Sunteți pe pagina 1din 97

Tangible Interfaces for Download:

initial observations on users’


everyday environments

Politecnico di Torino
III Facoltà di Ingegneria

Tesi di Laurea Magistrale in Ingegneria Informatica

Author:
Matteo Giaccone

Supervisor Ecole Polytechnique Relatore Politecnico di Torino:


Fédérale de Lausanne:

Enrico Costanza Prof. Aldo Laurentini

February 2010
Acknowledgements

This work would not be possible without the help and motivation
received in the years by my family, by Agnese and from a lot of old
time friends.

I have to thank also my friends in ACMOS that helped me to achieve


things that i would not thought possible being alone.

Thanks to the Media and Design Laboratory at EPFL to host my


master project, and to the Swiss hospitality given to me by Enrico,
Tal and Olivier that made my days in Lausanne be wonderful.

Thanks to Professor Jeffrey Huang in EPFL for accepting to host


me in the LDM lab and to all other people there for their hospitality.

Thanks to Professor Aldo Laurentini for giving to me the oppor-


tunity to go in Switzerland supervising my master project from the
Politecnico di Torino.

Finally, thanks to all the users of d-touch that had the courage of
trying this instruments and make this incredible adventure happen.

i
Abstract

Tangible User Interfaces (TUIs) have been studied and discussed


broadly in the Human Computer Interaction (HCI) community in
the past 15 years. Most reported TUI projects were studied only in
laboratories or in controlled environments for short amount of time.
With this study we distributed d-touch, a low cost tangible interface
for music composition and performance, through a website and we
remotely recorded interactions in users’ everyday environments. We
collected also user generated content deposed in public websites or
sent directly to us to get a better picture of the users’ impressions
about the experiment. We then qualitatively analyzed these content
with an approach grounded in data and we supported our theory
with a statistical analysis of interaction logs. The results show that,
even if the instruments were at an early stage, the tangible nature
of the interface has been perfectly accepted by the users as if it was
obvious to use.

ii
Contents

Acknowledgements i

Abstract ii

1 Introduction 1

2 Related Work 5
2.1 Early Work . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Audio TUI . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Markers . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 TUI User Studies . . . . . . . . . . . . . . . . . . . . 12
2.5 Large Scale User Studies . . . . . . . . . . . . . . . . 14
2.6 Grounded Theory . . . . . . . . . . . . . . . . . . . . 15

3 Audio d-touch 17
3.1 Working Principle . . . . . . . . . . . . . . . . . . . . 17
3.2 The Software . . . . . . . . . . . . . . . . . . . . . . 19
3.3 d-touch Drum Machine . . . . . . . . . . . . . . . . . 20
3.4 d-touch Sequencer . . . . . . . . . . . . . . . . . . . . 21
3.5 Logging System . . . . . . . . . . . . . . . . . . . . . 22
3.6 Distribution . . . . . . . . . . . . . . . . . . . . . . . 24
3.7 Diary of the Launch . . . . . . . . . . . . . . . . . . 26

4 Logs collection and analysis 29


4.1 Data Collection . . . . . . . . . . . . . . . . . . . . . 29
4.2 Data Analysis . . . . . . . . . . . . . . . . . . . . . . 30
4.2.1 Quantitative Log Analysis . . . . . . . . . . . 30
4.2.2 Qualitative Video Analysis . . . . . . . . . . . 34

iii
5 User Generated Content: collection and analysis 37
5.1 Data Collection . . . . . . . . . . . . . . . . . . . . . 37
5.2 Data Analysis . . . . . . . . . . . . . . . . . . . . . . 38
5.2.1 User Generated Content: Text . . . . . . . . . 38
5.2.2 User Generated Content: Videos and Photos . 45

6 Discussion 49

7 Future Work 52

8 Conclusions 54

A UGC texts 56

B Open Codes and Axial Coding 58

C UGC photos and videos analysis 61

D Riassunto in Italiano 69
D.1 Introduzione . . . . . . . . . . . . . . . . . . . . . . . 69
D.2 Audio d-touch . . . . . . . . . . . . . . . . . . . . . . 71
D.2.1 Come funziona . . . . . . . . . . . . . . . . . 72
D.2.2 d-touch Drum Machine . . . . . . . . . . . . . 73
D.2.3 d-touch Sequencer . . . . . . . . . . . . . . . 73
D.2.4 Sistema di Registrazione Remota . . . . . . . 74
D.2.5 Distribuzione . . . . . . . . . . . . . . . . . . 75
D.2.6 Il Lancio Online . . . . . . . . . . . . . . . . . 76
D.3 Raccolta e Analisi dei Log . . . . . . . . . . . . . . . 77
D.4 Raccolta e Analisi di Contenuti Creati dagli Utenti . 80
D.4.1 Analisi dei Testi . . . . . . . . . . . . . . . . . 81
D.4.2 Analisi delle Foto e dei Video . . . . . . . . . 82
D.5 Discussione . . . . . . . . . . . . . . . . . . . . . . . 83
D.6 Lavoro Futuro . . . . . . . . . . . . . . . . . . . . . . 85
D.7 Conclusione . . . . . . . . . . . . . . . . . . . . . . . 86

Bibliography 88

iv
Chapter 1

Introduction

Human Computer Interaction (HCI) is the field of work in which


this project is inscribed. The range of topics studied by HCI is
very broad and it includes various different subjects as computer
science, design, cognitive psychology, ergonomics and more. A clear
definition of HCI is given by ACM SIGCHI [1]:

Human-computer interaction is a discipline concerned with


the design, evaluation and implementation of interactive
computing systems for human use and with the study of
major phenomena surrounding them.

With the unstoppable spread of computing systems in all forms


and under every object, the interaction between humans and com-
puters is everyday more important and present. Computer science
should be interested in human-computer interaction studies at least
because a pleasurable and efficient interactive object is able to sell
better than others. It’s interesting to know that the studies in HCI
began during World War II when the first telecommunication de-
vices started to appear in war airplanes. The goals were to make this
new devices efficient and easy to use during high tension and danger
moments. Now the setting is completely changed and we are con-

1
stantly interacting with intelligent devices, from the personal com-
puter to the VCR and tens of other invisible intelligent companions.
This trend is described as ubiquitous or pervasive computing [2].

The importance of this studies is proved by the growing interest


of the major software and hardware companies to the academic de-
bate. All the major conferences on HCI, like CHI (ACM Conference
on Human Factors in Computing Systems) or UIST (ACM Sympo-
sium on User Interface Software and Technology) are sponsored by
companies like Microsoft, Google, Autodesk, IBM and Nokia1 .

One particular field of work of HCI is around Tangible User Inter-


faces (TUIs). The basic concept is that TUIs couple physical repre-
sentation (e.g., spatially manipulable physical objects) with digital
representation (e.g., graphics and audio), yielding user interfaces
that are computationally mediated but generally not identifiable as
“computers” per-se [3]. Tangible interaction is described by: tangi-
bility and materiality of the interface, physical embodiment of data,
whole-body interaction by the user, embedding of the user and the
interface in real spaces and contexts [4]. To give an analogy with
more familiar Graphical User Interfaces2 , in TUIs physical objects
represent digital informations and controls as in GUIs icons and
graphic metaphors are used.

TUIs are popular in HCI since the first half of 1990s when Well-
ner, Fitzmaurice, Ishii and Ullmer [5, 6, 3, 7] started developing
these new interfaces. TUIs are debated in the HCI community
since the pioneering works cited above, and since 2007 the ACM
participates to the organization of a specific conference on tangible
interfaces, the TEI conference3 .

