Mapping of Realism in Rendering
onto Perception of Presence in
Augmented Reality
DIPLOMARBEIT
zur Erlangung des akademischen Grades
Diplom-Ingenieur
im Rahmen des Studiums
Wirtschaftsinformatik
eingereicht von
David Schüller-Reichl, BSc
Matrikelnummer 00825849
an der Fakultät für Informatik
der Technischen Universität Wien
Betreuung: Assoc. Prof. Dr. Hannes Kaufmann
Mitwirkung: Mag Dr.techn. Peter Kán
Wien, 22. November 2017
David Schüller-Reichl Hannes Kaufmann
Technische Universität Wien
A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.ac.at
Die approbierte Originalversion dieser Diplom-/ 
Masterarbeit ist in der Hauptbibliothek der Tech-
nischen Universität Wien aufgestellt und zugänglich. 
 
http://www.ub.tuwien.ac.at 
 
 
 
 
The approved original version of this diploma or 
master thesis is available at the main library of the 
Vienna University of Technology. 
 
http://www.ub.tuwien.ac.at/eng 
 

Mapping of Realism in Rendering
onto Perception of Presence in
Augmented Reality
DIPLOMA THESIS
submitted in partial fulfillment of the requirements for the degree of
Diplom-Ingenieur
in
Business Informatics
by
David Schüller-Reichl, BSc
Registration Number 00825849
to the Faculty of Informatics
at the TU Wien
Advisor: Assoc. Prof. Dr. Hannes Kaufmann
Assistance: Mag Dr.techn. Peter Kán
Vienna, 22nd November, 2017
David Schüller-Reichl Hannes Kaufmann
Technische Universität Wien
A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.ac.at

Erklärung zur Verfassung der
Arbeit
David Schüller-Reichl, BSc
Beingasse 30/B05, 1150 Vienna, Austria
Hiermit erkläre ich, dass ich diese Arbeit selbständig verfasst habe, dass ich die verwen-
deten Quellen und Hilfsmittel vollständig angegeben habe und dass ich die Stellen der
Arbeit – einschließlich Tabellen, Karten und Abbildungen –, die anderen Werken oder
dem Internet im Wortlaut oder dem Sinn nach entnommen sind, auf jeden Fall unter
Angabe der Quelle als Entlehnung kenntlich gemacht habe.
Wien, 22. November 2017
David Schüller-Reichl
v

Acknowledgements
First of all, I would like to express my gratitude to Peter Kán for his extraordinary
support, motivation, discussions and feedback. Many thanks to Hannes Kaufmann for
being the supervisor of this thesis.
I also want to thank all members of the VR Group at Vienna University of Technology. I
always got valuable input, support and feedback.
I would like to thank my parents and sisters, Caroline, Tamara and Daniela. I am the
person today because of them which finally led me to this point. I am grateful to my
parents for enabling me a course of education. They always were open to my thoughts
and accepted how I organised my educational path. Additionally, I want to thank my
brother in law, and best man Simon Pichler for his advise, our discussions, the motivation
to stay focused, and the valuable entertaining distractions. Also, I want to thank all my
friends for their valuable feedback and support as well as all persons who participated in
the user study.
Especially, I want to thank my wife Johanna for her endless support and the motivation
to follow my thoughts and ideas. I could have never written this thesis without her
support. Finally, I want to thank my daughters Maya and Mira. Every day they show
me that the best way of learning is simply trying, learning from failures and repeat.
vii

Kurzfassung
Augmented Reality (AR) integriert virtuelle computergenerierte Objekte in die reale
Umgebung. Im Idealfall fügen sich virtuelle Objekte in die reale Welt so ein, dass der
Benutzer die virtuellen Objekte so wahrnimmt als wäre es Teil der realen Umgebung.
Dass sich ein virtuelles Objekt in die reale Umgebung einfügt ist auch unter dem
Konzept der Präsenz bekannt. Ein hohes Maß an Präsenz (oder auch Anwesenheit) ist
besonders wichtig, wenn eine AR-Anwendung virtuelle Personen verwendet, um mit dem
Benutzer zu interagieren. In dieser Arbeit wird untersucht ob eine visuelle realistische
Darstellung (im weiteren Realismus) in AR-Anwendungen wichtig ist um ein hohes Maß
an Präsenz zu erzielen. Zwei Hypothesen wurden erstellt um den Einfluss von Realismus
auf die Wahrnehmung der Präsenz und auf den Benutzerkomfort bzw. Benutzererfahrung
innerhalb von AR-Anwendungen zu untersuchen. H1: "Die Erhöhung des Realismus
erhöht das Gefühl von Präsenz und Benutzerkomfort / Benutzererfahrung.". H2: "Der
Uncanny Valley Effekt kann innerhalb des Experiments beobachtet werden.".
Um die aufgestellten Hypothesen zu untersuchen wurde eine Benutzerstudie durchgeführt,
in der die Teilnehmer eine virtuelle Person mit unterschiedlichen Realitätsniveaus beob-
achteten. Jedes Realitätsniveau unterschied sich in Geometrie, Textur und Lichtquellen.
Die entwickelte AR-Anwendung beinhaltete ein Rendering-System, mit dem das Realitäts-
niveau der virtuellen Personen eingestellt werden konnte. Die Ergebnisse unterstützten
teilweise die erste Hypothese (H1) und zeigen, dass visueller Realismus ein wichtiger
Faktor ist, um ein höheres Gefühl der Präsenz in einer AR-Anwendung zu erreichen. Die
zweite Hypothese (H2) wurde nicht unterstützt. Wahrscheinlich aufgrund von technischen
Einschränkungen, die eine realistische virtuelle Darstellung einer virtuellen Person nicht
zuließen, so dass der Teilnehmer glauben würde, dass es tatsächlich eine reale Person
sein könnte.
Das wichtigste Novum dieser Arbeit ist die Fokussierung auf das Präsenzgefühl gegenüber
virtuellen Personen in AR-Anwendungen. Anders als Studien im Bereich VR weißen
aktuelle Studien im Bereich AR darauf hin, dass der Einfluss von visuellen Realismus
auf die Wahrnehmung der Präsenz unterschiedlich ist. Zukünftige Entwicklungen von
AR-Anwendung mit dem Ziel hoher Präsenz zu erreichen können die Ergebnisse dieser
Arbeit als Basis verwenden. Vor allem, wenn virtuelle Menschen verwendet werden, um
mit Benutzern zu interagieren.
ix

Abstract
Augmented Reality (AR) is about seamless integration of virtual computer-generated
objects into the real-world view. Ideally, virtual objects should blend into the real world
so that the user feels like the virtual objects are "here". The sense of something being
”here” is also known as the concept of presence. Presence is especially important if an AR
application is using virtual humans to interact with the user. This thesis examines if the
visual realism is essential to achieve the highest possible presence in an AR application.
Two hypotheses were posed to examine the effect of realism on the perception of presence
and the convenience of users within AR applications. H1: ”Increasing the level of realism
increases the sense of presence and convenience of users.”. H2: ”The Uncanny Valley
effect can be observed within the experiment.”.
The approach of this thesis to examine these hypotheses was to conduct a user study in
which the participants experienced a virtual human with a specific visual realism levels.
Each visual realism level differs in geometry, texture and lights. The developed AR
application included a rendering system which allows the levels of realism of the virtual
human to be set. The results partially supported the first hypothesis (H1) and indicated
that visual realism is an important factor to achieve a higher sense of presence within
an AR application. The second hypothesis (H2) was not supported, most probably due
to technical limitations which did not allow such a realistic virtual representation of a
human so that the participant would believe it could be a real person.
The main novelty of this thesis is its focus on the presence of virtual humans within AR.
Recent studies showed that the influence of visual realism on the sense of presence if
different in the field of AR than in VR. Future presence demanding AR application can
take the results of this thesis of basis to achieve a higher sense of presence. Especially, if
virtual humans are used to interact with users.
xi

Contents
Kurzfassung ix
Abstract xi
Contents xiii
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Expected Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Background and Related Work 5
2.1 Augmented Reality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Presence in Augmented Reality . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Virtual humans in Augmented Reality . . . . . . . . . . . . . . . . . . 13
2.4 Presence and virtual humans . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Perception of visual realism in Augmented Reality . . . . . . . . . . . 15
2.6 Uncanny Valley . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3 Methodology 21
3.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Expected Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4 Virtual humans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5 Prestudy of virtual humans . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Participants and Procedure . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Results 39
4.1 Collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Perception of realism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Presence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4 Uncanny Valley . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
xiii
4.5 Open Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5 Discussion 69
5.1 Collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 Perceptions of Realism . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3 Presence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4 Uncanny Valley . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
6 Conclusion 75
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
List of Figures 79
List of Tables 81
Glossary 83
Acronyms 85
Bibliography 87
CHAPTER 1
Introduction
”The main concept of Augmented Reality (AR) is to overlay spatially aligned virtual
objects onto the real world. Ideally, the user perceives that the virtual and real objects
coexist in the same space” [Azu97] .
Seamless integration of virtual objects into the real world is often a goal of AR application.
Ideally, virtual objects should integrate into the real world so that the user feels like
the virtual objects are ”here”. The sense of something being ”here” is also known as
the concept of presence. To achieve the highest possible presence in an AR application,
the level of realism could be an important factor. The effect of visual realism on the
perception of presence in AR must be studied empirically.
1.1 Motivation
AR is a field that greatly depends on human perception. The usability of the final
application, as well as the convenience for users, is based on how the virtual objects are
interpreted and perceived by the users. Without knowing how users perceive virtual
objects it is very difficult to develop AR application with the goal of a high level of
presence. How can be a high level of presence in an AR application achieved if influencing
factors are not known? Therefore, studying how AR affects human perception plays an
important role in future AR application development.
The main goal of this thesis was to develop an AR rendering system with varying levels
of realism and to use it to study the effect of increasing realism on the perception of
presence and on the convenience of users in AR. The amount of realism was mapped
to the presence score and user convenience to investigate the final dependence curve.
During the study, different observations have been made. For example, the existence of
an Uncanny Valle effect was studied.
1
1. Introduction
Clarification how the level of realism affects the sense of presence can help future presence
demanding AR applications to pay attention to the influencing factors. The observation
of the Uncanny Valley effect can help developers to understand critical factors which
could make the user disturbed of an AR application.
This is especially important if virtual humans are used to interact with an user. The
user should concentrate on AR application and the content it tries to provide and should
not be distracted from visual representation or laking sense of presence.
1.2 Approach
To study the effect of increasing realism in an AR application on the perception of presence
and on the convenience of users, an experiment was designed. For the experiment, an
AR application was implemented. During the experiment, the participant observes a
virtual human (in this context, also known as a virtual agent). The AR application
included a rendering system which allows the level of realism of the virtual agent to be
set. The feeling of presence should be as high as possible by design. Therefore, and if
allowed by technical possibilities, the implementation followed several findings in previous
studies. The conducted experiment followed a ”within-group design”. Each participant
experienced a virtual human with all different levels of realism in a random order. To
measure the experiences, each participant filled out a questionnaire after each observation.
The questionnaire followed and adapted questionnaires used in previous studies. It covers
questions about collision, perception, presence, co-presence and social presence. They are
also designed and adapted to measure a possible occurrence of the Uncanny Valley effect.
1.3 Expected Results
Recent publications in the research field of AR show tendencies that visual realism is
important to achieve a higher sense of presence. It is expected that visual realism is an
important factor for the sense of presence and convenience of an user. Also, the Uncanny
Valley, which was not measured in a targeted manner in the past, is expected to be
observed. Nevertheless, it is expected that technical limitations could rise problem to
observe the Uncanny Valley and overall perception.
1.4 Structure
This thesis consists of six chapters. Chapter 1 ”Introduction” highlights the main idea,
motivation and the approach of this thesis. Chapter 2 ”Background and Related Work”
summarise the theoretical background of the field of research in AR, virtual humans,
perception, presence and Uncanny Valley. Chapter 3 ”Methodology” describes the
approach to study the effect of visual realism on the sense of presence and convenience of
the user. Mainly, it contains the hypotheses, expected results, the experimental design
of the conducted user study, the approach to create a virtual human with different
2
1.4. Structure
visual realism levels and the approach to analyse the measured data set. Chapter 4
”Results” contains the results of the made data analyzes following by its discussion in
Chapter 5 ”Discussion”. Finally Chapter 6 ”Conclusion” summarises the overall findings
and highlights important aspects for future work in this research field.
3

CHAPTER 2
Background and Related Work
2.1 Augmented Reality
”AR is able to bridge the gap between real and virtual objects.” [MS16]
Since the beginning of Augmented Reality (AR) in the 1960s [ABF+01], research in the
field of AR has been growing [ABF+01]. In 1997 Azuma published a survey [Azu97]
that ”defined the field, described many problems, and summarized the developments
up to that point.” Azuma complimented the paper in 2001 [ABF+01] ”denoting the
rapid technological advancements” since 1997. Especially regarding display technology,
interface, and interaction, visualization problems as well as new applications in AR.
A complete chapter about the History of AR until 2015 can be found in the survey
of Billinghurst et al. [BCL15]. This survey summarizes almost 50 years of research
and development in the field of AR. It provides an overview of common definitions,
recaps taxonomies of other related technologies, and shows how AR fits into them. The
aforementioned papers together provide an extensive introduction to the research field of
AR.
This section gives a short overview of the field of AR including definitions, classifications,
technical requirements, types of AR systems, available software libraries as well as
potential limitations and open problems.
”. . . (AR) aims to create the illusion that virtual images are seamlessly blended with the
real world” [BCL15].
The quote above indicates that AR is about registering real world objects and combining
them with virtual objects to the user’s view. This must be done in real-time as a user
looks at the real-world. These characteristics were stated by Azuma [Azu97] and were
revisited in 2001 [ABF+01]. In summary, the main properties which define AR as a
system are:
5
2. Background and Related Work
1. AR combines real and virtual objects in a real environment
2. AR runs interactively, and in real-time; and
3. AR registers (aligns) real and virtual objects with each other.
This definition “does not place any limitations on the type of technology used, nor is it
specific to visual experience” [BCL15]. Virtual objects can also be an audio or haptic
experience [BCL15].
Over time, different approaches to define and classify AR in relation of Virtual Reality
(VR) have been published. Azuma et al. [ABF+01] summarized the differences between
VR and AR as such: with VR “the surrounding environment is virtual, while in AR
the surrounding environment is real,” and also described AR as a variation of Virtual
Environment (VE). Other than in AR, VR completely fades out the real world and puts
the user inside an entirely simulated or virtual environment. Azuma [Azu97] described
this as to “completely immerse a user inside a synthetic environment” [Azu97]. In AR,
the real world can be seen as the basic layer for any other virtual layer. Therefore, the
real environment is constantly scanned, analyzed, and interpreted to superimpose upon
or composite virtual objects within it. Ideally, this can be done in such a realistic way
that a user thinks that the real objects coexist in the same space.
Looking at the different definitions and classifications, Rekimoto and Nagao [RN95]
define AR in the context of making the computer interface invisible, seeing AR as a
complement to VR. Milgram and Kishino [MK94] define AR ”in the context of other
technologies” [BCL15], establishing a concept of ”Mixed Reality” (Figure 2.1). Mann
[ABRK13, Man02] extends the Milgram MR continuum by establishing a concept of
Mediated Reality. The metaverse roadmap from Smart et al. [SCP07], which is based
on Neal Stephenson’s concept of the Metaverse [Ste94], established another approach
to classify AR experiences by putting AR into context with the Mediality Continuum
and Virtuality Continuum. Other classifications that “provide an alternative perspective
to characterize AR experience” [BCL15] are, for example, Hugues concept based on
functional purpose [HFN11], as well as Braz and Pereria’s TARCAST taxonomy [BP08].
When it comes to specifying technological requirements of AR systems, Billinghurst
et al. [BCL15] summarized that Azuma [Azu97, ABF+01] provided a useful definition.
This thesis follows this summary. Table 2.1 shows a mapping of Azuma’s properties to
technical requirements described by Billinghurst et al. [BCL15].
Figure 2.1: Milgram’s Mixed Reality continuum [MK94]
Various types of display technologies are available, as well as applications, to combine
and composite representations of virtual objects with the real-world scenes. Azuma
6
2.1. Augmented Reality
Properties of AR as System Technical requirements
Combines real and virtual objects in a real
environment
A Display is needed that combines real
objects with virtual objects
Runs interactively, and in real-time A computer system that can generate in-
teractive graphics and is able to react to
users input in real-time
Registers (aligns) real and virtual objects
with each other
A tracking system that tracks the users’
viewpoints, registers real objects, and com-
bines them with virtual objects.
