

(a) (b)

Figure 9: Task completion time over trial for condition 1

(a), condition 2 (b), condition 3 (c), and condition4 (C).

4.4.3 Joint Effect

The analysis of variance (ANOVA) reveals that the

experimental conditions have a signiﬁcant effect on

users performance (F(3,9) = 298.8, P < 0.0001).

The subjects of Group G1, G2, G3, and G4 have car-

ried out the task respectively in 86.9 seconds (std =

6.6), 56.2 seconds (std = 3.8), 82.2 seconds (std =

6.2), and 53.1 seconds (std = 2.9).

Results show that the joint effect of the tested pa-

rameters is not much signiﬁcant. Indeed, the sub-

jects of the group G3 (reversed image) performed the

task in 29.1 seconds faster than the subjects of group

G4 (non-reversed image) when the camera was posi-

tioned behind, while the subjects of group G1 (non

reversed image) performed the task in 30.7 seconds

more than the subjects of group G2 (reversed image),

when the camera was positioned in front.

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

350

4.4.4 Learning Process

Learning is deﬁned here by the improvement of sub-

jects performance during task repetitions. We asked

subjects to repeat 5 times the previously deﬁned task

4.2. The results show that the subjects of group G1

achieved the task in 102.5 seconds (std = 9.6) during

the ﬁrst trial and in 65.7 seconds (std = 6.5) during the

last trial. Subjects of groups G2, G3 and G4, achieved

the task respectively in 69.5 seconds (std = 6.2), 95.5

seconds (std = 9.0) and 67.4 seconds (std = 4.4) during

the ﬁrst trial and 49.1 seconds (std = 2.6), 64.8 sec-

onds (std = 5.9) and 47.5 seconds (std = 2.2) during

the last trial. This result in performance improvement

percentage of 34%, 29%, 32% and 30% respectively

for groups G1, G2, G3 and G4.

Figures 9(a), 9(b), 9(c) and 9(d) show learning

curves of each subject. We observed that for conﬁg-

urations in which the image is reversed, learning is

slower.

5 CONCLUSIONS

We presented the evaluation of a bimanual augmented

reality (AR) interaction technique, and focus on the

effect on viewpoint and image reversal on human per-

formance The interaction technique (called ﬂip-ﬂop)

allows the user to interact with a 3D object model by

using a V-shaped AR menu placed on a desk in front

of her/him. The menu is made of two complemen-

tary submenus. Both submenus (master and slave) are

made of four Artag ﬁducial markers. The functional-

ities of the slave submenu are the following : (1) in-

crease/decrease the size or rotate/stop the 3D object,

(2) apply a color (one over four) or (3) a 2D texture

(one over four) on the 3D object and (4) apply prede-

ﬁned material parameters. Each event is triggered by

a masking of a Artag marker by the the user’s right

or left hand. 40 participants were instructed to per-

form actions such as rotate the object, apply a tex-

ture or a color on it, etc. The results revealed some

difﬁculties due to the inversion of the image on the

screen. Finally, although the proposed interaction

technique is currently used for product design, it may

also be applied to other ﬁelds such as edutainment,

cognitive/motor rehabilitation, etc. Moreover, other

tasks than the ones tested in the experiment may be

archived using the menu.

REFERENCES

Azuma, R. (1997). A survey of augmented reality. In Pres-

ence: Teleoperators and Virtual Environments, pages

355–385.

Buchmann, V., Violich, S., Billinghurst, M., and Cockburn,

A. (2004). Fingartips: gesture based direct manipu-

lation in augmented reality. In GRAPHITE ’04: Pro-

ceedings of the 2nd international conference on Com-

puter graphics and interactive techniques in Australa-

sia and South East Asia, pages 212–221, New York,

NY, USA. ACM.

Dias, J. M. S., Santos, P., and Bastos, R. (2004). Gesturing

with tangible interfaces for mixed reality. In Camurri,

A. and (Eds.), G. V., editors, GW 2003,LNAI 2915,

page 399408, Berlin Heidelberg.

Fiala, M. (2004). Artag, an improved marker system based

on artoolkit. In NRC Publication Number: NRC

47166.

Henderson, S. and Feiner, S. (2008). Opportunistic controls

: Leveraging natural affordances as tangible user in-

terfaces for augmented reality. In VRST 2008, ACM

Symposium on Virtual Reality Software and Technol-

ogy, pages 211–218, Bordeaux, France.

Ishii, H. and Ullmer, B. (1997). Tangible bits towards seam-

less interfaces between people. In Bits and Atoms.

CHI 97., pages 234–241.

Kato, H. and Billinghurst, M. (1999). Marker tracking and

hmd calibration for a video-based augmented reality

conferencing system. In In 2nd IEEE and ACM Inter-

national Workshop on Augmented Reality, pages 84–

94, San Francisco, USA.

Klinker, G., Dutoit, A., Bauer, M., Bayer, J., Novak, V.,

and Matzke, D. (2002). Fata morgana - a presentation

system for product design. In ISMAR2002, pages 76–

85.

Lee, S., Chen, T., Kim, J., Kim, G. J., Han, S., and Z., P.

(2004). Using virtual reality for affective properties of

product design. In Proc. of IEEE Intl. Conf. on Virtual

Reality, pages 207–214.

Piekarski, W. and Thomas, B. H. (2002). Unifying aug-

mented reality and virtual reality user interfaces.

Regenbrecht, H., Baratoff, G., Poupyrev, I., and

Billinghurst, M. (2001a). A cable-less interaction de-

vice for ar and vr environnement. In Proc. ISMR,

pages 151–152.

Regenbrecht, H., Baratoff, G., and Wagner, M. (2001b). A

tangible ar desktop environment. In Computers and

Graphics, Special Issue on Mixed Realities.

Regenbrecht, H. T. and Wagner, M. T. (2002). Interaction

in a collaborative augmented reality environment. In

CHI 2002, Minneapolis, Minnesota, USA.

Slay, H., Thomas, B., and Vernik, R. (2002). Tangible user

interaction using augmented reality. In Third Aus-

tralasian User Interfaces Conference (AUIC2002),

Melbourne, Australia.

Szalavri, Z. and Gervautz, M. (1997). The personal inter-

action panel - a two-handed interface for augmented

AUGMENTED REALITY INTERACTION TECHNIQUES - Design and Evaluation of the Flip-Flop Menu

351

reality. In Publishers, B., editor, EUROGRAPHICS

97.

Xin, M., Sharlin, E., and Costa Sousa, M. (2008). Napkin

sketch - handheld mixed reality 3d sketching. In VRST

2008, ACM Symposium on Virtual Reality Software

and Technology, pages 223–226, Bordeaux, France.

GRAPP 2009 - International Conference on Computer Graphics Theory and Applications

352