Some areas of application are education [8, 9, 10], creativity [11,


12] and public expositions. Tangible interaction has been proved
1 CHI 2010 website: http://www.chi2010.org/ and UIST 2009 website: http://www.acm.org/uist/uist2009/
2 One classical example of GUI is the Desktop metaphor, used daily by millions of people.
3 Tangible and Embedded Interaction Conference. http://tei-conf.org/

2
interesting for certain type of educational tasks and improved cre-
ativity, especially when some kind of cooperation has been made
possible.

Tabletop tangible interaction is one branch of TUIs experiments,


in particular where the interface is horizontally placed and interac-
tion happens on it. This starts from the DigitalDesk of Wellner [5]
and comes to Reactable [11] as examples in the history of tabletop
TUIs.

Augmented Reality (AR) is another key aspect of TUIs and it


goes far beyond this specific application. With AR we are talking
about the process in which some digital (or virtual) information is
added to the perception of the real world. Some examples may be
the digital projection over real objects or the use of head-mounted
display to superimpose video to the real images. Some tangible in-
terface uses the possibilities given by AR to improve the interaction
with digital data through physical objects. A common approach
in AR uses Computer Vision algorithms to analyze markers that
are used as reference to draw the virtual images over the real en-
vironment through projection or in a video. Some different kind of
markers have been developed to support this kind of applications.
One of the most popular is the ARToolKit system developed in 1999
and used in hundreds of project, some of them in the mainstream
advertisement field4 . Audio d-touch uses the d-touch marker recog-
nition system, similar in concept to ARToolKit. The particular
point of d-touch is that the markers are human designable so that
they can convey information to humans and also to the machine.
This feature is useful for TUIs to enrich even more the object con-
taining data and to strengthen the link between virtual data and
physical object.

Until now no tangible interface has been tested in a real usage


environment in a large scale. The problems that blocked this testing
4 See for example: http://www.megabaile.com/ http://www.bmw.co.uk/z43d http://www.mini.de/webcam

3
were the high cost and the high technology that is usually needed by
a TUI, for example a sensing system or a retro-projected surface [13,
11]. Other typical usage of TUIs were in museums or in controlled
environments as school classes with trained teachers [14, 10]. The
suitability of TUIs in everyday environment with untrained users
remain an open question for everyday users’ environments.

A recent and emerging trend in HCI is on the Web potential to


create User Generated Content (UGC) and use them as a source
of data for user studies. Examples are uses of UGC to evaluate
commercial products and their emotional impact [15], the use of
crowd-sourcing web services, as the Amazon Mechanical Turk [16]
to recruit user study participants or the public Internet release of
GUI prototypes to run a user study on them [17].

The thesis work is a study on TUIs used in everyday environments


through an Internet-based logs observation and through UGC col-
lection and analysis. For that purpose we used Audio d-touch, a
previously existing TUI for musical composition and performance,
we made the software freely available through a Web site and we
gave instructions on how to build it and use it. We then logged all
the interaction produced by people usage, we collected UGC pro-
duced by users on Internet and after a statistical analysis of logs and
an analysis of all the collected data we concluded with some obser-
vations and future improvements on TUIs used in real settings. The
main interesting point that we have found from this study is that
the musical part of the application was not enough for our audience
and that, surprisingly, the tangible interface was hardly perceived,
as if it was completely obvious to use.

4
Chapter 2

Related Work

This chapter will analyze previous work on Tangible User Interfaces


and inspirational material for user studies.

2.1 Early Work

The first example of


tabletop tangible inter-
face documented is the
DigitalDesk by Pierre Well-
ner [5]. The aim of
this project was to cre-
ate a mixed reality ap-
plication, where a real
desktop was augmented
with computational ca-
pabilities. Two cameras Figure 2.1: DigitalDesk. 1991.
analyzed the desktop and
one projector sent the output directly on the desk. In this project
there are all the basic concepts found in many subsequent experi-
ments. The camera watching the scene from above (in later projects

5
camera and projectors are under the table), the projector that aug-
ment the surface near the camera and the computer vision program
to elaborate the data seen from the camera. This setup has inspired
also the d-touch applications that use the same strategy.

After the Wellner ex-


periment, Bricks, a project
by Fitzmaurice, Ishii and
Buxton, is another well
known experiment [6].
This experiment with the
related paper is very im-
portant for the history
of TUI, that were still
called Graspable User In- Figure 2.2: Bricks. 1995.
terfaces at the time of
Bricks, in 1995. The basic idea was to have a physical object (or
more than one), a brick, that you could grasp and use as an input
for the interface. The output was projected back under the bricks,
creating another tabletop interface as the DigitalDesk. The impor-
tance of this paper lays in the fact that for the first time a series
of important feature of TUIs (still called Graspable UIs) were an-
alyzed. From all the conclusions, two important points should be
highlighted. The experiments of Fitzmaurice et al. showed that hu-
mans have a lot of skills in bricks manipulation with two hands that
with traditional GUIs are not used at all. The obvious conclusion
for them was to use two bricks as input of the interface, showing also
that a TUI have the possibility to be more consistent than a GUI in
the interaction. The fact that GUIs have space-multiplexed output
and time-multiplexed input restrict the possibilities since the point
of interaction is always one instead of multiple possible. The TUI
instead, gives a space-multiplexed output and a space-multiplexed
input enriching the interaction as they showed in some practical
experiments with CAD programs.

6
The word Tangible User Interface was born with the paper Tan-
gible Bits by Ishii and Ullmer [7]. This paper is synergic with the
Weiser paper on Ubiquitous Computing (UbiComp) [18]. The focus
of the two papers is that with UbiComp and TUIs become possible
the use of natural human abilities like the ability to interact with
physical objects, and the use of background perception abilities of
humans. The Weiser objective is to put as much actions as possible
in the perceptive background. The concept behind is that if some-
one is doing something in his perceptive background, it means that
he can do something else in his focus, so doing more without using
more effort. With this idea in mind, Ishii and Ullmer developed
some experiments, like the ambientROOM, where the background
perception is used to react in case of ambient changes.

Another example is the


Live Wire of Jeremi-
jenko [19], where a wire
hanging from the ceiling
was connected to a little
motor activated by Eth-
ernet signals. So when
net traffic was higher the
wire was spinning faster
producing visual e au-
dio feedback in the room.
The people in the room
quickly became unaware
of the wire slow spin-
ning, but when a change
in speed happend every- Figure 2.3: Urp. 1999.

body could perceive it


and get the information of network activity without the need of
being actively watching the network statistics.

The project Urp [20] from Underkoffler and Ishii at the MIT Me-

7
dia Lab explores the possibility of tangible objects that express also
a physical meaning. Urp is a tangible table-top interface for urban
planning and design. The buildings in the applications are repre-
sented with objects with the shape of the real buildings in order
to understand how they look like and to see the shadows projected
on the table after Urp’s calculations. At the end of the paper the
authors explore the possibilities given by objects on TUIs. Here in
Urp objects can have the same shape of real buildings, or the shape
of a clock to regulate time used in shadow calculation and projec-
tion. Starting from these difference they explore the possibilities of
tangibles and their different possible meaning when used in TUIs.

Ullmer and Ishii in 2000


wrote a paper [3] that described
the state of the art of TUIs
and posed a discussion frame-
work, used afterwards in hun-
dreds of papers. A new interac-
tion model has been developed,
starting from the MVC (Model-
View-Controller) programming
Figure 2.4: Han’s multi touch screen. 2005.
paradigm. The new model
is called MCRpd, that means
Model-Control-Representation physical and Representation digital.
The idea of the MVC is that data in inputed in the Control part,
the Model has a digital representation of data, elaborates and gives
back an output in the View. The MCRpd has always a digital
Model of data, but the Control is tied with the physical Represen-
tation acting as an input and as an output at the same time. The
digital Representation is an added output (that might be optional)
as a digital information in form of video or audio. A good example
to explain this MCRpd is the abacus made in the paper. In the aba-
cus the representation is tightly coupled with the calculation model
and there is no distinction between input and output. Acting on the
physical representation is the same as acting directly on data, since

8
the model is directly connected to Representation and Control.