Table 2.1: Mapping properties of AR to technical requirements [BCL15]
characterized AR system displays based on optical and video-based, see-through Head
Mounted Display (HMD) technologies. See-through HMDs are devices that ”let the user
see the real world, with virtual objects superimposed by optical or video technologies”
[Azu97]. In contrast, closed-view HMDs ”do not allow any direct view of the real world”
[Azu97]. Video see-through systems capture the real world via a camera and present the
combined output to the user via a monitor. An Optical see-through system places the
combiner (i.e., a monitor) in front of the user’s eye. Figure 2.2 illustrates the concept
of an optical and video see-through system. Azuma also described a monitor-based
configuration with (i.e. Head-Up Displays) and without optical see-through (i.e., user
sits in front of a monitor). In 1997, Azuma did not categorize the displays based on
the location of the monitor or combiner. This was added in the complimentary paper
from Azuma et al. in 2001 [ABF+01], in which the authors classify displays into the
categories ”head-worn”, ”handheld”, and ”projective”. Head-worn displays (HMD) are
placed in front of the user’s eyes and are differentiated into optical see-through and video
see-through systems. Handheld displays are flat-panel, video see-through AR systems.
Projection display systems use projectors to overlay virtual information directly onto the
real/physical world.
Figure 2.2: Concepts of optical see-through (left) and video see-through displays (right).
Adapted from [ABF+01]
Billinghurst et al. [BCL15] took a detailed look into the approaches ”Video-based AR
Displays”, ”Optical See-through Displays,” ”Projection-based AR Displays,” and ”Eye
Multiplexed AR Displays.” AR displays are categorized based on where the display
is placed between the user’s eye and the real world to be consumed, including ”Head-
7
2. Background and Related Work
attached Displays,” ”Handheld and Body-attached Displays,” ”Spatial Displays,” and
also ”Other Sensory Displays.” Discussing these approaches in detail would be beyond
the scope of this thesis. The most important classification in the context of this thesis is
the distinction between optical see-through and video see-through, as well as head-worn,
handheld, and projective [BCL15].
Numerous libraries were developed in past that support the developer (or creator) with
tools to design and develop AR applications. Billingshurst et al. [BCL15] discussed
different development tools and arranged a couple of libraries and tools in “a hierarchy of
decreasing programming skills required to use them.” Table 2.2 illustrates the arranged
hierarchy, including required skills and example tools or libraries.
Type of Tool Skill required Example
Low-level software li-
brary/framework
Strong programming/coding
ability
ARToolkit, osgART,
Studierstube, MXRToolKit
Rapid prototyping tools Some programming ability,
but design/prototyping skills
FLARManager, Processing,
OpenFrameworks
Plug-in for existing de-
veloper tool
Skill with the developer tool
that the plug-in works with
DART, AR-Media plug-ins,
Vuforia and Metaio Unity
plug-ins
Stand-alone AR author-
ing tools
No programming ability, but
can learn stand-alone tool
BuildAR, Metaio Creator,
Layar Creator, Wikitude
Studio
Table 2.2: Hierarchy of AR development tools from most complex to least complex after
Billinghurst et al. [BCL15]
A tracking system determines ”the pose (position and orientation) of the viewer with
respect to an ’anchor’ in the real world” [BCL15]. This feature is crucial to registering
real and virtual objects with each other. The most common real world ”anchors” are
paper image markers, a GPS position, or even automatically-created markers based on
the registered environment. The term ”tracking” follows the understanding of Billinghurst
et al., that describes it as ”the process of registering the system in 3D,” [BCL15] which
may include one or two phases, depending on the underlying technology.
1. Registration phase: Determining “the pose of the viewer with respect to the real
world anchor” [BCL15]
2. Tracking phase: Updating “the pose of the viewer relative to a previously known
pose” [BCL15]
Billingshurst et al. described some of the most common tracking techniques in their
survey, like Visual Tracking, Inertial Tracking, GPS Tracking, Magnetic Tracking, and
8
2.2. Presence in Augmented Reality
Hybrid Tracking. The details of these tracking techniques can be found in the fourth
chapter, “AR Tracking Technology,” of the Billingshurst et al. survey [BCL15].
All systems have strengths and weaknesses, which are often based on technological limi-
tations. Besides technical limitations, other problem areas are user interface limitations
and social acceptance [ABF+01]. Also, the lack of a standard language to describe AR
content is a problem that limits the integration of AR content into existing tools. The
Augmented Reality Markup Language (ARML) is still not widespread [MS16]. Comparing
the problems noted in the papers from Azuma, from 1997 [Azu97] and 2001 [ABF+01],
and Manuri et al. [MS16], from 2016, it is noticeable that the concern for privacy, misuse
of data, and social acceptance increased over time. A detailed dive into current and
future problems, as well as challenges, can be found in [Azu97, ABF+01, BCL15, MS16].
Some AR applications require high realism of virtual objects, which should look like the
real objects themselves. The realism level of virtual objects greatly depends on rendering
quality. Important settings to achieve high-quality renderings are discussed in the next
sections.
2.2 Presence in Augmented Reality
Presence occurs when an individual overlooks or even forgets the use of technology during
a mediated experience. In relation to the concept of presence, the terms experience
and perception play an important role. This section explains the concept of presence as
related to Augmented Reality (AR) and how the concept of presence is different in AR
compared to Virtual Reality (VR).
The International Society for Presence Research provides an ”Explication Statement”
[II00] of the concept of presence. In the statement, definitions of the term presence, as
well as related terms, can be found. The following quote explains the term presence as a
technologically-created experience in which an individual does not recognize technology
to some degree; the term experience as an individual’s observation with objects; and the
term perception as interpretations of experience.
”Presence (a shortened version of the term ’telepresence’) is a psychological state or
subjective perception in which even though part or all of an individual’s current experi-
ence is generated by and/or filtered through human-made technology, part or all of the
individual’s perception fails to accurately acknowledge the role of the technology in the
experience. Except in the most extreme cases, the individual can indicate correctly that
s/he is using the technology, but at some level and to some degree, her/his perceptions
overlook that knowledge and objects, events, entities, and environments are perceived as
if the technology was not involved in the experience. Experience is defined as a person’s
observation of and/or interaction with objects, entities, and/or events in her/his environ-
ment; perception, the result of perceiving, is defined as a meaningful interpretation of
experience.” [II00]
9
2. Background and Related Work
Lombard and Ditton, examine the key concept of presence in their research [LD97]. They
argue that presence is a subjective property of a person, but also results from medium
and user characteristics. The authors provide a unifying definition of presence, which
can be applied to any medium and which also covers all six conceptualizations. The six
conceptualizations of presence are “Presence as social richness,” “Presence as realism,”
“Presence as transportation,” “Presence as social actor within medium,” and “Presence
as medium as social actor.” The full definition reads as follows:
“A medium that becomes invisible and produces a perceptual illusion of nonmediation
analogous to an open window can provide rich verbal and nonverbal information for social
interaction (presence as social richness); objects and entities in such a medium should
appear perceptually (if not socially) vivid and real (presence as realism); the illusion that
there is no medium at work means there is no border between “this side” and “the other
side” of the medium, so users can perceive that they have moved to the other side, that
objects/entities from the other side have entered their immediate environment, or that they
and other users are sharing a real or artificial environment (presence as transportation);
the illusion of nonmediation will be more complete if the medium is perceptually and
psychologically immersive (presence as immersion); and if we encounter people or entities
within such a medium, even if there is no possibility of true social interaction with them,
we are encouraged to respond to social cues they provide just as we would in nonmediated
communication (presence as social actor within medium). Finally, when the medium
itself presents us with social cues normally reserved for human-human interaction we are
likely to perceive it not as a medium but as an independent social entity, a transformed
medium (presence as medium as social actor).” [LD97]
In conjunction with visual realism, it is important to look at the conceptualization of
Presence as realism. It concerns the accurate representation of objects, events, and people
that should sounds and/or feels like the real object or as it would be expected. It can
be distinguished between social realism and perceptual realism. Perceptual realism is
increasing if the representations look as expected. This is important to note because it
gives a hint that visual realism is important for presence.
In conjunction with AR, it is important to look at the conceptualization of Presence as
transportation, especially types such as “You are there” and “It is here.” As Regenbrecht
et al. [Reg02] argued, while in VR, presence is more about the type “You are there,” in
which the user is transported to another place; in AR, the type “It is here” is much more
appropriate. The type “It is here” is about bringing a virtual object to the users’ real
environment.
Lombard et al. [LD97] also discuss cause and effect on presence. Therefore, they
identified variables that influence presence within mediated systems or applications.
Causes of presence are distinguished between “Causes of Presence as Invisible Medium”
and ”Causes of Presence as Transformed Medium.” Each is divided into form variables,
content variables, and media user variables. In conjunction with visual realism, the
content variables of “Causes of Presence as Invisible Medium” are important. The content
variables include the representation of “objects, human and nonhuman characters and
10
2.2. Presence in Augmented Reality
personae, task and activities, messages, stories, etc.” [LD97] that can influence presence
positively or negatively.
However, the influence of visual realism on presence is still not clear as can be seen by
different research results. This may be because of the different understanding of the
term “presence” in VR and AR. Slater [Sla09] and Blascovich [Bla02] defined presence in
the sense of ’being there.’ As mentioned above, Regenbrecht et al. [Reg02] argued that
presence in VR is about “You are there,” while in AR, it is about “It is here.” Sugano
et al. [SKT03] also describe the contrast of Presence in AR and VR. Presence in AR
“refers to how indistinguishable a virtual object seems from a similar real object played
in the same environment,” [SKT03] rather than in VR, where the term refers to “how
immersed a user feels in a virtual space.” They also point out presence as an important
metric to evaluate AR systems. Therefore, previous findings regarding the influence of
presence in VR should be reevaluated.
Vinayagamoorthy et al. [VB04] studied the consistency of visual realism between a virtual
environment and characters by testing their hypothesis that less repetitive textures of
the virtual environment and high fidelity of a virtual character lead to higher reported
presence. However, the result showed a different behavior and only partly supported
their hypothesis. The combination of the repetitive texture of the environment and high
fidelity of the visual appearance of the character showed the lowest reported presence.
One approach to explain that result was a discussion of possible occurrences of the
Uncanny Valley. The Uncanny Valley effect is discussed in Section 2.6. Visual realism of
characters is discussed in Section 2.5. Schuemie et al. [SvdSKvdM01] provide a useful
overview of presence research in VR. Zimmons et al. [ZPP03] determined no significant
influence of lighting and surface detail on presence in VR. Hendrix et al. [HB96], stated
that research studying presence in VR found a positive influence of higher visual realism.
Slater et al. “found greater behavioral and reported presence for static shadow compared
to no shadows and for dynamic shadows compared to static shadows,” [SUC95]. Mania
et al. found “no difference in reported presence among the three conditions,” [MR04].
In 2009, Slater et al. [SKMY09] concluded that there is “no clear message regarding
the impact of the visual realism level on reported or behavioral presence” [SKMY09].
With the goal of clarification, Salter et al. [SKMY09] set up an experiment to determine
the influence of recursive real-time ray tracing and ray casting and found presence to
be significantly higher for ray tracing. However, because it was not clear if a higher
illumination quality also influenced the reported presence, Slater et al. revisited this
open question in 2012 [YMK+12] and found that illumination quality does not have
an influence on presence, and so confirmed that only dynamic changes to shadows and
reflections influence presence in VR.
In AR, the level of realism seems to have an important influence on presence. Sugano
et al. [SKT03] found that shape and realistic shadow representations increase presence,
even if the shadow is not photo-realistic. They also found that the position of the light
source is not crucially important. Studying the effect of visual realism on search tasks in
an MR simulation, Lee et al. [LRM+13] included a presence questionnaire within their
11
2. Background and Related Work
experiment and could not find “any significant differences between the levels of realism”
[LRM+13]. Kán et al. [KDB+14] studied the influence of illumination model on presence
and found that presence is higher with global illumination than with direct illumination.
They found “significant correlation between the perception of realism and presence”
[KDB+14]. Kim and Welch [KW15] focused on finding characteristics of augmented
reality humans (AH) to maintain, enhance, or even destroy (also known as “break in
presence”) presence. In 2017, Kim et al. found that spatial and behavioral coherence
with (real) physical objects is important for increasing presence [KMB+17]. A more
detailed discussion about virtual humans in AR can be found in Section 2.3.
To study the effect of visual realism on presence, it is important to know the influencing
factors on users’ perception of virtual object realism. Those factors are discussed in
detail in Section 2.5. Effects of presence are distinguished between physiological and
psychological effects [LD97].
To measure presence in AR, Regenbrecht et al. [Reg02] followed the psychological
approach to developing a questionnaire. The goal of the questionnaire was to measure
the impression of how well a virtual object is integrated into the real world, from a
psychological point of view. The questionnaire, “Mixed Reality Experience Questionnaire,”
from Regenbrecht et al. [Reg13, RBBS17], provides a list of items that can be used to
measure Mixed Reality (MR) experiences. It provides a solid basis to measure presence
in AR and can be adapted and extended. The questionnaire from Regenbrecht was used
to measure presence in AR by Kán et al. [KDB+14], as well as Ling et al. [LNB+13].
Social presence and co-presence To measure the effectiveness of realistic virtual
humans several measurements are available. The two most common are social presence
and co-presence. Social presence describes the sense of being socially connected with
other present humans (virtual or real persons). Copresence describes the sense of another
person being present and being together [KW15, KMB+17, KBW17]. No universal
agreement on definitions of term social presence and co-presence could be found. Harms
and Biocca [HB04] categorized copresence as dimension of social presence.
“Social presence is a mutual interaction with a perceived entity refers to the degree of initial
awareness, allocated attention, the capacity for both content and affective comprehension,
and the capacity for both affective and behavioral interdependence with said entity.” [HB04]
“Copresence is the degree to which the observer believes he/she is not alone and secluded,
their level of peripheral of focal awareness of the other, and their sense of the degree to
which the other, and their sense of the degree to which the other is peripherally or focally
aware of them.” [HB04]
Blascovich et al. [Bla02] highlighted the typology of presence dimensions (social presence,
personal presence, environment presence) of Heeter [Hee92] and defined social presence
as the “degree to which one believes that he or she is in the presence of, and interaction
with, other veritable human beings.” [Hee92] Garau et al. [GSPR05] do not distinguish
between copresence and social presence.
12
2.3. Virtual humans in Augmented Reality
2.3 Virtual humans in Augmented Reality
“Virtual humans are used as interfaces as well as real-time augmentations (three-dimensional
computer-generated superimpositions) in real environments, as experienced by users
through specialized equipment for enhanced mobility (e.g., ultra-mobile PCs and video
see-through glasses).” [MTPM08]
Virtual humans in AR have been used in different research fields. This section provides
an overview of research which has been done with virtual humans.
Virtual humans can be distinguished between the virtual agent and the virtual avatar. A
virtual agent is a computer controlled virtual human. A virtual avatar is a virtual human
which is controlled by a real human [GSPR05].
Holz et al. [HDO09] surveyed different agent systems for social interaction along Milgram’s
Reality-Virtuality Continuum. In the survey, they discussed advantages and issues which
came along with social interaction with virtual humans. Obaid et al. [ONP11] studied
the perception of spatial relations and the perception of coexistence with virtual agents.
They conducted an experiment in AR and VR that measured the participants voice
compensation according to distances during interacting with a virtual agent. The authors
compared measurements taken in AR and VR and found that the compensation effect was
higher in AR. Kim et al. [KMB+17] combined these results with results from Loomis and
Knapp [LK03] and assumed that this might be because participants perceive the distance
between themselves and the AR agent with less perceptual errors. In another study,
Obaid et al. [ODK+12] investigated physiological response of users from different cultural
background during the interaction with a virtual agent. They found higher physiological
arousal if the virtual agent has not the same cultural background. Specifically, they
investigated interpersonal distance and eye-gaze and found that culturally appropriate
eye gaze has a higher arousal impact than interpersonal distance.
Several applications which include virtual humans were developed. The survey of
Magnenat-Thalmann et al. [MTPM08] summarized applications of interactive virtual
humans in mobile AR. Moreover, they discussed how different fields could benefit from a
virtual human interaction (i.e., in training scenarios). A check playing virtual human has
been introduced by Torre et al. [TFB+00]. SpaceTime, an AR telepresence framework
developed by Jo et al. [JKK15], significantly improved teleconference experience and
communications performance by adapting the position and motion of a virtual human
to the real environment in which it was placed. Room2Room, developed by Pejsa
[PKB+16], is a life-size telepresence system using digital projectors. It projects the remote
participants into the local participants’ environment without the need of participant
wearing special equipment.
2.4 Presence and virtual humans
The effect of virtual humans on real humans has been studied in VR and AR. Important
issues in this research field are the influence on presence, copresence, social presence
13
2. Background and Related Work
and how people react and respond to virtual humans. This section highlights important
studies which have been made in this field.
Visual realism and interaction influence presence. Vinayagamoorthy et al. [VB04] studied
presence responses across variations of visual realism in VR. They used virtual characters
and measured the response of users within a VR application. Garau et al. [GSPR05]
investigated the degree of virtual agents being treated as social entities rather than virtual
objects within VR. They measured presence, copresence, heart rate and electrodermal
activities (EDA) and found that users interact with virtual human more, even if only
little interaction mechanisms were implemented. At no time users perceived the virtual
human as real people. Yee et al. [YBR07] examined different studies to analyze if a
virtual human improves users’ performance. Despite the previous studies showing a
positive effect of interface agents on task performance, they found that the actual effect is
small. They stated that a visual representation of a virtual agent leads to more positive
social interactions than no visual representation, but also that the realism of the virtual
agent appears to matter very little.