In parallel with the approach of tangible objects as control inter-


face, completely different experiments were done, for example using
only touch screens. The Han’s multi-touch screen [21] is a good
example in this different trend. The objective is always to get a
better interaction and to use the human abilities not explored with
traditional interfaces, for example the dual hand interaction already
described above [6]. One strong difference is that in this approach
tangible objects are not (or less) important in the interaction, while
on TUIs the tangibles are the focus of the interface.

2.2 Audio TUI

As expressed in [3] one application field for TUI is the artistic/expressive.


In this case we analyze the musical applications that have gained a
lot of media attention in the last years1 .

Audiopad [13] is the first


tabletop TUI for musical perfor-
mance that have gained media
and scientific attention. The in-
terface is a tabletop TUI with a
sensing plane based on RF tag.
Inspired by the DigitalDesk [5],
the output is projected from
above on the interface allowing Figure 2.5: Audiopad. 2002.
the visual augmentation of the
knobs used to control the application. The performer could play
samples, use audio effects and regulate audio parameters with the
knobs on the interface. The TUI then produced a MIDI output
used to control the audio external application. The sensing table,
1 For an extended list: http://modin.yuri.at/tangibles

9
the video feedback and the complex interaction are parts of the
project common in several subsequent experiments. Audio d-touch
is different from this approach in the marker tracking (Computer
Vision approach) and in the simple and essential interface, allowing
an easy reproduction and installation.

Reactable [11] is by far


the most known and pub-
licized tangible interface,
used in concerts by fa-
mous groups2 and sold by
a company3 . The con-
cept of this application
is the same as others al-
ready presented. There is
Figure 2.6: Reactable. 2005.
a table with a camera and
a projector. The camera is used to track markers and hand gestures,
the projector is used to give a video feedback on the interface. The
setup is different from the others because the projector and the
camera are under the table, allowing Reactable to be a table with
nothing else around, easy to move, to expose and making possible
the selling of a single object. The audio application is a modu-
lar synthesizer controlled by knobs with markers (passive plastic
blocks, in contrast with Audiopad) placed on the table. The size of
the table (around 2 meters of diameter) allow cooperation in music
playing, exploiting another point of interest in TUIs.

Another very popular tangible interface for music is BeatBear-


ing [22]. Probably this application had a lot of attention because
was easy to understand and it has been released online with instruc-
tions and source. This is not a classic tabletop musical TUI, but
the tangibility of balls used to close electronic contacts that trigger
musical samples is inspiring. The output of the interface was an
2 Bjork used Reactable in numerous concerts, one is visible here:
http://www.youtube.com/watch?v=MHeX6yg95xU
3 http://www.reactable.com/

10
LCD screen used horizontally as the base of the instrument. Over
the screen were placed the electronic circuits and holes that were
closed with steel balls to trigger the music.

2.3 Markers

The use of markers to study novel interfaces


was born in the Augmented Reality (AR) field
where the work of Kato and Billinghurst [23]
has been fundamental. They started a new
trend in marker recognition that later became
the famous ARToolKit4 . Now a lot of AR
projects use this kind of markers that permit
tracking and recognition in a 3D space after Figure marker
2.7: ARToolKit

some camera calibration. The particularity of


these markers is the square border that characterize them. The al-
gorithm searches for the square and then it does a template recogni-
tion of the content to understand which marker has been recognized.
This marker recognition system and the AR applications done with
ARToolKit are very famous and easy to build since BuildAr from
HITLabNZ5 , that enable users to create very easily AR applica-
tions on a desktop PC. Probably the most famous example of AR
applications at this moment are FLARToolKit6 based web advertis-
ing. FLARToolKit is an ActionScript 3 port for Adobe Flash 9+7
of the ARToolKit system, allowing developers to easily build AR
applications with Flash. This framework has been heavily used to
build every kind of advertising on web, having probably the greatest
success with cars8,9 , or collectible cards for children10,11 .
4 http://www.hitl.washington.edu/artoolkit/
5 http://www.hitlabnz.org/wiki/BuildAR
6 http://www.libspark.org/wiki/saqoosha/FLARToolKit/en
7 http://www.adobe.com/products/flash/
8 Mini advertisement: http://www.youtube.com/watch?v=dBser6 gToA
9 BMW Z4 advertisement: http://www.youtube.com/watch?v=cTUJKvXIkSU
10 Pokemon collectible cards: http://www.youtube.com/watch?v=FqjSz-5P-a0
11 Baseball collectible cards: http://www.youtube.com/watch?v=I7jm-AsY0lU

11
Another framework to build AR applications on desktop is DART [24].
The focus of DART and DART-TUI is to easily prototype Tangible
Interface applications through Adobe Director12 extensions. In the
paper they also illustrate an example application built with their
framework about physics and mathematics experiments in a school.

Audio d-touch instruments use the


Computer Vision approach to iden-
tify the position of tangible con-
trollers, instead of the more expen-
sive and complex sensing table found
in other TUI experiments. The visual
marker recognition approach used in d-
touch [25] is similar to the one used
in Reactable [26]. The algorithm de-
tects markers using an adaptive image Figure 2.8: d-touch drum machine
thresholding, followed by region adja- marker
cency analysis. In this way neither geometry nor color are inherent
properties, giving freedom to design your own shape. The design
has been proved working and the results are encouraging [27]. For
the d-touch drum machine we designed two particular markers that
proved the algorithm flexibility. One is visible in Figure 2.8.

2.4 TUI User Studies

Paul Marshall tries to understand if tangible interfaces are a good


tool in learning [9]. He analyzes previous experiments, possible
interesting domains of action and he gives guidelines for possible
benefits and areas that still need a scientific study. To confirm the
suppositions of Marshall we can find a lot of user studies on tangible
interfaces and a lot of them are in the field of learning [28, 12, 8,
10, 14].
12 http://www.adobe.com/products/director/

12
Two user studies on TUIs that we have analyzed [29, 28], were
done with adults and so it was possible the handling of a ques-
tionnaire (seven point Likert-scale or NASA-TLX) and a statistical
analysis on answers. In either cases the results were used to see
that the interface was working as expected and to compare it to a
preexistent model to have a comparison. Instead, all the user stud-
ies analyzed used informal questions, or the analysis of videotaped
experiments to gain a better understanding of the study and to find
problems or possible improvements. We decided then to not use
questionnaires neither to ask people to fill them, since it was an ex-
ploratory study and we didn’t find good previous work to compare
in a large scale. We tried to gain some insight from the users answer-
ing emails or on the d-touch forum13 in form of informal questioning
or just reading the thought that people autonomously sent us.

One commont point that we tried to overcome was the small


number of people involved in user studies. In the cited papers the
number of participants was never larger than 50 people because
the user testing takes time and because the technology used was
expensive and difficult to reproduce. Often it was impossible to
reproduce a large number of experiment because the TUI setup was
difficult and researchers were needed to build it. Audio d-touch
instead goes in the opposite direction. What is needed is just a
webcam and paper. The setup of the system is easy and working
(although still not very well explained to users) as we experimented
with hundreds of users.

Another common problem that appears from time to time is that


users don’t have enough time to get acquainted with the interface
since user studies rarely are longer than one hour. With our experi-
ment we tried to overcome this problem giving the interface for free
to the users and giving them complete freedom to use it when they
wanted and as long as they wanted. So learning could be at the
user’s speed and we could gather as much data as possible.
13 http://d-touch.org/forum

13
2.5 Large Scale User Studies

Previous work has been done using UGC collected on Internet to


gain understanding on products from informal comments on web-
sites. One example is the work done by Blythe et al. on YouTube
comments regarding the launch of the iPhone 3G [15]. In this paper
the grounded theory approach has been used to gather and orga-
nize hundreds of YouTube comment to draw conclusion on it. This
paper has been very inspiring for our work and encouraged us to
walk the same path as them and analyze the UGC collected on the
Internet.

Another work that has marked good points using the distributed
intelligence of Internet users is the study on Mechanical Turk [16].
In this study the Amazon Mechanical Turk14 was used to under-
stand if the shared intelligence of the Internet users is valuable as
the intelligence of experts. Obviously the result is not completely
satisfactory, but if the work is thought expressly for the Mechanical
Turk, good results could be obtained. This paper is useful for us to
understand that if searching for the right things, the tests with a
large number of users are a valuable source of informations.

A more traditional user study through Internet is Note to Self [17]


that uses Internet to recruit users to study the impact of a particular
GUI in note taking tasks. The interesting part is that the user study
is almost completely held through usage logs taken remotely as we
did in the study of Audio d-touch. Then some statistical analysis
has been done on data and a questionnaire was requested. The
difference from our research is that the participants were explicitly
contacted and it was offered a prize while in our study no one has
been directly asked to use the system and the interest was purely
personal.
14 Amazon Mechanical Turk is a marketplace for small works that require human intelligence. It’s complete

name is Mechanical Turk, Artificial artificial intelligence. Website at http://www.mturk.com

14
A previous large scale user study on a particular interface or
device is the Scent Field Trial [30] by Nokia. This study is peculiar
because it’s held internally in the Nokia Intranet using a specific
Nokia mobile phone, available to 800 employees. The basis of this
study is that the device was already available to the users so that
they didn’t have to spend money. For a novel TUI this approach
is impossible except if the participants decide to buy the product
studied as might happen in a mobile phones company or with other
kind of commercial products.

2.6 Grounded Theory

To analyze the data that we collected on Internet in form of com-


ments, emails, blog posts, etc. we used a method inspired by a the-
oretical framework called Grounded Theory as explained by Sharp
et al. in [31]. This approach is useful to analyze qualitative data in
a systematic way and get an insight of data collected especially in
cases where the amount of data is significant as we experienced with
Audio d-touch, collecting and categorizing around 600 sentences
about the application. The method is strongly tied, or “grounded”,
with analyzed data. In the beginning data are collected, in form of
sentences for example, and iteratively categorized with open codes.
Each time that a new theme is encountered, a new code is used to
categorize the sentence. During the process of categorization the re-
searcher have to focus on a specific theme and find relations with the
categories already created. This process continues until the codes
are saturated, meaning that no new codes are added when process-
ing the data. The next phase, the axial coding, is the reduction
of open codes through grouping and relating existing categories.
The axial codes will support strongly the theory that is growing
grounded from data. The last step is the selective coding, where
the axial codes are organized around a central theme, the grounded
theory, that can be better supported if more data are present. In

15
our research this last step has produced two different ideas grounded
in data and not a real theory that could comprehend everything.

16
Chapter 3

Audio d-touch

The d-touch instruments are tabletop tangible interfaces for real-


time musical composition and performance, presented for the first
time in 2003 by Costanza, Shelley and Robinson [32, 33].

3.1 Working Principle

The working principle is similar to other tangible interfaces already


presented: a camera is positioned over a table looking to an interac-
tive table (or briefly the board ) that is an A4 paper characterized by
four markers used as calibration for the system. Then the user can
place over the board the interactive blocks (also called in literature
as Bricks [6], Phicons [7, 3] or tokens [13]) that are tracked by the
camera and from the position and angle of the blocks the application
produce the sounds accordingly to the logic of the instrument.

Both the markers on the blocks and the calibration markers on


the board are d-touch markers [27] used by the application to track
the blocks and to convey some information to the user, since the
markers are designable they can be drawn to give some cue to the
user. An example is the marker used in the drum machine, visible in

17
Figure 2.8, that represent one hand and is associated with medium
volume of sample reproduction. Instead the marker in Figure 3.1,
displaying two hands, was used to play sample at an higher volume.
The markers were tested to be recognized at a resolution of 320 by
240 with the A4 board completely visible. The speed of the webcam
is useful only to get a low latency response from the instrument
but anything that could work at 15 frame per second or more is
good. With this low requirements the webcam can be very cheap
and the recognition algorithm was developed in 2003 and roughly
tested for performance in previous papers [25]. We can say that the
recognition algorithm was never the bottleneck of the application,
the speed was regulated always by the camera frame rate.

The application is capable to track


in real-time (at 25/30 FPS) the four
calibration markers, to adjust calibra-
tion of the system in case of small
movements of the camera or of the in-
teractive board and, at the same time,
each block in the board is tracked with
its x-y position, its angle and its type
accordingly to the marker printed on
it. With the A4 board, to have good Figure 3.1: d-touch drum machine
marker, with two hands
recognition and have the minimum size
of the blocks, we had designed them to be 2.5 by 3 cm so that simul-
taneously on the board could be present and actively used around
25 blocks. The size of the system can be scaled to any desired size.

From the audio point of view, both applications are loop-based


sequencers. They scan the interactive board from left to right and
then they loop it indefinitely. It’s like if there was a virtual cursor
looping the board and each time that the cursor crosses a block the
application plays the sound associated to it. The different arrange-
ment of blocks on the board creates different musical compositions
and different patterns. Each application, as we will see, can have

18
different behaviors and have different uses.

Both instruments were designed to be simple and easy to learn.


One rule that we followed to be coherent through the interface was
the object-action paradigm from the GUI literature [34]. The idea
is that using this paradigm it’s easier to avoid mode problems and
the object is always in the attention focus of the user. The mode
problems derive from the action-object paradigm, when the user has
to select an action, so enter in a mode, and then the object. This
approach is error prone since it’s very easy for the user to forget
the current mode and do actions that bring to unexpected results.
We translated this modeless approach in our TUIs using interactive
blocks as objects and using horizontal surfaces as actions. In other
words if the blocks are positioned in some areas of the board they
activate actions on themselves, we will explore the particular behav-
iors and mapping in later sections when the particular instruments
will be described.

3.2 The Software

Audio d-touch is C++


written with some ex-
ternal libraries. The
Computer Vision engine
is publicly available on
SourceForge1 developed
in C++ by Enrico Costanza
with the support of differ-
ent libraries to have the
cross platform support of
Figure 3.2: d-touch Sequencer board
webcams. The audio en-
gine is written using the C++ STK library developed at CCRMA2 .
1 http://sourceforge.net/projects/libdtouch/
2 http://ccrma.stanford.edu/software/stk/

19
The last part of the software, the remote logging engine and the
activation GUI has been written in C++ with Qt3 . All the libraries
are cross platform and we didn’t experienced particular problems
using them.

3.3 d-touch Drum Machine

The d-touch drum ma-


chine is the first in-
strument that we de-
signed and published on-
line. The aim of this in-
strument is to be very
simple and playful, just
some kind of exploration
of possibilities given by
TUIs. The board is very
Figure 3.3: d-touch drum machine board
simple with a single ac-
tive area where all the ac-
tion takes place. The area is divided in eleven rows and sixteen
columns: each row is a different sound, identified by a description
written on rows, and each column correspond to a different time of
play relatively to the loop. The design of the board can be seen in
Figure 3.3. The objects used in this interface are of two different
kind, identified by the markers in Figures 2.8 and 3.1. The first
marker is used to play a normal drum hit, while the second one
plays a louder hit, used to create patterns with accented hits. The
playback of the sample is activated when a block is placed in the
board and the virtual cue that loops the board passes the position
where the marker currently is.
3 http://qt.nokia.com/

20
3.4 d-touch Sequencer

The d-touch Sequencer is an evolution of the drum machine instru-


ment. The basic concept of sample reproduction is the same, as in
the looping sequencer. The difference is that the samples that are
reproduced by the d-touch sequencer are not the drum hits but the
user’s recorded samples associated on the fly to a block that, from
the recording moment, will trigger the new sample. The interface,
as in the drum machine, is essential. The main areas are: the play
area, divided horizontally in two equal parts, where the blocks will
be played back by the instrument, the record area is where blocks
trigger the action of recording and finally the store area where blocks
trigger the recording of the currently played sequence in the block,
allowing the construction of recursively complex sequences. In the
play area the vertical axis maps the volume of the block and the
horizontal axis, as in the drum machine, maps the relative time in
the looped sequence when the sample will be played. The record
and store areas are sized in manner that only one block at the time
could fit. Another difference from the drum machine is that the
rotation of the blocks influences the speed of reproduction of the
sample associated with the block. The speed of the sample can be
doubled or halved rotating the block clockwise or counterclockwise.