Virtual humans socially influence the user in different ways. The degree of presence is
different when the virtual human is perceived as agent or as an avatar. Guadagno et al.
[GBBM07] studied social influence and persuasion in conjunction with virtual humans.
Basically, they found that higher behavior realism of a virtual human leads to higher
social presence. Also interesting is the finding that if participant believed that the virtual
human is an avatar the behavioral realism and social presence were higher rated. Fox et
al. [FAJ+15] examined the model of social influence in virtual environments of Blascovich
[Bla02] and found that ”avatars produced stronger responses than perceived agents.”
Kim and Welch [KW15] studied human-surrogate presence in AR and how perception of
real human and associated social influence are connected. They understood the term
human perception as how human users feel or interpret the human surrogate. Kim and
Welch [KW15] also concluded that in the inability of augmented human to physically
influence or control real objects in the real environment could be critical to establish or
maintain the level of perceived realism and the sense of social presence. Therefore, they
highlighted approaches to address this issue including a non-technical approach to the
context dependent behavior of virtual humans [KW15].
Virtual humans should not conflict with the surrounding physical world. Kim et al.
[KMB+17] used a virtual human to examine how the level of integration into the real-
world impacts presence. They presented an experiment in which the virtual human
did react to real physical objects. During the study the virtual human even asked the
participant to move a chair to get around it, even if a virtual human could go through
it. They concluded that “it is beneficial to have the virtual human’s natural occlusion
and proactive behavior (..) for higher social/copresence.” [KMB+17] The findings show
that to achieve a realistic virtual human in AR it is crucial to implement a virtual
human such that it reacts to physical objects in the environment [KMB+17]. Also,
Microsoft Developer guidelines for HoloLens Spatial Mapping [Mic17] highlights the
need for conflict-free relationship and interactions between virtual objects and the real
14
2.5. Perception of visual realism in Augmented Reality
environment.
Studying the influence of visual realism on presence depends on technological possibilities.
Guadagno et al. [GBBM07] pointed out that available technology is not capable of
creating highly realistic virtual humans.
In conjunction with virtual realism of characters, visual realism and behavioral realism
should be considered combined. Higher visual realism in character may lead to a higher
expectation of behavioral realism [TBS+98], [VB04]. This is also how Nowak and Biocca
[NB03] explained their VR experiment results that a higher anthropomorphism actually
leads to a lower sense of presence, copresence and social presence.
2.5 Perception of visual realism in Augmented Reality
“Perception, the recognition and interpretation of sensory stimuli, is a complex construct.”
[Cut03, KSF10].
AR mixes real-world objects with virtual objects. Therefore, it is often a goal to seamlessly
integrate virtual objects into the real world. One approach to reach this goal is to achieve
the highest possible visual realism of virtual objects. A virtual object’s rating as realistic
depends on the perception of the viewer. Which factors influence the perception of visual
realism is an important research topic in the field of VR and AR. This section determines
the influencing factors on users’ perceptions of virtual objects’ visual realism in AR.
2.5.1 Perception and perceptual Issues
“Each sensory modality provides a different kind of information on which we base our
interpretations and decisions.” [KSF10]
Modalities interact with each other and strongly influence how humans perceive the
world around them. Measuring these interactions is challenging. One approach is to
match obtained environmental cues. Sometimes, cues override or conflict with each other.
Conflicting cues often result in perceptually-incorrect augmentations [KSF10].
Perceptual issues are problems that arise while observing and interpreting information
from virtual objects. These issues can arise from a combination of virtual and real object
information, but also from the representations of real objects themselves. [KSF10]
Kruijff et al. [KSF10] classified perceptual issues in AR within the context of visual
processing and interpretation pipeline (what they called perceptual pipeline). They
organized issues into five classifications: environment, capturing, augmentation, display,
and individual user differences. Moreover, they identified problems that affect perception
of augmented content and stated that incorrect depth interpretation is the most common
perceptual problem in AR applications. Kruijff et al. [KSF10] presented a variety of
different approaches to address these issues. Among other things, they pointed out that it
is not clear how much rendering quality or fidelity is needed to address perceptual issues.
15
2. Background and Related Work
This is based on a study of Thompson et al. [TWG+04], who found no direct relationship
between the level of fidelity and the judgment of depth in digitally-reproduced graphics.
Besides rendering quality, Kruijff et al. [KSF10] stated that illumination can affect
fidelity of virtually augmented objects and their correct perception. Antialiasing can
improve fidelity but may lead to perceptual distortions [KSF10]. Differences in rendering
quality and antialiasing could lead to false stereoscopic disparity. Drascic and Milgram
[DM96] also found this effect if differences between the resolutions of the captured video
background and the rendered objects existed. Klein and Murray [KM08] found the
influence of color schema on perception. Regarding illumination, Kruijff et al. [KSF10]
pointed out that “correct illumination can also help to make the scenery more believable
preventing augmented objects from looking like cardboard mock-ups.”
2.5.2 Influencing factors
If a virtual object should look as realistic as possible (often referred to as photorealistic),
usually the question arises what are the influencing factors. This section determines the
influencing factors to create a realistic virtual object.
In computer graphics visual realism is defined as one of three levels of realism [Fer03]:
1. Physical realism: virtual objects provide the same visual simulation as the scene
2. Photo-realism: virtual objects provide the same visual response as the scene
3. Functional realism: virtual objects provide the same visual information as the scene
The questions of what does photo-realism mean in a perceptual context has not yet
been fully answered [KDT+11]. An experiment of Hattenberg et al. [HFJS09] tried to
find which algorithm creates the most realistic image. The results showed that indirect
illumination and noisier images were rated as more realistic, but the authors also pointed
out that these results cannot be generalized and there are other important influencing
factors (on the perception of a scene). Elhelw et al. [ENC+08] tracked eye movement
and found light reflections/specular highlights, 3D surface details, and depth visibilities
to be important image features.
To achieve visual realism, various factors must be considered. In VE (Virtual Environment)-
based application. Roussou et al. [DRÝ+03] found different factors to achieve an
immersive impression:
• Consistent lighting
• Use of shadow
• The realness factor of the virtual experiences
• Virtual people
16
2.5. Perception of visual realism in Augmented Reality
Haller [Hal04] stated that some other factors like shading, shadow integration and texture
effects have to be considered in an AR based application. Moreover, following the three
varieties of realism of Ferwerda [Fer03] we further focus on the visual perception. This
does not mean that behavior realism is not important. Durand [Dur02] concluded that a
virtual world should be more convincing and believable, rather than realistic.
Photogrammetric 3D reconstruction can be used to increase visual realism and with
that human-likeliness. Latoschik et al. [LRG+17] studied the effect of avatar realism in
VR on embodiment and social interactions and found that higher visual realism leads
to higher human-like rating and a stronger feeling of body ownership. They created
photogrammetry 3D scanned avatars and were able to show that ”current 3D scan
technology is capable of increasing the human-likeliness of avatars in immersive social
Virtual Realities.” [LRG+17]
Agusanto et al. [ALZN03] studied photorealistic rendering for AR using environment
illumination, by using image-based lighting techniques and environment illumination
maps. They were able to create a consistent and coherent virtual world reflecting the
real world. To achieve a convincing AR application, Bimber et al. [BGWK03] created a
consistent lighting situation between virtual and real objects. Key factors in their study
were to illuminate captured reflection information of real objects under new lighting
conditions and to approximately match direct and indirect lighting effects (such as shading,
shadows, reflections, and color bleeding). Haller [Hal04] created an AR framework that
uses photorealistic, as well as non-photorealistic, rendering techniques. Key factors of
the framework were to consider shading, shadows, and bump mapping.
Shading is important if a realistic virtual representation is required. Flat shaded objects
generally do not look realistic. Dynamic shading give a more realistic look than constant
shading. Shades depend on the light source. The exactly correct placement of the light
source does not have to be 100% accurate but it needs to approximately mimic the
direction of the light in the real world. [Hal04, LRM+13]
“Shadows are essential to improve the visual perception.” [Hal04]
Shadows “can dramatically improve the realism” [Hal04], and are essential to create a
3D impression and make the 3D perception of virtual objects easier. [NNMH02] Sugano
et al. [SKT03] highlighted the importance of shadows in an AR scenario. They studied
the effect of a shadow representation of virtual objects in AR and focused on providing
accurate shadows. They raised the question, “How different levels of optical consistency
can affect a user’s perception of virtual object realism?” [SKT03]. Regarding perception
of visual realism, they found that shadows of virtual objects have a positive effect on
spatial cues recognition, meaning that shadows are helpful “for recognizing position
relationships in depth direction” [SKT03]. Haller [Hal04] also believes that shadows are
equally important in a 3D stereo HMD (Head Mounted Display) because it helps to sense
the distance between virtual objects more precisely. Shadows improve intractability of
virtual objects [McC00]. Kruijff et al. [KSF10] also pointed out that “shadows can be
helpful, providing an important depth cue.” Similar to shading and shadowing geometry,
17
2. Background and Related Work
time, and illumination should be consistent within AR applications [NNMH02]. Bump
mapping simulates bumps or wrinkles on a surface and help to improve realism, without
complex geometry [Hal04].
Knecht et al. [KDT+11] created “a framework for studying photorealistic rendering
techniques in AR to investigate perceptual issues and visual cues.” Their implementation
included a rendering system in which different rendering modes and styles can be set
up. They divided photorealistic image features into two main categories: Visual cues
and augmentation style. Visual cues include inter-object spatial cues and depth cues.
Augmentation style includes illumination, color and contrast, tone-mapping, and camera
artifacts. They tested their framework by performing a user study testing three rendering
conditions. The first one was without any cast shadow or indirect illumination. The
second included shadows between real and virtual objects, but no indirect illumination.
The third one had inter-object shadowing and indirect illumination. They found no
significant effect from rendering conditions on task performance.
Lee et al. [LRM+13] studied the effects of visual realism on search tasks in a Mixed Reality
(MR) simulation. In their experiment, they performed an AR search task experiment and
repeated it within VR. For the VR application, the real-world location, where the AR
experiment took place, was modeled three times. Each model had a different perceived
level of visual realism. Regarding the goal of this section, it is interesting to look at
their approach to varying visual realism. They used the same models and varied only a
few factors of visual realism. They argued, “If we are careful not to inter-mix different
data formats and rendering techniques or change their levels in opposite directions, there
should be no interaction that could cause unanticipated responses in visual realism. If
we only increase the factors, we argue that this produces progressively and objectively
higher levels of visual realism.” The chosen data format was a polygon-based model,
and three factors of visual realism were varied: geometry, texture, and lighting. They
increased geometry realism by increasing polygon counts. Texture realism was increased
from no texture to simple textures, and up to high-resolution textures. Lighting realism
was increased from simple flat shading to baked-in realistic lighting.
2.6 Uncanny Valley
To the best of our our knowledge, there has been no study that studied the Uncanny
Valley effect specifically in AR. This is most likely due to technological boundaries and the
still ongoing research about visual realism, perception, and presence in AR. This section
explains the Uncanny Valley and presents related studies that mention the possible
appearance or influence of the Uncanny Valley.
The theory of the Uncanny Valley was proposed in 1970 by Mori [MMK12], who studied
the effect of human-like robot artifacts on humans. Ho et al. [HM10] describe the Uncanny
Valley as “a hypothetical graph describing a nonlinear relation between a character’s
degree of human likeness and the emotional response of the human perceiver” [HM10].
Meaning that the sense of affinity increases with the human likeness of a non-human
18
2.6. Uncanny Valley
representation (i.e., an artificial hand), but falls rapidly into a valley if the representation
hits around 80-100% of human likeness [MMK12]. In other words, if a human cannot
be sure if it is real or not, the human can suddenly feel a strong refusal or even disgust.
Figure 2.3 shows a graph which depicts the Uncanny Valley.
Figure 2.3: Uncanny Valley graph [MMK12]
.
The Uncanny Valley effect could also occur in virtual environments and with virtual
humans. Yee et al. [YBR07] raise the question of whether virtual agents are actually
useful in improving user performance. The study confirmed this hypothesis but also
pointed out that a realistic representation of a virtual agent has little influence on the
performance. Nevertheless, the study also pointed out that the Uncanny Valley may have
influenced these findings. The influence could not be proven and should be considered
in future studies [YBR07]. Although Guadagno et al. [GBBM07] did not study the
Uncanny Valley and could not investigate it empirically, they also pointed out that
it may occur if a mismatch between the appearance and behavior of a virtual human
appears. To measure the Uncanny Valley some studies used the Godspeed questionnaire.
The Godspeed questionnaire originally describes a series of questions to measure the
humans’ perception of robots [BKCZ09]. The included concepts, anthropomorphism and
likability, can serve as the x- and y-axes of Moris’s graph (Figure 2.3) [BKIH09]. After
empirical testing Ho et al. [HM10] rejected the Godspeed questionnaire as unsuitable to
measure the Uncanny Valley effect and developed a new set of Uncanny Valley indices.
The Uncanny Valley might affect presence. Vinayagamoorthy et al. [VB04] speculated
19
2. Background and Related Work
while discussing results of their VR user study that ”characters being notably different
in representation than the environment they inhabit, might lead the participant to think
the world ”inconsistent and thus might lead the participant to feel less present” [VB04],
and describes this speculation as an extension of the Uncanny Valley hypothesis. Kim
and Welch [KW15] mentioned that it could be argued that the Uncanny Valley ”could
be related to the failure of the sense of non-mediation or contextual inconsistencies”
[KW15]. Lugrin et al. [LLL15] studied the impact of virtual realism on the illusion of
body ownership (IVBO) in VR. The authors suspected the Uncanny Valley effect which
could cause that the machine-like avatars generated lower IVBO than human-like avatars.
20
CHAPTER 3
Methodology
To study the effect of increasing realism on the perception of presence and the convenience
of users in an AR application an experiment was conducted. The following chapter
describes the experiment and the AR application which was implemented for this purpose
in Subsection 3.3. Hypotheses (also see Subsection 3.1) were tested during the experiment.
The discussion about the results and the comparison to the expected results (Subsection
3.2) can be found in Section 3.7.
3.1 Hypotheses
Two hypotheses are posed to examine the effect of realism on the perception of presence
and the convenience of users within AR applications.
H1: Increasing the level of realism increases the sense of presence and convenience of
users.
H2: The Uncanny Valley effect can be observed within the experiment.
3.2 Expected Results
Related work to the field of presence and AR lead to the expectation that the results
will support the hypotheses (Subsection 3.1). Also, the Uncanny Valley, which was not
measured in a targeted manner in the past, is expected to be observed. Studies in the
field of AR often faced limitations due to technical possibilities of the used AR system.
Especially the HoloLens, which is the AR system to be used in this study, often shows
technical limitations. The limited field of view is expected to be the hardest challenge
for presence study.
21
3. Methodology
3.3 Experimental Design
The conducted experiment followed a within-group design. A within-group design was
chosen because interpersonal differences can affect results in social presence studies as
Kim et al. [KMB+17] argued. The virtual human was placed at a specific place in a room.
An idle animation was used to give the virtual human a more human-like behaviour
while standing. Additionally, the virtual human would also use a turn animation to align
its body in the direction of the participant. The AR application included a rendering
system which allows the levels of realism of the virtual human to be set. Each participant
experienced a virtual human with all different levels of realism in a random order.
After each observation the participant filled in a questionnaire. The goal was to measure
the perception of realism, presence, co-presence, social presence and the occurrence of
the Uncanny Valley effect. Therefore, a questionnaire was constructed using specific
questions designed for the particular purpose. Additionally, an open question was added
which can be answered with own words. The rest of the questions were ordinal questions
answered on 7-point Likert type items ranging from 1 (strongly disagree) to 7 (strongly
agree). Besides these questions, the participants were asked about their gender and age.
3.3.1 Perception of realism
To measure the level of perceived realism one question, used in the presence study of
Kán et al. [KDB+14] (based on [Reg02, Reg13, RBBS17]), was adapted. The question
used to measure perception of realism can be found in Table 3.1
PR-1 Overall, how would you rate the degree of realness achieved by the virtual
human; to what extent ”they seemed real to you.”
Table 3.1: Perception of realism questionnaire
3.3.2 Presence
To measure the overall sense of presence, the questionnaire followed the approach of a
mixed reality (MR) presence questionnaire designed by Regenbrecht et al. [Reg02, Reg13,
RBBS17]. To compare the virtual human with a real human, a question from [KDB+14]
was adapted. The presence questions can be found in Table 3.2
22
3.3. Experimental Design
P-1 Watching the virtual human (further ”guide”) was just as natural as watching
the real world.
P-2 I felt I could have touched the guide.
P-3 The guide was part of the same space.
P-4 The guide and surrounding walls, furniture, doors, etc. were in the same environ-
ment.
P-5 Overall, how would you rate the sense of presence generated by the guide; to what
extent “she/he was here”?
Table 3.2: Presence questionnaire
3.3.3 Co-presence
To measure co-presence, the questionnaire followed the approach of Kim et al. [KMB+17].
Their questions were based on questions used by Bailenson et al. [BBBL03, BHSS00].