The d-touch Se-


quencer blocks are
of two different
types: the Sound
Container blocks vis-
ible in Figure 3.4a
and the Control (a) Sound Container (b) Start/End Control (c) Time Control
Block Block Block
blocks visible in
Figures 3.4b, 3.4c. Figure 3.4: d-touch Sequencer markers

There are 18 different sound blocks representing 18 different sounds,


more than one block of the same time can be present simultaneously

21
in the interface but they will reproduce the same sample. The col-
ored arrow, if put pointing at the right, as the play symbol, indicates
the normal speed of reproduction of the sample associated with the
block. The Start block has a twin called End block. The two blocks
can be used together or one at a time and they indicate that only
the subset of the board that they delimit will be looped to get the
sequence. These blocks can be used to break the constant duration
of loop, or in conjunction with the store action, to store in a block
a subset of the played sequence. The Time Control blocks are four,
differentiated by the different times (2 sec, 4 sec, 8 sec, 16 sec), and
are used to change the loop duration of the sequencer. The default
duration of the entire loop is 2 seconds, but can be changed to 16
seconds with 4 seconds steps. To use the Control blocks the user
has to place them in the board and they will change the behavior of
the instrument. If the Time Control blocks are covered or removed
the last time that has been set up is kept, while if the Start or End
blocks are removed the sequence is played completely and not only
in the last subset indicated.

3.5 Logging System

To enable the remote user study, a new part of the d-touch appli-
cations that sends usage logs to a remote server that stores them in
a database has been developed.

To have as accurate data as possible we developed also a registra-


tion and software activation mechanism. The users has to register
in our website to download the software and the same activation
data are requested from the d-touch instruments that, on startup,
interrogate the server to verify the account data. If everything is
correct the instrument starts, otherwise it shuts down. The logging
part in the application has been developed in C++ using the Qt
library, both for the TCP/IP implementation and for the GUI used

22
to input activation data. To limit the possibilities of shared ac-
counts between users a MAC address check on activation has been
implemented. When the user activates the software for the first
time it sends to the server an hash of the username, the password
and the MAC address of the machine, then the hash is stored. The
next time that the application is started the hash is created and
sent to the server. If it’s different another activation is required. So
if someone shares an account with someone else each time that he
starts the application he gets annoyed by the activation. We hoped
that this behavior could push the users to register and use their
own account in order to preserve usage patterns of single users, as
the time span between the first play and the last play, that we’ll
see in the next section. With these solutions we noticed, from IP
addresses belonging to different countries, only one case of shared
account and we removed it from our database.

The registration phase on the web has been useful also to gather
some background information for the users. We asked to users their
age, sex, profession, musical knowledge and practice, TUI knowl-
edge and practice. These data were later used to group logged
data and try to understand how people with different background
could differently interact with the d-touch instruments. With the
applications, as with the registration, we decided to collect only the
essential data interferring as little as possible with the users’ pri-
vacy. The application registers the marker and calibration position
and angle, the timestamp, the user and the IP address of the con-
nection. Every 25 frames the logging part of the application sends
data to the remote server that stores it along with the user iden-
tifier. We never gathered pictures from the webcam to reconstruct
the usage. For the sequencer we decided to log also the registered
sounds, otherwise we could not understand what the user was do-
ing but we thought that this was not an invasive practice since the
audio should only be recorded during playing. In fact we never reg-
istered personal conversations. However we always stated clearly
and as better as possible that the application was free because we

23
were doing a user study and that we were registering user actions
during usage. Also the fact that the application could work only
with an Internet connection raised attention and understanding on
this point. We received a lot of emails and messages of user com-
plaining on the need of Internet connection, but only once a user
asked to be removed from our database.

From the server point of view, the logging was done modifying
an existing system based on TikiWiki4 a PHP based content man-
agement system. A PHP script has been developed to get the data
arriving from the applications in form of HTTP POST calls. The
data is gathered, associated with the user that has sent the data
and then stored in a MySql database. The system has revealed it-
self fairly robust since it gathered simultaneously data from multiple
users apparently without losing data.

3.6 Distribution

The low cost and easy to set-up nature of the d-touch instruments
were design goals from the beginning of the development. No sol-
dering, particular electronic components nor special handcraft is
needed, just a webcam (or a consumer grade DV camera), a com-
puter with Windows, Mac OS X or Linux, an audio card and an
ink-jet printer are needed. The complete package with software,
images with blocks and board to print is downloadable in less than
10 Mega Bytes.

The cross-platform compatibility of the software was needed for


the nature of the project, often music makers or artists are Mac users
and then the Linux version was done with almost no change, but
never publicly released, just given to people that explicitly asked for
it. The fact of distributing the software out of the lab on multiple
4 http://tikiwiki.org/

24
unknown hardware and software configurations has revealed itself a
non-trivial technical challenge. The system has been heavily tested
and some deep software bug has been found. We can say that the
public release of the software forced us to clean and review a large
part of the audio and computer vision code of d-touch. Then for
the first time we tested the cross-platform capabilities on OS X
and finally we had to prepare installation packages with external
libraries for Windows and OS X.

To have an easy to
setup tangible interface
we prepared the ready to
cut blocks, visible in Fig-
ure 3.5, the instructions
to build a webcam stand
with the paper outline to
be glued on cardboard
and the image to print
the board. We provided
also the plain markers for
the lazy users, suggest-
ing an easy and quick way
to build the d-touch sys-
tem: the webcam could
be hanged to a lamp, the Figure 3.5: d-touch marker on a ready to build block
plain markers could be
glued on objects like small chocolate bars or nuts and we demon-
strated it with a video that we posted on YouTube5 where we used
chocolate and walnuts as marker support. Otherwise we suggested
to fill the blocks with lentils to give some weight to the blocks and
have a better control of them6 . We did these videos mainly to con-
vey the easy to build and low cost materials needed for the project
as well as to get some media attention on the project.
5 http://www.youtube.com/watch?v=lCv0TvnVUHg
6 http://www.youtube.com/watch?v=AmGP3eCOF5A

25
3.7 Diary of the Launch

Since its beginning in


2003 the Audio d-touch
instruments have been in-
formally tested by a cer-
tain number of people,
including real musicians.
Given the easy to build
nature of the project
the situations were nu-
merous: laboratory open
houses, friends from our
laptop and also live on Figure 3.6: The complete d-touch setup
stage. The responses
have always been enthusiastic, people loved the playfulness of the
system, the simplicity and the low cost nature. Also in the musical
field it received good comments. The professional composer and
cellist Giovanni Sollima used the d-touch Sequencer in some con-
cert starting from 20067 . The musicians appreciated the computer
interaction without the need of a screen and the physical approach
to music that reminds the use of audio effects pedals. Some d-touch
videos were posted on YouTube at the end of 2006 and in 2 years,
without any promotion, they received more than 30000 views over-
all.