They also added two own of their questions. The question used to measure co-presence
can be found in Table 3.3.
CP-1 I perceived that I was in the presence of the guide in the room with me.
Table 3.3: Co-presence questionnaire
3.3.4 Social presence
To measure social presence, the questionnaire followed the approach of Kim et al.
[KMB+17]. Their questions are based on the social presence questionnaire from Bailenson
et al. [BBBL03]. The questions used to measure social presence can be found in Table 3.4.
For the question number 3 "The thought that the guide is not a real person crosses
my mind often." is a negative question. This question has to be inverted during result
analysis.
SP-1 I felt the guide was watching me and was aware of my presence.
SP-2 The thought that the guide is not a real person crosses my mind often.
SP-3 I felt like I want to talk to the guide or asked him something
Table 3.4: Social presence questionnaire
3.3.5 Uncanny Valley
To identify the occurrence of the Uncanny Valley effect, scales, designed by Ho and
MacDorman [HM10, HM17], were used. They created indices (further dimensions) to
measure the appearance of the Uncanny Valley. These dimensions are Humanness,
Eeriness, and Attractiveness. Each of these dimensions originally included multiple scales.
The originally created scales can be found in Table 3.5. To prevent biased answers based
23
3. Methodology
on possible tiredness of the participant (after answering several questions before) we
decided to limit the possible scales. Therefore, each participant was only asked about
three dimensions. For each dimension a question about the dimension itself was created.
Further, the participants were able to answer the question using the already used scale
1 (strongly disagree) and 7 (strongly agree). The questions used to measure Uncanny
Valley can be found in Table 3.6. The questions number 2 "The guide seems eerie/creepy
to me." and 3 "The guide created spine-tingling moments." are negative questions. These
questions have to be inverted during result analysis.
# Dimension Scale (-3 - +3)
1 Humanness Inanimate/Lifeless – Living/Alive
2 Humanness Synthetic/Simulated – Real
3 Humanness Mechanical movement – Biological movement
4 Humanness Human-made – Humanlike
5 Humanness Without definite lifespan/Immortal – Mortal
6 Eeriness(Eerie) Dull/Boring – Freaky/Cool
7 Eeriness(Eerie) Predictable – Eerie/Unpredictable
8 Eeriness(Eerie) Plain/Simple – Weird
9 Eeriness(Eerie) Ordinary – Supernatural
10 Eeriness(Spine-tingling) Boring – Shocking
11 Eeriness(Spine-tingling) Uninspiring – Spine-tingling/Fascinating
12 Eeriness(Spine-tingling) Predictable – Thrilling
13 Eeriness(Spine-tingling) Bland/Boring – Uncanny
14 Eeriness(Spine-tingling) Unemotional – Hair-raising
15 Attractiveness Ugly – Beautiful
16 Attractiveness Repulsive/Disgusting – Agreeable/Enjoyable
17 Attractiveness Crude/Raw/Uncut – Stylish
18 Attractiveness Messy – Sleek/Decent
Table 3.5: Original Uncanny Valley scales adapted from Ho and MacDorman [HM10,
HM17]
# Dimension Question
UV-1 Humanness The guide seems to be human/alive/living.
UV-2 Eeriness(Eerie) The guide seems eerie/creepy to me.
UV-3 Eeriness(Spine-tingling) The guide created spine-tingling moments.
UV-4 Attractiveness The guide was attractive/good looking/handsome.
Table 3.6: Uncanny Valley questions used in the experiment
24
3.3. Experimental Design
3.3.6 Open Question
To collect possible other influencing factors of participants’ perception of the experiment
an optional open question was added. If preferred, the participants were allowed to
answer in keywords. The used open question can be found in Table 3.7.
OQ-
1
How would you describe the overall perception of the experiment? You can answer
in keywords (of preferred).
Table 3.7: Open question of participants’ perception
3.3.7 Questionnaire
All questions were put together and participants were asked to answer one questionnaire
for each realism level. The complete questionnaire can be found in the Table 3.8. The
questionnaire was put into an online form and the participants were asked to answer the
questions electronically using a laptop.
25
3. Methodology
Perception of realism
PR-1 Overall, how would you rate the degree of realness achieved by the virtual human;
to what extent ”they seemed real to you.”
Presence
P-1 Watching the virtual human (further ”guide”) was just as natural as watching
the real world.
P-2 I felt I could have touched the guide.
P-3 The guide was part of the same space.
P-4 The guide and surrounding walls, furniture, doors, etc. were in the same environ-
ment.
P-5 Overall, how would you rate the sense of presence generated by the guide; to what
extent “she/he was here”?
Co-Presence
CP-1 I perceived that I was in the presence of the guide in the room with me.
Social Presence
SP-1 I felt the guide was watching me and was aware of my presence.
SP-2 The thought that the guide is not a real person crosses my mind often.
SP-3 I felt like I want to talk to the guide or asked him something
Uncanny Valley
UV-1 The guide seems to be human/alive/living.
UV-2 The guide seems eerie/creepy to me.
UV-3 The guide created spine-tingling moments.
UV-4 The guide was attractive/good looking/handsome.
Open question
OQ-1 How would you describe the overall perception of the experiment? You can answer
in keywords (of preferred).
Table 3.8: Questionnaire used in the user study
3.3.8 Experimental Setup
The experiment was conducted in a room with about 6-meter width and 6-meter depth.
The setup consists of a Microsoft HoloLens which is an optical see-through Head Mounted
Display (HMD) system (also see Section 2.1) with a field of view of 30◦x17.5◦. The
HoloLens is a fully closed-loop system which does not need external tracking systems.
The HoloLens contains an inertial measurement unit (IMU) including an accelerometer,
gyroscope, and a magnetometer, four sensors for environmental scanning, a depth
camera with a 120◦x120◦ angle of view, a 2.4-megapixel photographic video camera, four
microphones, and a light sensor. Besides the Intel Cherry Trail SoC which contains the
CPU and GPU the HoloLens also includes a custom Microsoft Holographic Processing
Unit (HPU) which is a coprocessor. The Windows 10 operating system runs on the 64GB
eMMC controlled by the SoC. SoC (1GB LPDDR3) and HPU (1GB LPDDR3) sahre a
8MB SRAM. The internal battery lasts about 2-3 hours (active using). Although the
26
3.4. Virtual humans
hardware is very potent for an HMD of its size, it is still a mobile device which suffers
from potential technical limitations. These have to be kept in mind if an AR application
is developed for the HoloLens. Unity1 was used to develop the AR application for the
experiment.
3.4 Virtual humans
3.4.1 Model
The goal was to provide a virtual human with different levels of realism. The sense of
presence should be as high as possible by design. Therefore, the implementation followed
several findings from previous studies. Sadly, due to the technical limitations (also see
Subsection 6.1.1) of the used AR system HoloLens global illumination or shadows could
not be used. To reduce the risk that different models could influence the results, the main
structure of the particular virtual humans stayed the same for each virtual realism level
except for the lowest realism level. Only a few factors were chosen (see Table 3.9) to be
changed for each realism level. Changes of these factors were only made in one direction.
In other words, increasing the level of visual realism would increase all factors. The
geometries varied in different fidelity settings. The goal was to increase the visual realism
only by increasing the chosen factors. This should minimize the risk of unanticipated
responses in visual realism. This approach followed the approach of varying the level of
realism of Lee et al. [LRM+13]. Table 3.10 shows the different visual realism levels and
the factor settings. Figure 3.1 shows examples of the virtual avatar for each realism level.
To create the virtual humans, the software Fuse from Adobe 2 was used. The Adobe
general terms of use 3 allow the free use of every created character and animation in every
embedded media. The generated character and animation were imported into Unity.
Geometry
Texture
Lighting
Shade and Shadow
Illumination
Table 3.9: Visual realism factors
1https://unity3d.com
2https://www.adobe.com/products/fuse.html
3https://www.adobe.com/legal/terms.html
27
3. Methodology
Geometry Texture Lights, Illumination and Shadows
Level 1 100% 1024x1024 5 light sources; direct illumination; no shadows
due to technical limitations
Level 2 60% 128x128 4 light sources; direct illumination; no shadows
due to technical limitations
Level 3 40% 64x64 3 light sources; direct illumination; no shadows
due to technical limitations
Level 4 20% To texture;
only plain
color
2 light sources; direct illumination; no shadows
due to technical limitations
Level 5 Block avatar To texture;
only plain
color
1 light sources; direct illumination; no shadows
due to technical limitations
Table 3.10: Visual realism levels and factors
28
3.4. Virtual humans
(a) Level 1 (b) Level 2 (c) Level 3
(d) Level 4 (e) Level 5
Figure 3.1: Used virtual human at different visual realism levels
3.4.2 Geometry
Geometry was varied by polygon counts. Starting with the highest possible polygon
settings, they were reduced accordingly. Figure 3.2 and Figure 3.3 shows examples of a
virtual human with different counts of polygons at the different levels.
29
3. Methodology
(a) 32323 Polygons (b) 12928 Polygons (c) 6464 Polygons
Figure 3.2: Examples of polygons of a virtual human at different levels
(a) Polygons of a virtual hu-
mans hand
(b) Polygons of a virtual hu-
mans hand after 80% polygon
reduction
Figure 3.3: Example of polygons of a virtual humans hand
3.4.3 Texture
The texture was varied by resolution. We started with the highest possible resolution
setting and reduced it accordingly. Figure 3.4 shows examples of a virtual human with
varying resolutions of texture at the different realism levels.
(a) Texture resolu-
tion 1024x1024
(b) Texture resolu-
tion 128x128
(c) Texture resolu-
tion 64x64
(d) No texture
Figure 3.4: Examples of a virtual humans with different texture resolutions
30
3.4. Virtual humans
3.4.4 Lights
Light sources are necessary to make the virtual humans visible. Light sources at the
highest realism level were placed based on real light conditions in the actual room where
the experiment took place. Therefore, five light sources were placed with different colors
and angles. Table 3.11 shows the different lights with their color, angle setting as well as
in which level they were visible. Light number 3,4 and 5 were light spots in the area of
the virtual human, which is why the all have the same color and angle settings. Figure 3.5
shows an example view of the light sources color and angle.
Light number Light color [HEX] Light angle (Euler rotation) Level visibility
1 DDCACAFF (100,-125,0) Level 1
2 C7C179FF (30,-80,0) Level 1-2
3 FFFFFFFF (25,-160,0) Level 1-3
4 FFFFFFFF (25,-160,0) Level 1-4
5 FFFFFFFF (25,-160,0) Level 1-5
Table 3.11: Light sources
Figure 3.5: Light sources
.
3.4.5 Illumination
Only direct illumination was taken into account. While Kán et al. [KDB+14] found a
significant positive impact of global illumination on the sense of presence in AR environ-
ments, unfortunately, due to the limited technical capabilities (also see Subsection 6.1.1)
of the used AR system global illumination could not be used.
3.4.6 Shadow
Due to technical limitations (also see Subsection 6.1.1) and the resulting risk to negatively
influence the perception of the participant it was decided not to use shadows. Nevertheless,
it should be mentioned here that the occurrence of shadows significantly influence the
realistic appearance of a visual object and it is important for the perception of AR
scenes. [KDB+14] It helps users to better understand the placement of a virtual object
in the real world. The visual realism of the shadows depends strongly on its resolution.
31
3. Methodology
Unfortunately, the HoloLens uses light-based displays. These displays add virtual objects
into the field of view of the user by adding light to the real-world, but it cannot reduce
light. This means that a traditional dark shadow would be transparent to the user’s eye.
If the virtual object would be a static object, there are some workarounds which can
be used. The first possibility is to place a plane right under the virtual object and the
shadows would be shown on the plane. The second possibility is to implement a so-called
’negative shadow’. The goal of a negative shadow is to add a glow to the floor around the
virtual object. The negative shadow would appear darker than its surrounding. Microsoft
describes the possibility of using a negative shadow in their documentation as well:
”Many designers have found that they can even more believably integrate holograms by
creating a "negative shadow" on the surface that the hologram is sitting on. They do this
by creating a soft glow on the ground around the hologram and then subtracting the
"shadow" from the glow. The soft glow integrates with the light from the real world and
the shadow grounds the hologram in the environment.” 4.
Another problem which occurs if using shadows with the HoloLens is the limited per-
formance which results in a bad resolution of the shadow. The virtual human uses idle
animation and also turns to the user’s direction. In the context of shadows, this would
mean that a shadow should dynamically change accordantly to the idle animations and
further has to be calculated in real-time. These real-time calculations are performance
expensive. This results in a limited resolution of the shadows. It was decided not to use
shadow since it appeared that the resolutions of dynamic shadows was to low and would
most likely influence the perception of the user.
3.4.7 Animations
Since users expect a realistic behavior of photorealistic rendered virtual objects [Hal04],
all used animations are animations which have been recorded with a real human using
motion capture systems. Table 3.12 shows the used animations during the user study.
The animations were obtained from the online service Mixamo Adobe5. The Adobe
general terms of use 6 allow the free use of every created character and animation in every
embedded media. The generated character and animation were imported into Unity.
4https://docs.microsoft.com/en-us/windows/mixed-reality/hologram
5https://www.mixamo.com/
6https://www.adobe.com/legal/terms.html
32
3.5. Prestudy of virtual humans
Name Description
Idle Used while virtual human does not do any action.
Turn right 1 Character turns 45◦ towards user
Turn right 2 Character turns 90◦ towards user
Turn right 3 Character turns 135◦ towards user
Turn right 4 Character turns 180◦ towards user
Turn left 1 Character turns -135◦ towards user
Turn left 2 Character turns -90◦ towards user
Turn left 3 Character turns -45◦ towards user
Table 3.12: Animations used in user study
3.4.8 Occlusion
Due to the experimental design, it was decided that it is not necessary to implement
occlusion features. Nevertheless, it is important to know that Kim et al. [KMB+17]
showed that virtual humans should be implemented concerning the physical objects in
the environment to achieve a higher sense of presence.
3.5 Prestudy of virtual humans
The approach to create different realism levels of a virtual human and order was desinged
by author of this thesis. To avoid a possible biased view on the realism level a prestudy
was conducted. To goal of the prestudy was to confirm the designed order.
3.5.1 Method
The main aim of the prestudy was to confirm the used approach to create different
realistic levels of a virtual human. A web application was implemented in which the user
initially see pictures of the different avatars in random order. The task was to order the
avatars according to the users’ perception of realism from the most realistic to the least
realistic one.
3.5.2 Hypothesis
The prestudy hypothesized that the different visual realism levels created in Subsection 3.4
and further the desinged order (from lowest to highest realism level) listed in Table 3.10
agrees with the perception of real persons (i.e. participants perceive the realism levels
increasing in the same order as designed by the author).
3.5.3 Expected result
It was expected that the submitted order of realism levels indicated by the participants
matches with the visual realism levels created in Subsection 3.4 and further listed in
33
3. Methodology
Table 3.10. Figure 3.6 shows the expected order.
Figure 3.6: Expected order of visual realism levels
.
3.5.4 Experimental Design
Participants were asked to order the pictures of the virtual humans according to the
realistic levels from the most realistic to the least realistic. Using buttons the user was
asked to move the pictures up or down / left to right (depending on the layout). Finally,
the user could submit his answers. The design of the web application can be seen on
Figure 3.7.
Figure 3.7: Prestudy web application
.
3.5.5 Results
20 individual persons participated in our prestudy. 19 ordered the pictures in the expected
order. One person submitted precisely the other way round as expected. We assumed
that the person misunderstood the task.
34
3.6. Participants and Procedure
3.5.6 Conclusion
The result confirmed the expected order and so the different realism levels of virtual
humans created by us. The expected order of realism can further be used during the
experiment.
3.6 Participants and Procedure
Thirty-one people aged between 19 to 71 participated in our main presence study, with
an average mean of M = 33.29, SD = 11.61. The group of participants consisted out
of 12 women and 19 men. All participants were informed about the study and signed a
consent to participate in the research study. After fitting the HMD on the head of the
participant, she was asked to stand at an initial position in the room. Right next to the
starting point virtual button could be found. The participants could start the experiment
by interacting with the button. The virtual human was placed at a certain place in the
room. The participant could walk around to observe the avatar from different distances
and angles. She was asked to observe the virtual human for at least two minutes. After
the participant felt ready, she could walk to a table to fill out the questionnaire. After
filling out the questionnaire, the was asked to go back to the initial starting point and
push/click the virtual button. These steps were repeated five times (once for each realism
level). Each participant experienced all five different realism levels in a random order
(within-group design).
3.7 Data Analysis
All in all 15 questions were recorded. These can be broken down into four categories:
Collision, Perception of realism, Presence and Uncanny Valley. Each category is statisti-
cally analyzed depending on the underlying data. In each category, the realism levels
(1-5) can be understood as sample groups or conditions. The within-group design of
the experiment, in which participants are the same in each sample group, is decisive to
choose the correct statistical test. [Fie17]
3.7.1 Collision
The first question the participants could answer by a single choice. They could choose
one of three options. These options are further transformed into nominal data 1, 2 and 3.
Table 3.13 shows the mapping of the single choice answer and the transformed number.