From this informal positive feedback we were encouraged to or-


ganize a large scale user study, since the appealing audience looked
potentially wide enough, thanks also to the success of Reactable [11].
The fact that the instruments were low cost and completely down-
loadable was the key to push us towards the online publishing. The
ongoing large scale user studies done in the HCI community [15, 16,
17] convinced us to go for the remote logging and usage analysis,
7 http://www.neural.it/art/2008/09/physical sequencer sequencing.phtml

26
with the collection of informal comments.

The first step that we did was the creation of the d-touch.org
website to promote and publish the applications. On the site we
made available for download the Windows and OS X applications,
together with the PDF file of blocks and board, after a registration
procedure. In a first phase we made publicly available only the
drum machine while we were completing also the revision of the
sequencer. To explain to people how d-touch works and to promote
the diffusion we did a short video that we published on YouTube8
and we featured it in the d-touch.org homepage.

We planned the release day to be on June 28 2009, so the days


before we prepared the online material to promote d-touch. We
thought that the Do-It-Yourself part of the project might interest
someone and so we posted an entry on the popular instructables.com
website, with a complete set of photos and instructions to build it,
with a link to our site to download the software. We used Twitter9 ,
the popular micro-blogging platform, to communicate with other
people and on the release day we sent emails to popular blogs telling
about the new d-touch project gone online. In a few days we were
featured in homepage of instructables.com, getting more than 5000
view of the instructions, the YouTube video went fast over 20000
views and we started to get followed on Twitter. Then d-touch
started to appear in blogs, until Engadget.com and HackADay.com
published a post on our system and we received a huge response.
By July 5, a week after the launch, we had 671 users registered
to our website, 208 started the application and 112 used d-touch
successfully (more than 1 minute of usage).

During the following 6 weeks, until mid August, we released 4


new updates of the drum machine, fixing bugs that we discovered
only thanks to the incredible feedback of our users, while in the
meantime we continued working on the sequencer. The website
8 http://www.youtube.com/watch?v=AmGP3eCOF5A
9 http://www.twitter.com

27
continued to be viewed and totally we received 25000 visits and the
project was featured in more than 30 blogs including 2 hands-on
original reviews10 .

On August 17 we officially launched the sequencer, a new version


of the drum machine and we renewed all the promotional material.
A new video was done11 , with new footage and modified accordingly
to the comments we received on the first one. We prepared high
quality photos12 and a press kit to contact better as we could the
blogs and the magazines. Probably due to the second launch, the
project gained less attention than the first time and in the same
time span we received 7800 visits on our website roughly a half than
the first time. Surprisingly the YouTube videos had more attention
than the actual website, as if the talking about the system was more
interesting than the system itself. Despite the less visibility, after
the second launch we had more time to gather data to analyze, since
we decided to stop on December 15 2009. The numbers after the
second launch are still interesting. The registered users are 1252,
389 tried the interface at least once and 273 for more than one
minute, playing a total of 199 hours.

10 http://d-touch.org/coverage/
11 http://www.youtube.com/watch?v=cKd8NXWwvKI
12 http://d-touch.org/audio/material/

28
Chapter 4

Logs collection and analysis

In this chapter we will talk about the interaction logs gathered by


the system explained in Section 3.5. The collection and analysis of
the User Generated Content collected through Internet sites will be
covered in the next chapter.

4.1 Data Collection

The data analyzed in the next section is what we collected from


users that registered from August 17 to December 15. We could not
take into consideration also previous data because we discovered a
bug that made useless what we collected before. All the following
analysis will be done on the 4 months of correct data that are stored
in a database bigger than 3 GB. Each entry in the database is char-
acterized by username, timestamp, time relative to the start of the
session for each frame, position and angle of markers for each frame.
We will cover two different approaches that we had to data analy-
sis. The first is a statistical approach that brought us quantitative
results useful for future design guidelines. The second approach is
qualitative towards the interaction videos that we reproduced. The
latter is similar to direct observation of users and it has been used

29
to try to understand the actions taken by the people. The mixture
of the two approaches, with the the UGC analysis done in the next
chapter will help drawing some conclusion.

4.2 Data Analysis

In this section, interac-


tion logs analysis will be
described. To elaborate
statistics on data, and
draw graphs, we used
Python1 scripts with the
Matplotlib2 library. An-
other sort of analysis has
been done through the
playback of logs. We
wrote a Python log player Figure 4.1: Example of frame produced by the log player
that sent data to the
drum machine, exactly as the computer vision engine does during
normal usage, so we could play and record all the sounds that users
produced. In parallel we draw all the frames that we received with
another Python script so that we ended up with a video track and
an audio track for each session played. We combined them together
and we could playback all the logs that we have collected in form
of videos. One example frame can be seen in Figure 4.1.

4.2.1 Quantitative Log Analysis

The quantitative analysis of logs started with the self-reported reg-


istration data about demographics. We report percentages from
1 http://python.org/
2 http://matplotlib.sourceforge.net/

30
Sequencer Drum Machine Overall
Avg session length (min.) 8.75, (σ = 10.12) 6.75, (σ = 8.05) 7.34, (σ = 8.76)
Avg no. sessions per user 3.33, (σ = 3.13) 5.45, (σ = 6.35) 5.55, (σ = 6.34)
Max no. sessions per user 15 30 39
Avg blocks in session 1.80, (σ = 1.71) 3.06, (σ = 3.75) 2.69, (σ = 3.33)
Max blocks in session 5.49 7.11 6.64
Sessions with 8 blocks or more 96, (8.35%) 318, (27.70%) 414, (25.43%)
No. of sessions 479, (29.42%) 1149, (70.58%) 1628, (100%)
Minutes of usage 4193.25, (35.10%) 7752.62, (64.90%) 11945.87, (100%)

Table 4.1: Audio d-touch usage data, gathered from the interaction logs. In brackets the
standard deviation of the average values.

the 273 users that successfully started Audio d-touch and used it
for more than one minute, except 7 people that filled forms with
useless data. 27% of our users were under 20 years of age, 73%
under 30 and 94% under 40. Females were very few, only the 2% of
users. Few users reported previous knowledge with TUIs and even
less reported previous experience with them.

Going further with quantitative analysis of data we tried to un-


derstand the engagement of people towards this kind of tangible
interfaces. We collected statistical data from sessions, like the aver-
age session duration, the number of sessions per user, the number of
blocks per session. All the data gathered is summed up in Table 4.1.

We used the boxplot graph to understand better the overall data


and to make some comparison between sequencer and drum machine
usage. From the data in boxplots shown in Figure 4.2 and from the
Table 4.1 we can get some insights on user preferences. Even if the
drum machine has been used much more than the sequencer, in the
same time span, we can see that the sequencer has been used for
longer sessions than the drum machine in the large majority of cases.
For the number of blocks it’s controversial. The mean is higher for
the drum machine as we see in Table 4.1, but from the Figure 4.2a
we see that the majority of sequencer sessions have more blocks that
the majority of drum machine sessions. In the drum machine we
have a high number of outliars, probably due to the “tests” done

31
(a) Boxplot of maximum number of blocks.

(b) Boxplot of session length.

Figure 4.2: Comparison between instruments. The boxplot used in these pictures indicates:
with the box the lower and upper quartile (25% to 75%), in red the median and the whiskers
are 1.5 times 50%. Usually in the box will be the majority of data, with the whiskers we see
the minimum and the maximum and with blue crosses are marked the outliars.

32
with the complete A4 paper full of markers that we provided in
PDF. We can say that on average the sequencer sessions showed
more blocks, probably because of the two available tracks to place
markers and to the more interesting instrument that engages more
the users.

Generally usage of d-
touch was infrequent: 21%
of users interacted with
d-touch over a period
longer than 2 days and
11% for more than one
week. We can also see,
from the histogram in
Figure 4.3, that the us-
Figure 4.3: Histogram of number of sessions in a time
age was compressed in slot. The bins are half an hour big in the global image
the large majority in less and 2 minutes long in the zoomed part.
than two hours. All the
quantitative analysis is useful to understand better the pattern of
usage that we will later see and the UGC that will be presented in
the next chapter. In the final discussion, we will review this data
in conjunction with the others to get a complete picture over the
experiment.