35
3. Methodology
On the way to the table... Transformed number
... you crossed/went through the virtual human 1
... you went around the avatar to avoid a collision 2
... you did not use the direct path and went around
the corner to avoid a collision
3
Table 3.13: Collision answer mapping into nominal data
To analyze the behavior of the participants the frequencies of the answers for each realism
level were calculated. The mode was used to calculate the central tendency. Due to the
within-group design of the experiment, it can be assumed that the answers may be biased
after a participant answered the question about the collision in the first condition. To
analyze this circumstance the first conditions where evaluated separately by analyzing
the answers after the first observation of each participant.
3.7.2 Perception of realism
One question measured the level of perceived realism. It was a question answered
on a 7-point Liker type item ranging from 1 (strongly disagree) to 7 (strongly agree).
Ordinal data limits the possibilities of statistical tests. To analyze this question the
frequency of answers were investigated. To analyze if differences across the realism
levels are significantly different the Friedman test was used. The ANOVA tests only
provide information about the differences between the different conditions but does not
provide any information which specific differences are in fact significant. [Fie17] Wilcoxon
signed-rank tests were used to examine the specific differences between the conditions.
[Fie17] The decision of using non-parametric tests was made because the data violated
the assumption of normal distribution.
3.7.3 Presence
The presence questionnaire consists out of nine questions to measure presence. Cronbach’s
alpha was utilized to analyze the internal consistency of the questions. All questions were
answered on a 7-point Liker scale ranging from 1 (strongly disagree) to 7 (strongly agree).
The presence questionnaire can be broken down into three categories: General presence,
Co-Presence and Social-Presence. Each category consits out of one or more questions.
Five questions about general presence, one about co-presence and three about social
presence. Question number two of the category Social presence "The thought that the
guide is not a real person crosses my mind often." was inverted from 1-7 to 7-1 because
of the negative nature of this question (also see Subsection 3.3)
The questions were merged using the mean for analyzing the overall representation of
presence. The repeated-measures ANOVA test was used to analyze if the means are
significantly different across the realism levels. In case a repeated-measures ANOVA
showed significant differences the parametric paired-sample t-test was used to verify this
36
3.7. Data Analysis
trend. The parametric tests could be used because the means between the conditions
were normally distributed.
3.7.4 Uncanny Valley
Four questions measured the effect of the Uncanny Valley. Cronbach’s alpha was used to
analyze internal consistency of the questions for each realism level. These were ordinal
questions answered on a 7-point Liker type item ranging from 1 (strongly disagree) to
7 (strongly agree). The questions used to measure Uncanny Valley can be found in
Table 3.6. The questions number 2 "The guide seems eerie / creepy to me." and 3 "The
guide created spine-tingling moments." were inverted by recoding from 1-7 it to 7-1
because of the negative nature of this question. Further the median of these question
were calculated, because it is a subgroup one group Eerieness (also see Section 3.3).
The questions were merged using the mean for analyzing the overall representation of
presence. The repeated-measures ANOVA test was used to analyze if the means are
significantly different across the realism levels. In case a repeated-measures ANOVA
showed significant differences the parametric paired-sample t-test was used to verify this
trend. The parametric tests could be used because the means between the conditions
were normally distributed.
3.7.5 Open question
The last question was an open question. The participants were allowed to answer in
keywords or not at all (optional). To analyze the answers the quantity how many
actually answered were measured. Further, for each level the keywords were analyzed
and categorized based on the notion.
37

CHAPTER 4
Results
The following chapter shows the specific results of the statistical examination of the
experiment.
4.1 Collision
For each realism condition participants were asked if they avoided a collision with the
avatar. This section evaluates this collision question and shows the results for each
realism level from 1 (highest realism) to 5 (lowest realism). The results of a frequency
analysis can be found in Table 4.1. Additionally, Figure 4.1 shows the histograms for
each realism levels.
Level 1
15 participants out of 31 went through the virtual human. 14 participants went straight
to the table but tried to avoid a collision. 2 participants did not go the shortest way to
the questionnaire and went around a corner to avoid a collision. The calculated mode
value was 1 (collision).
Level 2
11 participants out of 31 went through the virtual human. 15 participants went straight
to the table but tried to avoid a collision. 5 participants did not go the shortest way to
the questionnaire and went around a corner to avoid a collision. The calculated mode
value was 2 (no collision).
Level 3
17 participants out of 31 went through the virtual human. 11 participants went straight
to the table but tried to avoid a collision. 3 participants did not go the shortest way to
the questionnaire and went around a corner to avoid a collision. The calculated mode
value was 1 (collision).
Level 4
15 participants out of 31 went through the virtual human. 10 participants went straight
39
4. Results
to the table but tried to avoid a collision. 6 participants did not go the shortest way to
the questionnaire and went around a corner to avoid a collision. The calculated mode
value was 1 (collision).
Level 5
15 participants out of 31 went through the virtual human. 13 participants went straight
to the table but tried to avoid a collision. 3 participants did not go the shortest way to
the questionnaire and went around a corner to avoid a collision. The calculated mode
value was 1 (collision).
1 (Collision) 2 (No collision) 3 (Detour) Mode
Level 1 15 14 2 1
Level 2 11 15 5 2
Level 3 17 11 3 1
Level 4 15 10 6 1
Level 5 15 13 3 1
Table 4.1: Collision - Contingency table
40
4.1. Collision
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.1: Collision histograms
The answers of the first rounds condition can be found in the Table 4.2. Additionally,
Figure 4.2 shows the histograms for each realism levels.
Level 1
6 persons experienced level 1 during the first condition. 4 participants went through
the avatar. 2 participants went straight to the table but tried to avoid a collision. 0
participants did not go the shortest way to the questionnaire and went around a corner
to avoid a collision. The calculated mode value was 1 (collision).
Level 2
8 persons experienced level 2 during the first condition. 4 participants went through
the avatar. 2 participants went straight to the table but tried to avoid a collision. 2
participants did not go the shortest way to the questionnaire and went around a corner
41
4. Results
to avoid a collision. The calculated mode value was 1 (collision).
Level 3
5 persons experienced level 3 during the first condition. 4 participants went through
the avatar. 1 participants went straight to the table but tried to avoid a collision. 0
participants did not go the shortest way to the questionnaire and went around a corner
to avoid a collision. The calculated mode value was 1 (collision).
Level 4
2 persons experienced level 4 during the first condition. 0 participants went through
the avatar. 1 participants went straight to the table but tried to avoid a collision. 1
participants did not go the shortest way to the questionnaire and went around a corner
to avoid a collision. The calculation of the mode results in two possible values: 2 (no
collision) and 3 (detour).
Level 5
10 persons experienced level 5 during the first condition. 6 participants went through
the avatar. 3 participants went straight to the table but tried to avoid a collision. 1
participants did not go the shortest way to the questionnaire and went around a corner
to avoid a collision. The calculated mode value was 1 (collision).
No of par-
ticipants
1 (Collision) 2 (No collision) 3 (Detour) Mode
Level 1 6 4 2 0 1
Level 2 8 4 2 2 1
Level 3 5 4 1 0 1
Level 4 2 0 1 1 2 and 3
Level 5 10 6 3 1 1
Table 4.2: Collision in the first round - Contingency table
42
4.2. Perception of realism
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.2: First round collision histograms
4.2 Perception of realism
To measure perception of realism the question "Overall, how would you rate the degree
of realness achieved by the virtual human; to what extent ”they seemed real to you." was
used. Frequency analysis shows that the list of top one frequent answers for each realism
level mostly equals the expected order, but the frequencies and their plots showed that
the differences between level 2 and 3 are most likely not significant. Figure 4.3 shows a
histogram of the answers for each realism level. The contingency table (Table 4.3) shows
the frequency of the answers for each level. The top one frequency of each realism level
is highlighted as bold. The median values from 1 to 5 are M1 = 4, M2 = 3, M3 = 3, M4
= 2, M5 = 1. Figure 4.4 shows a plot of the median of the perceived realism over all
43
4. Results
realism levels.
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.3: Histograms of perceived realism
1 2 3 4 5 6 7
Level 1 2(6%) 4(12%) 7(22%) 4(12%) 10(32%) 3(9%) 1(3%)
Level 2 4(12%) 8(25%) 6(19%) 7(22%) 4(12%) 2(6%) 0(0%)
Level 3 3(9%) 11(35%) 8(25%) 6(19%) 1(3%) 2(6%) 0(0%)
Level 4 6(19%) 13(41%) 6(19%) 3(9%) 2(6%) 1(3%) 0(0%)
Level 5 20(64%) 3(9%) 4(12%) 3(9%) 1(3%) 0(0%) 0(0%)
Table 4.3: Contingency table - Perceived realism scores from 1-7 (horizontal) for each
realism level (vertical)
44
4.3. Presence
Figure 4.4: Median of perceived realism scores from 1-7 (vertical) for each realism level
1-5 (horizontal)
Friedman’s ANOVA showed that perception of realism of each level did significantly
change, χ2
F(4) = 47.259, p < 0.001. Wilcoxon tests were used to follow up this finding.
It appears that the perception of realism from level 2 to level 3 has not significantly
changed, T = 92.5, Z = -1.158, p = 0.247. All other differences were significant. The
perception of realism from level 1 to level 2 has significantly changed, T = 66, Z = -2.224
p = 0.026. From level 3 to level 4 it did significantly changed, T = 26, Z = -2.025, p =
0.043. From Level 4 to 5 perception of realism has significantly changed,T = 76, Z =
-2.374, p = 0.018.
4.3 Presence
As mentioned in Subsection 3.7.3 the presence questionnaire can be broken down into
three categories: General presence, Co-Presence and Social-Presence. The following
45
4. Results
section first analyzes the overall representation of presence by using the calculated mean
value. The Subsections 4.3.1, 4.3.2, and 4.3.3 further analyze the categories.
The questionnaire about presence consists out of nine questions. Cronbach’s alpha showed
a high internal consistency of the presence questionnaire. Level 1 α = 0.906, Level 2
α = 0.873, Level 3 α = 0.886, Level 4 α = 0.886, Level 5 α = 0.848. The means of
the answers of each participant were calculated to analyze the data. The histograms in
Figure 4.5 shows a wide range of answers in each level. Especially level 4 in Figure 4.5d
and level 5 in Figure 4.5e.
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.5: Presence histogram
The overall mean of each level, plotted in Figure 4.6, shows that presence decreases with
a lower realism level. Table 4.4 shows the median, mean, confidence interval and standard
46
4.3. Presence
deviation of each level.
Figure 4.6: Mean of presence across realism levels
Median Mean 95% confidence interval Standard deviation
Level 1 4.2222 3.9926 [3.514, 4.471] 0.234
Level 2 3.9444 3.7741 [3.341, 4.207] 0.212
Level 3 3.1667 3.4556 [2.978, 3.934] 0.234
Level 4 3.3889 3.3111 [2.859, 3.763] 0.221
Level 5 3.1667 3.2625 [2.796, 3.729] 0.228
Table 4.4: Mean and standard deviation of presence
The repeated-measures ANOVA showed that there were significant changes across the
means, F = 10.055, p = <0.01. The assumption of normal distribution of the differences
between the levels was fulfilled. The Shapiro-Wilk test for the differences from level 1
to level 2 returned a statistic value SW = 0.963, p = 0.345. Shapiro-Wilk of differences
between level 2 and 3, SW = 0.95, p = 0.16. Shapiro-Wilk of differences between level 3
and 4, SW = 0.974, p = 0.664. Shapiro-Wilk of differences between level 4 and 5, SW =
0.9465, p = 0.119. The assumption of sphericity has not been violated. The Mauchly
test shows no significant variance changes, p = 0.095. The estimators of sphericity,
Greenhouse-Geisser, ε = 0.825, and the Huynh-Feidt, ε = 0.944, showed a good fit. A
value of 1 of a sphericity estimator would be perfectly spherical. Paired-sample t-test was
used to follow up findings of the repeated-measures ANOVA test, which does not provide
47
4. Results
information which levels are actually significantly different. The difference of mean of
level 1 and 2 was not significant, t = 1.481, p = 0.149, BCa 95% CI[-0.08333, 0.52037],
with a difference of 0.21852. The difference of mean of level 2 and level 3 was significant,
t = 2.707 , p = 0.011, BCa 95% CI[0.007789, 0.555915], with a difference of 0.31825. The
difference of mean between level 3 and level 4 was not significant t = 1.485, p = 0.148,
BCa 95% CI[-0.05443, 0.34332], with a differences of 0.1444. The difference of mean
between level 4 and 5 was not significant, t = 0.371, p = 0.714, BCa95% CI[-0.21956,
0.31679], with a difference of 0.04861.
4.3.1 General Presence
The questionnaire about the general presence consists out of five questions. Cronbach’s
alpha showed a high internal consistency of the presence questionnaire. Level 1 α = 0.913,
Level 2 α = 0.890, Level 3 α = 0.913, Level 4 α = 0.923, Level 5 α = 0.876. The means
of the answers of each participant were calculated to analyze the data. The histograms
in Figure 4.7 shows a wide range of answers in each level.
48
4.3. Presence
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.7: General presence histogram
The overall mean of each level, plotted in Figure 4.8, shows that presence decreases with
a lower realism level. Table 4.5 shows the median, mean, confidence interval and standard
deviation of each level.
49
4. Results
Figure 4.8: Mean of general presence across realism levels
Median Mean 95% confidence interval Standard deviation
Level 1 4.2 3.9933 [3.47, 4.516] 1.40097
Level 2 3.6 3.7 [3.204, 4.196] 1.32847
Level 3 3.1 3.38 [2.839, 3.921] 1.44828
Level 4 3.2 3.3467 [2.796, 3.897] 1.47432
Level 5 3.4 3.1950 [2.656, 3.734] 1.4432
Table 4.5: Mean and standard deviation of general presence
The repeated-measures ANOVA showed that there were significant changes across the
means, F = 7.938, p = <0.01. The assumption of normal distribution of the differences
between the levels was fulfilled. The Shapiro-Wilk test for the differences from level 1
to level 2 returned a statistic value SW = 0.959, p = 0.276. Shapiro-Wilk of differences
between level 2 and 3, SW = 0.964, p = 0.383. Shapiro-Wilk of differences between level
3 and 4, SW = 0.892, p = 0.005. Because of the low p-value, a graphical analysis was
made. The graphical analysis showed that the statistical Shapiro-Wilk test is disturbed
by outliers and a high quantity of a specific difference. Figure 4.9 shows the histogram
of the differences including a normal distribution curve, a normal Q-Q Plot, and a
Box-Plot. Shapiro-Wilk of differences between level 4 and 5, SW = 0.931, p = 0.051.
The assumption of sphericity has not been violated. The Mauchly test shows significant
variance changes, p = 0.380. The estimators of sphericity, Greenhouse-Geisser, ε = 0.847,
and the Huynh-Feidt, ε = 0.972, showed a good fit. A value of 1 of a sphericity estimator
would be perfectly spherical. Paired-sample t-test was used to follow up findings of the
repeated-measures ANOVA test, which does not provide information which levels are
significantly different. The difference of mean of level 1 and 2 was not significant, t =
50
4.3. Presence
1.735, p = 0.093, BCa 95% CI[-0.05238, 0.63905], with a difference of 0.29333. The
difference of mean of level 2 and level 3 was significant, t = 2.183 , p = 0.037, BCa 95%
CI[0.02025, 0.61975], with a difference of 0.32. The difference of mean between level 3
and level 4 was not significant t = 0.257, p = 0.799, BCa 95% CI[-0.23204, 0.29870], with
a differences of 0.0333. The difference of mean between level 4 and 5 was not significant,
t = 0.99, p = 0.33, BCa95% CI[-0.16161, 0.46494], with a difference of 0.33.
51
4. Results
(a) Histogramm with normal distribution curve
(b) Normal Q-Q Plot
(c) Box-Plot
Figure 4.9: Graphical normal distribution analysis between differences between general
presence mean values of realism level 3 and 4
4.3.2 Co-Presence
One question was used to measure Co-Presence. The means of the answers of each
participant were calculated to analyze the data. The histograms in Figure 4.7 shows a
wide range of answers in each level.
52
4.3. Presence
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.10: Co-Presence histogram
The overall mean of each realism level plotted in Figure 4.11 shows that presence decreases
with a lower realism level, except for level 4 to 5. Table 4.6 shows the median, mean,
confidence interval and standard deviation of each level.
53
4. Results
Figure 4.11: Mean of Co-Presence across realism levels
Median Mean 95% confidence interval Standard deviation
Level 1 4.47 5.0 [3.80, 5.13] 1.776
Level 2 4.3 4.0 [366, 4.94] 1.725
Level 3 4.2 4.0 [3.51, 4.89] 1.846
Level 4 3.9 4.0 [3.14, 4.66] 2.023
Level 5 4.03 3.5 [3.25, 4.815] 2.092
Table 4.6: Mean and standard deviation of CoPresence
The repeated-measures ANOVA showed that there were no statistically significant changes
across the means, F = 2.076, p = 0.088. The assumption of normal distribution of the
differences between the levels was fulfilled. The Shapiro-Wilk test for the differences
from level 1 to level 2 returned a statistic value SW = 0.91, p = 0.015. Shapiro-Wilk of
differences between level 2 and 3, SW = 0.912, p = 0.017. Shapiro-Wilk of differences
between level 3 and 4, SW = 0.868, p = 0.002. Shapiro-Wilk of differences between level 4
and 5, SW = 0.77, p <0.001. Because of the low p-values, a graphical analysis was made.