Before doing the video analysis we had to find some criteria to


lead it. We could not watch the entire video production because it
would be too time consuming and too less informative as the starting
sessions were often similar. So we decided to do a quantitative
analysis before, to watch only a meaningful subset of all the sessions.

To select a user that may be interesting we plotted the number


of blocks present in a frame on time, that resulted in graphs like
the Figure 4.4a. We selected users that had sessions with a number
of blocks greater than 5 or if some particular pattern was evident
from the general usage. Then if a user had a lot of sessions, we

33
(a) Usage graph. The number of blocks in the (b) Usage plot. This is a heat map presenting
interface are mapped on the y-axis while the time where markers has been positioned during a session
is mapped on the x-axis. to understand if some movement has happened.

Figure 4.4: Selection principles. The video analysis has been done selecting videos that pre-
sented interesting patterns in above pictures.

drew heat maps of usage, based on where the blocks were more
present during a session. The plotting resulted in images like the
Figure 4.4b. The pixels were more colored if the block was present
for a relevant number of second in a certain position. From this
plots we can see if a user has moved blocks during a session or if the
session was all equal because he forgot the instrument on an he went
away. With this previous analysis we found interesting sessions and
then we produced the videos regarding the history of these users to
see the learning curve and the possibly interesting results. Using
this selection approach we managed to analyze around 24 hours of
videos, instead of the 90 present at the time of video generation and
we were satisfied with the quality of the sessions. The results of the
video analysis will be presented in the next section.

4.2.2 Qualitative Video Analysis

Since we had not analyzed all the videos but only a part of them
we cannot give percentages on what we have seen, also because we
report here mostly observed patterns, so we will give trends of usage
that might be useful for the development of d-touch and other TUIs
with similar setup.

From the people that managed to start successfully d-touch we

34
saw that everyone managed to produce a basic rhythm with few
blocks or that they enjoyed themselves with one or two blocks try-
ing to understand how the system is working. We will refer to this
latter kind of behavior calling it exploration of the interface. Al-
most everyone started the first time the application and explored
its possibilities moving one block in front of the camera and try-
ing to calibrate the system showing all the markers of the board
to the camera. Only a few minority started immediately with a
rhythm creation, probably users with a background in tangible in-
terfaces. The exploration of the interface might also happen with
the uncut paper with markers that we provided on the website. This
behavior leads to a high number of markers seen (around 30) for a
small amount of time and an indistinct noise. This kind of pattern
was observed multiple time as it is the easiest way to test if the
instrument is able to produce some sound only having printed a
paper. We have observed sometime that after this exploration the
board remains empty and markers appear slowly in the interface,
as probably the user were cutting markers and placing them in the
instrument as soon as they were cutted. These exploration patterns
are visible only in the firsts sessions of the users. Later sessions
start with the already calibrated setup and sometimes with already
some blocks in the interface, as if the setup was left untouched from
the previous experience in a fixed setup.

Beside these exploration patterns and the basic rhythm gener-


ation, as we saw in Table 4.1, few sessions contained a significant
number of blocks allowing the creation of a complex sequence. Any-
way these sessions were often long and musically interesting, demon-
strating that the users grasped well the concept of the instruments
and that good results are achievable.

From observations we noticed another recurring phenomenon, the


markers where recognized intermittently indicating difficult lighting
conditions. This worsen the user experience because the blocks lay-
out seen by the users doesn’t correspond to the block positions seen

35
by the applications, leading to incorrect audio generation. Given
that the algorithm has been tested for five years and a tracking
filter actively smooth this behavior, we suppose that the lighting
conditions were really poor in these situations. After having ob-
served this phenomenon, we realized that we didn’t explain clearly
how to check if the recognition was happening correctly, beside a
small sign on markers visible on screen. Related to this lack we
realized also that we didn’t provide a sign on the GUI to indicate
that the calibration of the system was acting correctly.

In general we observed rhythmical patterns if the sessions were


correctly calibrated and with good lighting conditions. In the other
cases users could not get the correct responses from the system and
they were able only to explore basically the system. Even if they
tried with a significant number of markers, since the sound was
not coherent with the tangible setup, the users quickly moved from
music making, to audio exploration and then they abandoned the
instruments.

36
Chapter 5

User Generated Content:


collection and analysis

The focus of our research has been the observation of tangible in-
terface users in their everyday environments. Through interaction
logs we gathered information on their actual usage, but probably
the most interesting informations arrived from their informal com-
ments and from the videos and photos they shared with us. In this
chapter this type of content will be analyzed.

5.1 Data Collection

Around the d-touch instruments raised a lot of interest through the


Internet promotion. We observed and gathered all data that we
were able to find in the period from the pre-launch on June 28 2009
until September 9 2009. In this time we collected more than 120
emails, more than 330 posts on the d-touch.org forum, more than
50 blog posts, hundreds of comments on social web sites and more
than 220 Twitter posts. Through Twitter and email we asked our
users to share with us photos and videos of their setup to see how
they were using the system in order to have more information on the

37
real usage scenario. Six videos were posted on YouTube1 , one photo
on Flickr2 , and one photo was sent to us via email. We found also
two hands-on reviews of the drum machine which included various
photos of the used setup. The contents that we have found were
from all around the world, mostly from USA, Germany, UK, Chile,
Brazil, Japan. We tracked the most interesting ones in our website3 .

5.2 Data Analysis

All the data that we have col-


lected, as explained before, has
been analyzed with an approach
inspired to the grounded the-
ory [31, 15], especially for text-
based content, while the videos
and photos have been analyzed
to see how the setup was built
and if some recurring usage pat-
tern emerged. In two cases we
saw interesting examples of user
appropriation of d-touch.

5.2.1 User Generated Content:


Text

To analyze the text-based con- Figure


Flickr
5.1: A user setup publicly posted on

tents we divided them at the


sentence level and we catego-
rized them with the open coding method. The open coding phase
consists in categorizing each sentence in an appropriate group and
1 http://www.youtube.com/channel/dtouchshow
2 http://www.flickr.com/photos/85596380@N00/3689219303/
3 http://d-touch.org/coverage

38
Axial Codes Sentences
Actual Usage Feedback 305
Technical 202
Improve and extend 40
Audio 25
Physicality of the interface 21
Applications and real use 10
Field trial related 7
Generic Comments 286
Viral sharing 153
First impressions 118
Personal data sharing 15
Total 591

Table 5.1: Axial Coding and the two main groups.

create a new group if the sentence doesn’t fit a previously created


one. In this way from a large amount of data it’s possible to get
a significantly smaller number of categories that describe the com-
plexity of all the sentences. The next step is called the axial coding
and consists in grouping previously created categories to have a
small amount of groups that can be described.

Using this grounded approach we collected 591 sentences and we


grouped them with 50 open codes. With the axial coding we man-
aged to create 9 broader categories: physicality of the interface, au-
dio, technical, improve and extend, field trial related, applications
and real use, first impressions, personal data sharing, viral shar-
ing. To present better the results we divided them in two groups:
actual usage feedback, that refers to real usage, real experience on
d-touch and generic comments, that contains all the impressions on
the system, not grounded with real usage experience, based only on
what has been seen on the web. A summary of the categories can
be found in Table 5.1, while the contents of each category will be
explored later.

39
Actual Usage Feedback

Almost every comment in this group was directly sent to us via


email or on the d-touch.org forum. Some exception come from the
hands-on reviews and from some Twitter post sent between users.

Technical The large majority of technical messages are bug reports


or problems related to the setup not working. We received also
questions on software architecture and suggestions. We tried to
reply as much as possible to this kind of questions to help our users
to use d-touch. We noticed that people interacting with us for bug
reporting, showed a good technical competence and helped us to
solve problems that in a normal laboratory condition would never
be found.