The graphical analysis showed that the statistical Shapiro-Wilk test is disturbed by
outliers and a high quantity of specific differences. The graphical analysis in Figures 4.12,
4.13, 4.14, 4.15 show the histogram of the differences including a normal distribution curve,
a normal Q-Q Plot, and a Box-Plot for each difference. The assumption of sphericity has
not been violated. The Mauchly test shows significant variance changes, p = 0.238. The
estimators of sphericity, Greenhouse-Geisser, ε = 0.85, and the Huynh-Feidt, ε = 0.977,
showed a good fit. A value of 1 of a sphericity estimator would be perfectly spherical.
Paired-sample t-test was used to follow up findings of the repeated-measures ANOVA test,
which does not provide information which levels are actually not significantly different.
54
4.3. Presence
The difference of mean of level 1 and 2 was not significant, t = 0.841, p = 0.407, BCa 95%
CI[-0.239, 0.572], with a difference of 0.167. The difference of mean of level 2 and level 3
was not significant, t = 0.474, p = 0.639, BCa 95% CI[-0.331, 0.531], with a difference of
0.1. The difference of mean between level 3 and level 4 was not significant t = 1.725, p
= 0.095, BCa 95% CI[-0.056, 0.656], with a differences of 0.3. The difference of mean
between level 4 and 5 was not significant, t = 0.643, p = 0.526, BCa95% CI[-0.558, 0.291],
with a difference of -0.133.
55
4. Results
(a) Histogramm with normal distribution curve
(b) Normal Q-Q Plot
(c) Box-Plot
Figure 4.12: Graphical normal distribution analysis between differences between Co-
Presence mean values of realism level 1 and 2
56
4.3. Presence
(a) Histogramm with normal distribution curve
(b) Normal Q-Q Plot
(c) Box-Plot
Figure 4.13: Graphical normal distribution analysis between differences between Co-
Presence mean values of realism level 2 and 3
57
4. Results
(a) Histogramm with normal distribution curve
(b) Normal Q-Q Plot
(c) Box-Plot
Figure 4.14: Graphical normal distribution analysis between differences between Co-
Presence mean values of realism level 3 and 4
58
4.3. Presence
(a) Histogramm with normal distribution curve
(b) Normal Q-Q Plot
(c) Box-Plot
Figure 4.15: Graphical normal distribution analysis between differences between Co-
Presence mean values of realism level 4 and 5
4.3.3 Social Presence
The questionnaire about the social presence consists out of three questions. Cronbach’s
alpha showed a low internal consistency of the presence questionnaire. Level 1 α = 0.440,
Level 2 α = 0.387, Level 3 α = 0.311, Level 4 α = -0.011, Level 5 α = 0.013. Deleting the
second question SP-2 ”The thought that the guide is not a real person crosses my mind
59
4. Results
often.” results in higher Cronbach’s alpha values: Level 1 α = 0.713, Level 2 α = 0.653,
Level 3 α = 0.760, Level 4 α = 0.769, Level 5 α = 0.572. The means of the answers
of each participant were calculated to analyze the data. The histograms in Figure 4.16
shows a wide range of answers in each level.
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.16: Social presence histograms
The overall mean of each level plotted in Figure 4.17 shows that the sense of presence
decreases with a lower realism level, except for level 4 to 5. Table 4.7 shows the median,
mean, confidence interval and standard deviation of each level.
60
4.3. Presence
Figure 4.17: Mean of Social Presence across realism levels
Median Mean 95% confidence interval Standard deviation
Level 1 4.0 3.83 [3.367, 4.3] 1.15
Level 2 4.0 3.72 [3.274, 4.171] 1.20
Level 3 3.33 3.33 [2.853, 3.814] 1.28
Level 4 3.16 3.06 [2.674, 3.437] 1.02
Level 5 3.16 3.12 [2.668, 3.565] 1.20
Table 4.7: Mean and standard deviation of social presence
The repeated-measures ANOVA showed that there were significant changes across the
means, F = 7.013, p = <0.01. The assumption of normal distribution of the differences
between the levels was fulfilled. The Shapiro-Wilk test for the differences from level 1
to level 2 returned a statistic value SW = 0.965, p = 0.404. Shapiro-Wilk of differences
between level 2 and 3, SW = 0.946, p = 0.132. Shapiro-Wilk of differences between level
3 and 4, SW = 0.958, p = 0.272. The Shapiro-Wilk of differences between level 4 and
5, SW = 0.968, p = 0.497. The assumption of sphericity has not been violated. The
Mauchly test shows significant variance changes, p = 0.354. The estimators of sphericity,
Greenhouse-Geisser, ε = 0.836, and the Huynh-Feidt, ε = 0.959, showed a good fit. A
value of 1 of a sphericity estimator would be perfectly spherical. Paired-sample t-test was
used to follow up findings of the repeated-measures ANOVA test, which does not provide
information which levels are actually significantly different. The difference of mean of
level 1 and 2 was not significant, t = 0.691, p = 0.495, BCa 95% CI[-0.21793, 0.44015],
with a difference of 0.11. The difference of mean of level 2 and level 3 was significant, t =
2.183 , p = 0.026, BCa 95% CI[0.05064, 0.72714], with a difference of 0.39. The difference
of mean between level 3 and level 4 was not significant t = 1.664, p = 0.107, BCa 95%
61
4. Results
CI[-0.06362, 0.61917], with a differences of 0.28. The difference of mean between level 4
and 5 was not significant, t = 0.364, p = 0.718, BCa95% CI[-0.40412, 0.2819], with a
difference of 0.33.
4.4 Uncanny Valley
Cronbach’s alpha first showed very low internal consistency of the questionnaire. Level 1
α = 0.494, Level 2 α = 0.200, Level 3 α = 0.038, Level 4 α = 0.297, Level 5 α = 0.314.
Deleting the eeriness values result in higher Cronbach’s alpha values. Level 1 α = 0.700,
Level 2 α = 0.642, Level 3 α = 0.753, Level 4 α = 0.294, Level 5 α = 0.572.
The means of the answers of each participant were calculated to analyze the data. The
histograms in Figure 4.18 shows a wide range of answers in each level.
The overall mean of each level plotted in Figure 4.19 shows that presence decreases with
a lower realism, except a significant increase from level 4 to level 5. Table 4.4 shows the
median, mean, confidence interval and standard deviation of each level.
Figure 4.19: Means of Uncanny Valley scores across realism levels
Median Mean 95% confidence interval Standard deviation
Level 1 4.64 4.64 [4.228, 5.0553] 1.11
Level 2 4.0 4.11 [3.7336, 4.4831] 1.0
Level 3 3.5 3.49 [3.1587, 3.8246] 0.89
Level 4 3.25 3.18 [2.835, 3.5317] 0.93
Level 5 4.0 4.34 [4.0649, 4.6185] 0.74
Table 4.8: Mean and standard deviation of Uncanny Valley scores
62
4.4. Uncanny Valley
(a) Level 1 (b) Level 2
(c) Level 3 (d) Level 4
(e) Level 5
Figure 4.18: Uncanny Valley histogram
The repeated-measures ANOVA showed that there were significant changes across the
means, F = 13.792, p = <0.01. The assumption of normal distribution of the differences
between the levels was fulfilled. The Shapiro-Wilk test for the differences from level 1
to level 2 returned a statistic value SW = 0.949, p = 0.158. Shapiro-Wilk of differences
between level 2 and 3, SW = 0.961, p = 0.333. Shapiro-Wilk of differences between level
3 and 4, SW = 0.934, p = 0.062. Shapiro-Wilk of differences between level 4 and 5, SW
= 0.964, p = 0.383. The assumption of sphericity has not been violated. The Mauchly
test shows no significant variance changes, p = 0.083. The estimators of sphericity,
Greenhouse-Geisser, ε = 0.792, and the Huynh-Feidt, ε = 0.9, showed a good fit. A value
of 1 of a sphericity estimator would be perfectly spherical. Paired-sample t-test was used
to follow up findings of the repeated-measures ANOVA test, which does not provide
63
4. Results
information which levels are actually significantly different. The difference of mean of
level 1 and 2 was significant, t = 3.456, p = 0.002, BCa 95% CI[0.21773, 0.84894], with
a difference of 0.53333. The difference of mean of level 2 and level 3 was significant, t
= 2.873, p = 0.008, BCa 95% CI[0.17762, 1.05571], with a difference of 0.61667. The
difference of mean between level 3 and level 4 was not significant t = 1.406, p = 0.17, BCa
95% CI[-0,14016, 0.75683], with a differences of 0.30833. The difference of mean between
level 4 and 5 was significant, t = -4.897, p < 0.001, BCa95% CI[-1.64215, -0.67452], with
a difference of -1.15833.
4.5 Open Question
Participants were asked about the overall perception of the experiment using the question:
"How would you describe the overall perception of the experiment? You can answer
in keywords (of preferred)." The question was optional to answer. Is was answered 83
times of 155 (53.5%). Most of the participant answered the question using keywords. 16
participants answered the open question for level 1 (answers can be found in Table 4.9.
Most of the keywords mentioned are interaction or movement (4/16) and realistic
representation (4/16). 16 participants answered the open question for level 2 (answers
can be found in Table 4.10. Most of the keywords mentioned are about that the virtual
humans made the user unconformable (6/16) and about the realistic representation
(4/16). 18 participants answered the open question for level 3 (answers can be found in
Table 4.11. Most of the keywords mentioned are about that virtual humans made the user
unconformable (8/18) and about the unrealistic representation (6/18). 16 participants
answered the open question for level 4 (answers can be found in Table 4.12. Most of
the keywords mentioned are about the unrealistic representation (9/16) and about that
virtual humans made the user unconformable (5/16). 17 participants answered the open
question for level 5 (answers can be found in Table 4.13). Most of the keywords mentioned
are about the unrealistic or game-like representation (7/17), and the sympathetic or
funny feeling participants had (4/17).
64
4.5. Open Question
Level 1 involved more movement than level 2
Noticeable more interaction. Want to experiment with virtual human, i.e. I shook
my head and waited until the avatar does the same
had the feeling that it imitated my movements
realistic
Interesting
more realistic
the eyes have often reminded that he was not real
best representation so far; best experience; However, the avatar (as well as all the
others) seemed to shine, which unfortunately greatly reduces realism.
Most realistic representation. The more realistic the more sinister. Missing blink of
the avatar looks spurious. General: Near Plane is too far away for a realistic feeling.
I would like to get in contact more
sympathetic but without movement feedback
Above all, I felt strangely watched
Better than the level 3
Preppy dressed, thus competent charisma. nice facial features
no occlusion -> that makes it less present in the same room.
more realistic
Table 4.9: Answers to the open question - Level 1
eyes did not move or blink; proportion between shoes and body is not correct (shows
are too small)
he eyes have stared at me unpleasant
enjoyable
exciting
Maybe little more realistic, one realized that the object itself was strange
impressive
no difference to Level 3
Compared to the previous Avatar, the eyes were more conspicuous here
Avatar is clearly artificial. More interaction would improve the presence
unusual, impressive that a human is standing with me in the room
strange
surreal
Black background and thus less transparency have made the experience "real".
Same as Level 3
eerie
Hardly any interaction, slow response to my position
Table 4.10: Answers to the open question - Level 2
65
4. Results
Head is moving, this was not the case in previous levels
something scary avatar due to the eyes
normal
disappointed
you could see the pixels and notice that it is not real
first not sure if it the same, little bit more impressive than the last one
same as level 2?, except the eyes were strange, the shape a bit ’angular’ / ’edgy’
looks like previous level 2, but less quality
surrealistic, exciting, wanted to try how the avatar react
Low poly makes him very unrealistic
edgy
unpleasant
From a distance, not sure if is the same, it seems the avatar had no eyes. Something
scary
the starring was strange
Very distorted facial structures, eerie overall appearance
least interesting avatar
Eerie face
scary, because black face, if you are close
Table 4.11: Answers to the open question - Level 3
seemed a bit nervous
Due to the lack of eyes, a feeling of discomfort arose
very pleasant presence
strange
boring
not so realistic
the avatar did not have eyes
Less details on the avatar. Clearly artificial. Reminiscent of [Nitendo] Wii graphic.
edgy person
slippery
As if watching a mannequin.
Similar to other Avatars
I was not sure how I should behave during the study
Spooky, the lack of shadows hampers the credibility of the Avatar’s existence through
its low-detail appearance - I noticed more in Level 4 than in the others
Weird, emotionless face
because of closed eyes / missing pupils I do not feel looked at, clearly non-human (it
looks like the avatar should not represent a human being)
Table 4.12: Answers to the open question - Level 4
66
4.5. Open Question
Lego-avatar is looking towards me
computer generated figure, nice representation but nevertheless never perceived as
real
absolutely neutral
funny, playful
cheerfully
imperceptible
interesting life experience
Minecraft style. Seems like being part of a game.
very unrealistic
unreal character
unrealistic
Because the Avatar did not try to look like a real person, his presence was a lot more
believable.
surreal, avatar seems like a ghost
Nice, playful, sympathetic
looks wobbly, depicts people (as lego), disappointing that Level 5 does not close his
eyes when approaching
Very interesting, human movements depending on how I moved were impressive
unusual, new experience
Table 4.13: Answers of open question - Level 5
In Section 4.2 no significant differences of perceived realism between level 2 to level 3
could be found. The answers of the open question of participants which consumed these
realism level in this order is additionally discussed in the following. These participants
stated that they had problems to see the differences. 4 out of 31 participants consumed
level 2 and 3 (or vice versa) in this order during the experiment. Table 4.14 shows the
answers to the open question of these 4 participants.
67
4. Results
Level order Answer level 2 Answer level 3
1 3 -> 2 "Maybe little more realistic,
one realized that the object it-
self was strange"
"you could see the pixels and
notice that it is not real"
2 2 -> 3 "impressive" "first not sure if it the same,
little bit more impressive than
the last one"
3 2 -> 3 "Compared to the previous
Avatar, the eyes were more
conspicuous here"
"same as level 2?, except the
eyes were strange, the shape a
bit ’angular’ / ’edgy’"
4 2 -> 3 "Black background and thus
less transparency have made
the experience "real"
From a distance, not sure
if is the same, it seems the
avatar had no eyes. Something
scary."
Table 4.14: Answers of participants which consumed level 2 and level 3 in this order (or
vice versa)
68
CHAPTER 5
Discussion
The main goal of this thesis was to develop an AR rendering system with varying levels
of realism and to use it to study the effect of increasing realism on the perception of
presence and on the convenience of users in AR. Therefore, two hypotheses were created
and studied. The first Hypothesis "Increasing the level of realism increases the sense of
presence and convenience of users." was partially supported by the significant differences
in the results. Nevertheless, tendencies can be shown. The second hypothesis "The
Uncanny Valley effect can be observed within the experiment." was not supported. To
conduct the experiment an AR rendering system with varying levels of realism was
implemented and a questionnaire was developed (Subjection 3.3). In Subsection 3.4 the
visual realism factors for the different realism level were created. The results of the
experiment are discussed in the following chapter.
5.1 Collision
The results in Section 4.1 show that there are no real tendencies recognizable. The mode
values visibility does not have any correlation with the realism levels, which is why no
further statistical tests were carried out. It is assumed that technical limitations (also
see Subsection 6.1.1) of the used AR headset and especially the limited field of view
prevented the participants from seeing the whole virtual avatar at once. It could be that
the user looked at the floor and so lost sight of the virtual human when heading to the
questionnaire. Nevertheless, some participants unconsciously went around the virtual
human without even noticing it. These observations give a hint that a virtual human
somehow was perceived as ’real’ spatial object.
69
5. Discussion
5.2 Perceptions of Realism
For the majority, perceived realism levels match with the predetermined realism level,
and the differences were significantly different (except for between level 2 and level 3).
The results of the perception of realism followed the expected results as well as the result
of the prestudy (also see 3.5). The perception of realism between level 2 and level 3 was
not significantly different. Looking at the histogram in Figure 4.3b and Figure 4.3c and
the frequencies in Table 4.3 it can be seen that the answers are diversified, and no real
tendency is recognizable. It is assumed that the visual differences between these realism
level were not different enough or the participants may not be able to see the differences
due to technical limitations (discussed in Section 6.1.1). Looking at the answers to the
open questions (Section 4.5) some participants who consumed the realism levels in this
order (level 2 to level 3 or vice versa) stated that they could not see any differences or at
least they were not sure if there were any. Most of them automatically moved closer, but
then they could not see the whole virtual human. One participant stated the reason why
he could immediately see that the virtual human is not real that he could see the pixels.
5.3 Presence
As mentioned in Subsection 3.7.3 the presence questionnaire can be broken down into
three categories: General presence, Co-Presence and Social-Presence. The following
section discusses the results of the overall merged representation of presence (also see
Subsection 4.3) as well as the results of the subcategories (also see Subsections 4.3.1,
4.3.2, 4.3.3).
The positive influence of visual realism on the sense of presence was studied to clarify
how important visual realism for future AR applications is and which influencing factors
should be taken into account. Additionally, the correlation between the sense of presence
and visual realism was evaluated. The results in subsection 4.2 suggest that there is an
effect of visual realism on the sense of presence. The presence decreases with a lower
realism level. However, a specific analysis of the differences between the realism levels
shows that only one difference (between realism level 2 and 3) is statistically significant.
Interesting is the difference between the sense of presence from realism level 4 to 5. The
gradient becomes much flatter than the differences before. The mean values between
realism level 4 and 5 are clearly not statistically significant and also a graphically very
close together. This is notable because it seems like that a virtual human with clearly
unrealistic geometry like the virtual human used in realism level 5 can score almost the
same sense of presence than a low-quality virtual human with more realistic geometry.
Looking at the answers to the open question (Section 4.5) some participants stated
that they clearly recognized the unrealistic or game-like representation. At the same
time, they liked the virtual human in level 5 often more than the others. An interesting
statement was made about virtual human used in realism level 5 "Because the avatar did
not try to look like a real person, his presence was a lot more believable."
70
5.3. Presence
The results of the categories of overall presence show similar trends, but also some
interesting differences. The mean values of the category ’General Presence’ show that
presence decreases with a lower realism level and also shows statistically significant
differences only between level 2 and 3 (same as the overall merged presence results). The
results of the category ’Co-Presence’ show also the trend that presence decreases with a
lower realism level, except for realism level 4 and 5. Here it is called a ’trend’ because
the repeated-measures ANOVA test shows that there were no statistically significant
changes across the means. Nevertheless, the mean values between realism level 4 and 5
are interesting because they are actually increasing with a lower realism level. Even if the
difference is statistically not significant, it seems like that a virtual human with cleary
unrealistic geometry can score a higher Co-Presence than a low-quality virtual human
with more realistic geometry. The category ’Social Presence’ also shows that the sense of
presence decreases with a lower realism level, except between the realism level 4 to 5. The
results show statistically significant differences only between level 2 and 3 (same as the
overall merged presence results). Same as in the category ’Co-Presence’, the mean values
between realism level 4 and 5 are interesting because they are actually increasing with a
lower realism level. Even if the difference is statistically not significant, it seems like that
a virtual human with cleary unrealistic geometry can score a higher Social Presence than
a low-quality virtual human with more realistic geometry. Sadly the reliability of the
questions in this category was too low. If the second question SP-2 ”The thought that
the guide is not a real person crosses my mind often.” is left out of the reliability test the
reliability increases dramatically, so it can be assumed that participants misunderstood
this question due to its negative nature (also see Subsection 3.7).
Also interesting is to map the overall presence score to the hypothesized Uncanny Valley
graph by Mori [MMK12]. It roughly followed the original hypothesized Uncanny Valley
graph, except that the slope may not be that steep as in the original graph (Figure 2.3).
The theoretical background of the Uncanny Valley effect is discussed in Section 2.6. We
mapped the original term ’Familiarity’ to the means of the overall sense of presence.
Figure 5.1 shows the mapping of the overall presence score to the Uncanny Valley graph.
The mapping has been done by calculating the presence percentage values (y-axis) and
calculating the realism value based on the means of perceived realism (x-axis).
71
5. Discussion
Figure 5.1: Hypothesized mapping of the mean of overall sense of presence to the Uncanny
Valley graph
.
All in all, the hypothesis H1 was partially supported. These results show a clear trend
and indicate partly significant differences between the realism levels. These results are
different from results in the past from Lee et al. [LRM+13] and Slater et al. [SKMY09].
5.4 Uncanny Valley
During the experiment design (also see Subsection 3.3) the original scales designed by
Ho and MacDorman [HM10, HM17] were reduced to three questions about the overall
sense of Humanness, Eeriness, and Attractiveness. Sadly, this attempt was not successful,
and the reliability of the questions was too low. It could also be that participants
misunderstood the questions since question number two and three were recoded due to
their negative nature (also see Subsection 3.7). Nevertheless, the results of the mean
value analysis in Subsection 4.4 is discussed in the following section. Additionally, this
section discusses the possibility to map the scores to the original assumed plot of the
Uncanny Valley effect by Mori [MMK12]. The theoretical background of the Uncanny
Valley effect is discussed in Section 2.6. In the original plot (Figure 2.3) Mori used the
terms ’Human likeliness’ and ’Familiarity’. In this discussion, we understand ’Human
likeness’ as visual realism levels of the virtual human and ’Familiarity’ as the mean scores
of the Uncanny Valley questionnaire.
Even if a form of the Uncanny Valley effect could not be observed the results allow other
interesting interpretations. The results suggest that there is an effect of visual realism
on the familiarity of the user. The repeated-measures ANOVA test showed significant
72
5.4. Uncanny Valley
changes across the means. Nevertheless, a specific analysis of the differences between the
realism levels shows that only the differences of the means between realism level 3 and 4
were not significant. From realism level 1 to 4 it can be seen the higher the realism, the
higher is the familiarity score. Interesting is the difference between the realism level 4
to 5 as well as the mean value from realism level 5 itself which is about as high as the
value of realism level 2. It is statistically significant that a virtual human with clearly
unrealistic representation (realism level 5) can score a higher familiarity than low-quality
virtual human with more realistic representation (in this case level 2,3 and 4).
Mapping the results to the hypothesized graph of the Uncanny Valley effect by Mori [MMK12]
indicate that the first part of the curve may not be as steadily rising as assumed by
Mori [MMK12]. This thesis hypothesizes that the curve from 0% realism to about 50%
is not steadily rising, but could include a hump (something like a local maximum).
Figure 5.2 shows the mapping of the relative familiarity score with the Uncanny Valley
graph. The mapping has been done by calculating the familiarity percentage values
(y-axis) and calculating the realism value based on the means of perceived realism (x-axis).
Apparently, the is shows no tendencies of an Uncanny Valley effect.
Figure 5.2: Hypothesized mapping of the quentionnaire score to the Uncanny Valley
graph
.
The results do not support hypothesis H2 since no tendencies of the Uncanny Valley
effect is detectable. Nevertheless, the findings support the findings stated in Section 5.3
and further also partly supports hypothesis H1. Further, a new hypothesis for future
studies was created.
73

CHAPTER 6
Conclusion
6.1 Summary
The main goal of this thesis was to study the effect of increasing realism on the perception
of presence and on the convenience of users in AR. To evaluate the effect of visual
realism to the sense of presence, an experiment was conducted in which each participant
experienced a virtual human for each realism level. The results indicate that visual
realism is an important factor to reach a higher sense of presence within AR application.
6.1.1 Technical limitations
The technical setup of the used AR system (HoloLens) stresses some technical limitations
which could potentially lead to biased answers. Perhaps the most important lesson
learned was the difficulty to assess what kind of experimental design is actually possible
due to technical possibilities.
Providing the same performance and user experience in all realism levels is a challenge.
Initially, it was planned that participants should be navigated through a building with
the help of a virtual human. The feature ’Navigation’ and ’Walking’ are with no doubt
possible if drawbacks due to realism and animation are acceptable. In the case of this
thesis, we could not provide these features at the highest visual realism level without
frame drops and further loose user experience. For the same reason, neither global
illumination nor shadows could be used.
Additionally, the limited field of view of the HoloLens in combination with a walking
virtual human in space cause that the participant could lose sight of the virtual human.
This would most likely break the presence and produce biased answers. Most of the
participants claimed about the limited field of view. The limited field of view prevented
the participants from seeing the whole virtual human during the whole experiment. Some
participants reported that the virtual humans seems to be transparent or would shine.
75
6. Conclusion
This is due to the light-based displays build into the HoloLens. These displays add virtual
objects into the field of view of the user by adding light to the real-world, but it cannot
reduce light. If the surrounding environment is too bright, the virtual objects can appear
transparent. If the surrounding environment is too dark virtual objects tend to look like
they would shine. Before the study other optical as well as video see-throw HMD systems
were tested, but no system provided a tracking system as robust as the HoloLens offers.
After several tests and due to the technical problems mentioned above the navigation
and walking features were discarded. It was decided to concentrate on a standing virtual
human which looks at the participant at any time of the experiment.
The technical capabilities of today’s AR systems limit the experiment design possibilities.
In 2007 Guadagno et al. [GBBM07] pointed out that the available technology is not
capable of creating highly realistic virtual humans. Ten years later the technology has
developed rapidly, but this problem still exists today. Especially the limited field of view
most probably causes biased measurements. The ideal AR system for such a study should
be a mobile head held AR system which at least provides accurate tracking, powerful
high-quality rendering, shadows, global illumination, at least three times wider field of
view, etc. Future research should consider repeating this user study as soon as new AR
systems solve the above mentioned technical problems. The influence of field of view on
the presence could also be validated in the future.
6.2 Future Work
During the implementation, we ran into some technical problems due to the technical
limitations of the used AR system. Also, a part of the questionnaire was not reliable
enough. Further, some participants of the user study provided feedback which uncovered
potential improvement. This section discusses these and other problems and suggestions
for future research.
6.2.1 Uncanny Valley
During the discussion of the results in Section 5.4 we hypothesized that the curve from
0% to 50% is not steadily rising but could include a hump. Future work should focus on
this finding to validate the occurence of this ’hump’. This could be done by conducting a
user study which focus on the human likeness between 0% to 50% of the Uncanny Valley
graph.
6.2.2 Virtual human
The realism levels among the virtual humans differ in a few factors which were changes
in each realism level. Although this approach produces useful different realism levels
it is not possible to determine which factor exactly scored a different sense of presence
afterwards. Additionally, the visual differences between level 4 and 5 maybe were too big.
Future work could provide finer gradations between the low poly, no texture only colour
76
6.2. Future Work
virtual human (level 4) and a game-like, block style virtual human (level 5). Future work
should address these circumstances.
A limitation of this research was that only one male virtual human was used for the
study. Studies in the past have shown a significant difference in perception and task
performance of a virtual human with a different gender than the participant was used
[BBBL03, ZPP03, GBBM07, SSSVB10, LLL15]. To minimize the influence of unequal
gender, future work should consider having multiple virtual humans of each gender.
6.2.3 Measuring experience
Past studies mentioned the problematic of biased measures if perception and presence or
even experience overall is measured through a questionnaire. There are other approaches
to measure experiences like EEG, brainwaves, etc. but so far no standard procedure has
been developed. These measurement could be very interesting for future work.
The attempt to create a reduced version of the Uncanny Valley scales created by Ho
and MacDorman [HM10, HM17] did not show enough reliability in our study. Future
work should either use with all available dimensions or, if a reduced version is necessary,
should test the reduced version before using it in further studies.
77

List of Figures
2.1 Milgram’s Mixed Reality continuum [MK94] . . . . . . . . . . . . . . . . . 6
2.2 Concepts of optical see-through (left) and video see-through displays (right).
Adapted from [ABF+01] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Uncanny Valley graph [MMK12] . . . . . . . . . . . . . . . . . . . . . . . 19
3.1 Used virtual human at different visual realism levels . . . . . . . . . . . . 29
3.2 Examples of polygons of a virtual human at different levels . . . . . . . . 30
3.3 Example of polygons of a virtual humans hand . . . . . . . . . . . . . . . 30
3.4 Examples of a virtual humans with different texture resolutions . . . . . . 30
3.5 Light sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 Expected order of visual realism levels . . . . . . . . . . . . . . . . . . . . 34
3.7 Prestudy web application . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.1 Collision histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 First round collision histograms . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Histograms of perceived realism . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4 Median of perceived realism scores from 1-7 (vertical) for each realism level
1-5 (horizontal) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.5 Presence histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.6 Mean of presence across realism levels . . . . . . . . . . . . . . . . . . . . 47
4.7 General presence histogram . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.8 Mean of general presence across realism levels . . . . . . . . . . . . . . . . 50
4.9 Graphical normal distribution analysis between differences between general
presence mean values of realism level 3 and 4 . . . . . . . . . . . . . . . . 52
4.10 Co-Presence histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.11 Mean of Co-Presence across realism levels . . . . . . . . . . . . . . . . . . 54
4.12 Graphical normal distribution analysis between differences between Co-Presence
mean values of realism level 1 and 2 . . . . . . . . . . . . . . . . . . . . . 56
4.13 Graphical normal distribution analysis between differences between Co-Presence
mean values of realism level 2 and 3 . . . . . . . . . . . . . . . . . . . . . 57
4.14 Graphical normal distribution analysis between differences between Co-Presence
mean values of realism level 3 and 4 . . . . . . . . . . . . . . . . . . . . . 58
79
4.15 Graphical normal distribution analysis between differences between Co-Presence
mean values of realism level 4 and 5 . . . . . . . . . . . . . . . . . . . . . 59
4.16 Social presence histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.17 Mean of Social Presence across realism levels . . . . . . . . . . . . . . . . . 61
4.19 Means of Uncanny Valley scores across realism levels . . . . . . . . . . . . 62
4.18 Uncanny Valley histogram . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.1 Hypothesized mapping of the mean of overall sense of presence to the Uncanny
Valley graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2 Hypothesized mapping of the quentionnaire score to the Uncanny Valley graph 73
80
List of Tables
2.1 Mapping properties of AR to technical requirements [BCL15] . . . . . . . 7
2.2 Hierarchy of AR development tools from most complex to least complex after
Billinghurst et al. [BCL15] . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.1 Perception of realism questionnaire . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Presence questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.3 Co-presence questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Social presence questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.5 Original Uncanny Valley scales adapted from Ho and MacDorman [HM10,
HM17] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.6 Uncanny Valley questions used in the experiment . . . . . . . . . . . . . . 24
3.7 Open question of participants’ perception . . . . . . . . . . . . . . . . . . 25
3.8 Questionnaire used in the user study . . . . . . . . . . . . . . . . . . . . . 26
3.9 Visual realism factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.10 Visual realism levels and factors . . . . . . . . . . . . . . . . . . . . . . . 28
3.11 Light sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.12 Animations used in user study . . . . . . . . . . . . . . . . . . . . . . . . 33
3.13 Collision answer mapping into nominal data . . . . . . . . . . . . . . . . . 36
4.1 Collision - Contingency table . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Collision in the first round - Contingency table . . . . . . . . . . . . . . . 42
4.3 Contingency table - Perceived realism scores from 1-7 (horizontal) for each
realism level (vertical) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4 Mean and standard deviation of presence . . . . . . . . . . . . . . . . . . 47
4.5 Mean and standard deviation of general presence . . . . . . . . . . . . . . 50
4.6 Mean and standard deviation of CoPresence . . . . . . . . . . . . . . . . . 54
4.7 Mean and standard deviation of social presence . . . . . . . . . . . . . . . . 61
4.8 Mean and standard deviation of Uncanny Valley scores . . . . . . . . . . 62
4.9 Answers to the open question - Level 1 . . . . . . . . . . . . . . . . . . . . 65
4.10 Answers to the open question - Level 2 . . . . . . . . . . . . . . . . . . . . 65
4.11 Answers to the open question - Level 3 . . . . . . . . . . . . . . . . . . . . 66
4.12 Answers to the open question - Level 4 . . . . . . . . . . . . . . . . . . . . 66
4.13 Answers of open question - Level 5 . . . . . . . . . . . . . . . . . . . . . . 67
81
4.14 Answers of participants which consumed level 2 and level 3 in this order (or
vice versa) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
82
Bibliography
[ABF+01] Ronald Azuma, Reinhold Behringer, Steven Feiner, Simon Julier, and
Blair Macintyre. Recent Advances in Augmented Reality. IEEE Computer
Graphics and Applications, 2011(December):1–27, 2001.
[ABRK13] B. Anis, F. Börrnert, M. H. Ruümmeli, and C. A. Kuntscher. Role of
the pressure transmitting medium on the pressure effects in DWCNTs.
Physica Status Solidi (B) Basic Research, 250(12):2616–2621, 2013.
[ALZN03] Kusuma Agusanto, Li Li, Chuangui Zhu, and Wan Sing Ng. Photorealistic
rendering for augmented reality using environment illumination. In
Proceedings - 2nd IEEE and ACM International Symposium on Mixed
and Augmented Reality, ISMAR 2003, 2003.
[Azu97] Ronald Azuma. A survey of augmented reality. Presence: Teleoperators
and Virtual Environments, 6(4):355–385, 1997.
[BBBL03] Jeremy N. Bailenson, Jim Blascovich, Andrew C. Beall, and Jack M.
Loomis. Interpersonal Distances in Virtual Environments. Personality
and Social Psychology Bulletin, 29(7):819–833, 2003.
[BCL15] Mark Billinghurst, Adrian Clark, and Gun Lee. A Survey of Augmented
Reality. Foundations and Trends R© in Human–Computer Interaction,
2015.
[BGWK03] Oliver Bimber, Anselm Grundhofer, Gordon Wetzstein, and Sebastian
Knodel. Consistent illumination within optical see-through augmented
environments. In Proceedings - 2nd IEEE and ACM International Sym-
posium on Mixed and Augmented Reality, ISMAR 2003, 2003.
[BHSS00] Cagatay Basdogan, Chih-Hao Ho, Mandayam A Srinivasan, and Mel
Slater. An Experimental Study on the Role of Touch in Shared Virtual
Environments. ACM Trans. Comput.-Hum. Interact., 7(4):443–460, 2000.
[BKCZ09] Christoph Bartneck, Dana Kulić, Elizabeth Croft, and Susana Zoghbi.
Measurement instruments for the anthropomorphism, animacy, likeability,
perceived intelligence, and perceived safety of robots. International
Journal of Social Robotics, 1(1):71–81, 2009.
83
[BKIH09] Christoph Bartneck, Takayuki Kanda, Hiroshi Ishiguro, and Norihiro
Hagita. My robotic doppelgänger - A critical look at the Uncanny Valley.
In Proceedings - IEEE International Workshop on Robot and Human
Interactive Communication, 2009.
[Bla02] Jim Blascovich. Social influence within immersive virtual environments.
In The Social Life of Avatars, pages 127–145. Springer London, London,
2002.
[BP08] José M Braz and João M Pereira. TARCAST: Taxonomy for Augmented
Reality CASTing with Web Support. The International Journal of
Virtual Reality, 7(4):47–56, 2008.
[Cut03] James E. Cutting. Reconceiving perceptual space. Looking into pictures:
An interdisciplinary approach to pictorial space, 2003.
[DM96] David Drascic and Paul Milgram. Perceptual Issues in Augmented Reality
Reality-Virtuality Continuum. Virtual Reality, (March 2001):123–134,
1996.
[DRÝ+03] George Drettakis, Maria Roussou, Maria Roussou Ý, George Dret-
takis, Maria Roussou, and George Drettakis. Photorealism and non-
photorealism in virtual heritage representation. . . . Workshop on Graph-
ics and Cultural Heritage . . . , (May 2013):47–56, 2003.
[Dur02] Frédo Durand. An invitation to discuss computer depiction. Npar ’02,
pages 111–124, 2002.
[ENC+08] Mohamed Elhelw, Marios Nicolaou, Adrian Chung, Guang-Zhong Yang,
and M. Stella Atkins. A gaze-based study for investigating the perception
of visual realism in simulated scenes. ACM Transactions on Applied
Perception, 2008.
[FAJ+15] Jesse Fox, Sun Joo Grace Ahn, Joris H. Janssen, Leo Yeykelis, Kathryn Y.
Segovia, and Jeremy N. Bailenson. Avatars versus agents: A meta-analysis
quantifying the effect of agency on social influence. Human-Computer
Interaction, 30(5):401–432, 2015.
[Fer03] Ja Ferwerda. Three varieties of realism in computer graphics. Proceedings
SPIE Human Vision and Electronic, 2003.
[Fie17] Andy P. Field. Discovering statistics using IBM SPSS statistics. Sage
Publications Ltd., 4th edition, 2017.
[GBBM07] Rosanna E Guadagno, Jim Blascovich, Jeremy N Bailenson, and Cade
McCall. Virtual humans and persuasion: The effects of agency and
behavioral realism. Media Psychology, 10(1):1–22, 2007.
84
[GSPR05] Maia Garau, Mel Slater, David-Paul Pertaub, and Sharif Razzaque.
The Responses of People to Virtual Humans in an Immersive Virtual
Environment. Presence, 14(1):104–116, 2005.
[Hal04] Michael Haller. Photorealism or / and Non-Photorealism in Augmented
Reality. Proceedings of the 2004 ACM SIGGRAPH international confer-
ence on Virtual Reality continuum and its applications in industry, pages
189–196, 2004.
[HB96] Claudia Hendrix and Woodrow Barfield. Presence within virtual environ-
ments as a function of visual display parameters. Presence: Teleoperators
and Virtual Environments, 1996.
[HB04] Chad Harms and Frank Biocca. Internal Consistency and Reliability
of the Networked Minds Measure of Social Presence. Seventh Annual
International Workshop: Presence 2004, pages 246–251, 2004.
[HDO09] Thomas Holz, Mauro Dragone, and G. M.P. O’Hare. Where robots
and virtual agents meet: A survey of social interaction research across
milgram’s reality-virtuality continuum. International Journal of Social
Robotics, 1(1):83–93, 2009.
[Hee92] Carrie Heeter. Being There: The Subjective Experience of Presence.
Presence: Teleoperators and Virtual Environments, 1(2):262–271, 1992.
[HFJS09] Timothy J Hattenberger, Mark D Fairchild, Garrett M Johnson, and
Carl Salvaggio. A Psychophysical Investigation of Global Illumination
Algorithms Used in Augmented Reality. ACM Trans. Appl. Percept.,
6(1):2:1—-2:22, 2009.
[HFN11] Olivier Hugues, Philippe Fuchs, and Olivier Nannipieri. New Augmented
Reality Taxonomy: Technologies and Features of Augmented Environ-
ment. In Handbook of Augmented Reality. 2011.
[HM10] Chin Chang Ho and Karl F. MacDorman. Revisiting the uncanny valley
theory: Developing and validating an alternative to the Godspeed indices.
Computers in Human Behavior, 26(6):1508–1518, 2010.
[HM17] Chin-Chang Chang Ho and Karl F. MacDorman. Measuring the Uncanny
Valley Effect. International Journal of Social Robotics, 9(1):129–139,
2017.
[II00] International Society for Presence research and ISPR. The Concept of
Presence: Explication Statement, 2000.
[JKK15] Dongsik Jo, Ki Hong Kim, and Gerard Jounghyun Kim. SpaceTime:
Adaptive control of the teleported avatar for improved AR tele-conference
experience. In Computer Animation and Virtual Worlds, 2015.
85
[KBW17] Kangsoo Kim, Gerd Bruder, and Greg Welch. Exploring the Effects of
Observed Physicality Conflicts on Real–Virtual Human Interaction in
Augmented Reality. Proceedings of the 23rd ACM Symposium on Virtual
Reality Software and Technology, 717:31:1–31:7, 2017.
[KDB+14] Peter Kán, Andreas Dünser, Mark Billinghurst, Christian Schönauer,
and Hannes Kaufmann. The Effects of Direct and Global Illumination on
Presence in Augmented Reality. Challenging Presence - Proceedings of
15th International Conference on Presence (ISPR 2014), pages 223–230,
2014.
[KDT+11] Martin Knecht, Andreas Dünser, Christoph Traxler, Michael Wimmer,
and Raphael Grasset. A Framework For Perceptual Studies In Photoreal-
istic Augmented Reality. Proceedings of the 3rd IEEE VR 2011 Workshop
on Perceptual Illusions in Virtual Environments, 2011.
[KM08] Georg Klein and David Murray. Compositing for small cameras. In Pro-
ceedings - 7th IEEE International Symposium on Mixed and Augmented
Reality 2008, ISMAR 2008, 2008.
[KMB+17] Kangsoo Kim, Divine Maloney, Gerd Bruder, Jeremy N. Bailenson, and
Gregory F. Welch. The effects of virtual human’s spatial and behavioral
coherence with physical objects on social presence in AR. Computer
Animation and Virtual Worlds, 28(3-4):1–9, 2017.
[KSF10] Ernst Kruijff, J. Edward Swan, and Steven Feiner. Perceptual issues in
augmented reality revisited. In 9th IEEE International Symposium on
Mixed and Augmented Reality 2010: Science and Technology, ISMAR
2010 - Proceedings, 2010.
[KW15] Kangsoo Kim and Greg Welch. Maintaining and enhancing human-
surrogate presence in augmented reality. In Proceedings of the 2015 IEEE
International Symposium on Mixed and Augmented Reality Workshops,
ISMARW 2015, 2015.
[LD97] Matthew Lombard and Theresa Ditton. At the Heart of It All: The
Concept of Presence. Journal of Computer-Mediated Communication,
3(2):0–0, jun 1997.
[LK03] Jack M Loomis and Joshua M Knapp. Visual perception of egocentric
distance in real and virtual environments. In Virtual and Adaptive
Environments. 2003.
[LLL15] Jean Luc Lugrin, Johanna Latt, and Marc Erich Latoschik. Avatar
anthropomorphism and illusion of body ownership in VR. In 2015 IEEE
Virtual Reality Conference, VR 2015 - Proceedings, 2015.
86
[LNB+13] Yun Ling, Harold T. Nefs, Willem Paul Brinkman, Chao Qu, and Ingrid
Heynderickx. The effect of perspective on presence and space perception.
PLoS ONE, 8(11), 2013.
[LRG+17] Marc Erich Latoschik, Daniel Roth, Dominik Gall, Jascha Achenbach,
Thomas Waltemate, and Mario Botsch. The Effect of Avatar Realism
in Immersive Social Virtual Realities. 23rd ACM Symposium on Virtual
Reality Software and Technology (VRST 2017), 2017.
[LRM+13] Cha Lee, Gustavo A. Rincon, Greg Meyer, Tobias Höllerer, and Doug A.
Bowman. The effects of visual realism on search tasks in mixed reality
simulation. IEEE Transactions on Visualization and Computer Graphics,
19(4):547–556, 2013.
[Man02] Steve Mann. Mediated Reality with implementations for everyday life.
Presence: Teleoperators and Virtual Environments, pages 1–13, 2002.
[McC00] MD McCool. Shadow volume reconstruction from depth maps. ACM
Transactions on Graphics (TOG), 2000.
[Mic17] Microsoft. Spatial Mapping (Windows Holographic)., 2017.
[MK94] Paul Milgram and Fumio Kishino. A Taxonomy of Mixed Reality Visual
Displays. IEICE Trans. Information Systems, vol. E77-D:1321–1329,
1994.
[MMK12] Masahiro Mori, Karl F. MacDorman, and Norri Kageki. The uncanny
valley. IEEE Robotics and Automation Magazine, 19(2):98–100, 2012.
[MR04] Katerina Mania and Andrew Robinson. The effect of quality of render-
ing on user lighting impressions and presence in virtual environments.
Proceedings of the 2004 ACM SIGGRAPH International Conference on
Virtual Reality Continuum and Its Applications in Industry, 1(212):200–
205, 2004.
[MS16] Federico Manuri and Andrea Sanna. A Survey on Applications of Aug-
mented Reality. Advances in Computer Science, 5(1):18–27, 2016.
[MTPM08] Nadia Magnenat-Thalmann, George Papagiannakis, and Parag Chaud-
huri Miralab. APPLICATIONS OF INTERACTIVE VIRTUAL HU-
MANS IN MOBILE AUGMENTED REALITY. Encyclopaedia of Multi-
media, 2nd edn, 2008.
[NB03] Kristine L. Nowak and Frank Biocca. The Effect of the Agency and
Anthropomorphism on Users’ Sense of Telepresence, Copresence, and
Social Presence in Virtual Environments. Presence: Teleoperators and
Virtual Environments, 12(5):481–494, 2003.
87
[NNMH02] T Naemura, T Nitta, A Mimura, and H Harashima. Virtual shadows in
mixed reality environment using flashlight-like devices. Trans Virtual
Reality Society of Japan, 7(2):227–237, 2002.
[ODK+12] Mohammad Obaid, Ionut Damian, Felix Kistler, Birgit Endrass, Johannes
Wagner, and Elisabeth André. Cultural Behaviors of Virtual Agents in
an Augmented Reality Environment. In Yukiko Nakano, Michael Neff,
Ana Paiva, and Marilyn Walker, editors, Intelligent Virtual Agents: 12th
International Conference, IVA 2012, Santa Cruz, CA, USA, September,
12-14, 2012. Proceedings, pages 412–418. Springer Berlin Heidelberg,
Berlin, Heidelberg, 2012.
[ONP11] Mohammad Obaid, Radosław Niewiadomski, and Catherine Pelachaud.
Perception of Spatial Relations and of Coexistence with Virtual Agents. In
Hannes Högni Vilhjálmsson, Stefan Kopp, Stacy Marsella, and Kristinn R
Thórisson, editors, Intelligent Virtual Agents: 10th International Confer-
ence, IVA 2011, Reykjavik, Iceland, September 15-17, 2011. Proceedings,
pages 363–369. Springer Berlin Heidelberg, Berlin, Heidelberg, 2011.
[PKB+16] Tomislav Pejsa, Julian Kantor, Hrvoje Benko, Eyal Ofek, and Andrew D
Wilson. Room2Room: Enabling Life-Size Telepresence in a Projected
Augmented Reality Environment. In Proceedings of the 19th ACM Con-
ference on Computer-Supported Cooperative Work & Social Computing -
CSCW ’16, 2016.
[RBBS17] Holger Regenbrecht, Cristina Botella, Rosa Baños, and Thomas Schubert.
Mixed Reality Experience Questionnaire (MREQ)—Reference. 2017.
[Reg02] Holger Regenbrecht. Measuring presence in augmented reality environ-
ments: design and a first test of a questionnaire. Annual International
Workshop Presence, (August 2017):1–4, 2002.
[Reg13] T. Regenbrecht, H., Botella, C., Banos, R., & Schubert. Mixed Reality
Experience Questionnaire (MREQ) Version 1.0, 23/05/2013, 2013.
[RN95] Jun Rekimoto and Katashi Nagao. The world through the computer. In
Proceedings of the 8th annual ACM symposium on User interface and
software technology - UIST ’95, pages 29–36, 1995.
[SCP07] J. Smart, J. Cascio, and J. Paffendorf. Metaverse Roadmap: Pathways to
the 3D Web. Metaverse: a cross-industry public foresight project, 2007.
[SKMY09] Mel Slater, Pankaj Khanna, Jesper Mortensen, and Insu Yu. Visual Re-
alism Enhances Realistic Response in an Immersive Virtual Environment.
IEEE Computer Graphics and Applications, pages 76–84, 2009.
88
[SKT03] Natsuki Sugano, Hirokazu Kato, and Keihachiro Tachibana. The effects of
shadow representation of virtual objects in augmented reality. Proceedings
- 2nd IEEE and ACM International Symposium on Mixed and Augmented
Reality, ISMAR 2003, pages 76–83, 2003.
[Sla09] M. Slater. Place illusion and plausibility can lead to realistic behaviour
in immersive virtual environments. Philosophical Transactions of the
Royal Society B: Biological Sciences, 364(1535):3549–3557, 2009.
[SSSVB10] Mel Slater, Bernhard Spanlang, Maria V. Sanchez-Vives, and Olaf Blanke.
First person experience of body transfer in virtual reality. PLoS ONE,
5(5), 2010.
[Ste94] Neal Stephenson. Snow crash, 1994.
[SUC95] Mel Slater, Martin Usoh, and Yiorgos Chrysanthou. The influence of
dynamic shadows on presence in immersive virtual environments. Selected
papers of the Eurographics workshops on Virtual environments, 1995.
[SvdSKvdM01] Martijn J. Schuemie, Peter van der Straaten, Merel Krijn, and
Charles A.P.G. van der Mast. Research on Presence in Virtual Reality:
A Survey. CyberPsychology & Behavior, 2001.
[TBS+98] Jolanda Tromp, Adrian Bullock, Anthony Steed, Amela Sadagic, Mel
Slater, and Emmanuel Frécon. Small group behavior experiments in the
coven project. IEEE Computer Graphics and Applications, 1998.
[TFB+00] Rémy Torre, Pascal Fua, Selim Balcisoy, Michal Ponder, and Daniel Thal-
mann. Interaction Between Real and Virtual Humans: Playing Checkers.
In Jurriaan Mulder and Robert van Liere, editors, Virtual Environments
2000: Proceedings of the Eurographics Workshop in Amsterdam, The
Netherlands, June 1–2, 2000, pages 23–32. Springer Vienna, Vienna,
2000.
[TWG+04] William B Thompson, Peter Willemsen, Amy A Gooch, Sarah H Creem-
Regehr, Jack M Loomis, and Andrew C Beall. Does the quality of the
computer graphics matter when judging distances in visually immersive
environments? Presence: Teleoperators and Virtual Environments, 2004.
[VB04] Vinoba Vinayagamoorthy and Andrea Brogni. An Investigation of
Presence Response across Variations in Visual Realism. Proceedings
of Presence 2004: The 7th Annual International Workshop on Presence,
(September 2017):119–126, 2004.
[YBR07] Nick Yee, Jeremy N Bailenson, and Kathryn Rickertsen. A meta-analysis
of the impact of the inclusion and realism of human-like faces on user
experiences in interfaces. Proceedings of the SIGCHI conference on
Human factors in computing systems, pages 1–10, 2007.
89
[YMK+12] Insu Yu, Jesper Mortensen, Pankaj Khanna, Bernhard Spanlang, and
Mel Slater. Visual realism enhances realistic response in an immersive
virtual environment - Part 2. IEEE Computer Graphics and Applications,
pages 36–45, 2012.
[ZPP03] Paul Zimmons, Abigail Panter, and D Ph. The Influence of Rendering
Quality on Presence And Task Performance in a Virtual Environment
at Chapel Hill at Chapel Hill. Proceedings of the IEEE Virtual Reality
2003, 2003:2–3, 2003.
90