Improve and extend With d-touch we targeted as a potential audience


the Do-It-Yourself people, the ones that enjoy to spend time build-
ing things and improving existing objects. For this kind of people,
the time spent building the system is the more interesting part of
the project as we can see with some comments:

The system is made mostly of paper and cardboard; soft-


ware and instructions of how to build the system are given
on the website, in pure style Do-It-Yourself (probably the
most appealing part of the project, from the user perspec-
tive).

or

I spent a happy half an hour cutting out the shapes and


putting the little boxes together.

40
and

Really enjoyed the glue, lentil and paper time though, re-
minds me how good it can be to get down and dirty with
materials.

Instead of discouraging people the DIY side of the project pushed


people to suggest more complex and advanced setups, as to demon-
strate the amusement in this:

Currently I made a stand with 2 pvc pipes, 1 pvc L joint,


and a metal flange for the base, and an adapter to go from
the pvc pipe to the metal flange. I then threaded the cam-
era cord through that, cut notches out at the top to secure
the webcam ’clip’, and the height is just enough to accom-
modate the board being on a 8.5 by 11 inch paper board.
The fiducials are just taped onto 1x1x1 inch cubes I had
from a board game.

We were worried by the fact that people could be discouraged by


the time needed in building the interface but we never received
comments on that point. We saw only someone, short in time, that
tries to convince himself to do the project as soon as possible:

I’ll get back to this on the weekend. Pretty much blocked


for now - a slightly ironic turn of phrase given the nature
of the project.

Audio The audio part of the project received a good number of


comments. Someone was about the easiness and playfulness of the
project, suggesting enjoyment from the users, as:

41
Everyone in this house has now put together a radio-worthy
beat by pushing little scraps of paper around under a web-
cam.

Other people suggested to use d-touch in schools to teach music


or that they used the instruments to play with their children.

The audio part of d-touch received the largest part of the criti-
cism. The audio synthesis part was too limited and without enough
features for a long engagement and for real musical use. Several
users requested the ability of sending MIDI or OSC signal out of
the drum machine, or the ability to load custom samples. One post
that we received on our forum is explicative of these problems for
musicians:

it’s just a short-time toy because: 1) can’t send midi 2)


can’t even load custom samples

or again from the forum:

I Could probably live without the Midi Sync if there was a


BPM selector in the D-touch program in the video window.

We collected sentences that externalized


Physicality of the interface
the understanding of the tangible interface and appreciation of a
tangible approach in music creation. One example might be:

When I showed my good bands and songwriters the setup,


they loved the realness and ability to touch and move some-
thing real to make the sound

or

42
Software doesn’t have to mean virtualizing everything and
letting go of physical objects. On the contrary, it can create
all sorts of imaginative, new ways of mapping musical ideas
to the physical world. And that’s how we wind up with a
walnut drum sequencer.

Others appreciated the ease of use of the interface:

The Audio d-touch is a collection of tools and applications


which allow you to compose music in real time and the
interfaces are extremely user friendly

and the simple design of the projects:

It’s easy and it’s really fun. All other examples of similar
technologies involved being a computer genius and the step
by step instructions are just pain awesome for everyone, this
is the right way to do this kind of things, by making them
available and enjoyable to everyone.

We received some proposals from users on ed-


Application and real use
ucational projects and ideas on how to market d-touch in the real
world. This shows that users were interested in real world applica-
tions of d-touch and they proposed real usage scenarios related to
their personal projects and ideas.

Regarding our user study we received some positive


Field trial related
and negative comments. Some users complained about the registra-
tion questions being to invasive or about the logging system, but
this comments were very few (only two in total). The majority
of issues were related to the necessity of Internet connection when
playing the instrument that limited possible use on stage of d-touch.

43
This was often accepted well as the applications were freely down-
loadable and the setup was really cheap. We received also proposals
of small payments to remove the logging system to allow free usage
of the applications.

Generic Comments

In this section we’ll analyze comments not directly related to usage


experience but on the informations related to the d-touch appli-
cations. The media promotion provoked emotional reactions from
users before, or instead of, actual usage of the system. The follow-
ing material is less interesting than the one presented above, but it
shows that tangible interfaces can attract a lot of media attention,
especially if they are freely available for download.

We used the Twitter micro-blogging platform


Viral sharing (Twitter)
to promote and publicize d-touch. As a consequence many Twitter
users posted content related to d-touch and their feelings about it.
Often the interest was expressed with just a link to our website or
to their blog post related to d-touch. In any case this viral form of
communication gathered a substantial number of users that actually
tried the system.

Often d-touch was surrounded by a wow-effect, es-


First impressions
pecially at the first launch. This provoked a series of emotional
comments as: “Sci-Fi??? No more! You can actually build it!” or
“This is an awesome UI.” These impressions are representative of
the common feelings of the people that saw videos or photos about
d-touch, probably even before trying it.

44
Figure 5.2: This is the first video that
we received. It’s done with an old ver-
sion of the drum machine and the user
is hearing wrong sounds caused by a
bug in the audio card settings, that
we solved with later versions. We can
see that the user has a very small
space to place the board and it’s in
the middle of two different computers.
The markers have been done folding
the paper blocks that we proposed on
our website. This is a typical case of
technology interested user, since the
music is not reproduced correctly but
no comments are done on this.

The personal data sharing refers to the fact that


Personal data sharing
our users often shared with us and between themselves personal in-
formations like age, location or other data. We find this interesting
in the measure that around a project might quickly gather a com-
munity of users, that can become affectionate to it and help the
development. We discovered this especially in our forum that was
heavily used during the launch phase and until we provided news,
updates to the software or replies to specific questions.

5.2.2 User Generated Content: Videos and Photos

The photos and the videos that we received were probably the most
informative part of our study, since we could see where the user
setup the system and how they used it. To analyze them we initially
created a set of questions to answer for each video and then we
observed all of them trying to answer our questions, that were:
What was used for the blocks? What kind of stand? Near the
computer? What kind of room? Are they on table? Audio setup?
Operating system? A summary of the analysis and other material
can be found on Appendix C.

On the DIY side, we observed that some people preferred to stay


as easy as possible, without folding the 3D paper blocks, just using

45
Figure 5.3: This user built wooden
blocks to have a polished setup and
we can also hear in the video a very
good rhythm created in few seconds.
The ability and control of the inter-
face are good as if he practiced a bit
before doing the video. He is inter-
ested more in the musical part, since
while speaking in the video he always
refers to the drum machine and never
to the tangible interface.

Figure 5.4: This is probably the most


artistic video done on d-touch. The
purpose was to promote a creativ-
ity day in Aosta, Italy, where people
could interact with novel interfaces
in a new multimedia library. In the
video is shown all the building process
and at the end a bit of playing. As
discussed, strong lighting conditions
lead sometimes to bad recognition, as
happened during this video shooting,
leading to a inconsistent music pro-
duction. In this video we see also se-
quencer markers, the only appearance
up to now.

Figure 5.5: In this video we see two


people using the drum machine in a
small place, with a quick setup. The
markers are cut from paper without
folding blocks, or even used in strips.
The purpose of the playing is surely
for fun as we see the board put in a
wrong way, the webcam placed on a
music stand and a light that makes
the interaction even more awkward.
Nevertheless the two users have fun
playing, they don’t even try to make a
rhythm and at the end they congrat-
ulate with us, as if the only purpose
of the drum machine was to be a toy.

46
the flat markers printed on paper. Sometimes they even used the
uncut strip of markers to construct repetitive patterns or to test
quickly the interface, as we also observed from the interaction logs.
On the opposite end the DIY lovers built a polished setup with
wooden blocks and professional tripods. We never saw the card-
board stand that we provided as a PDF, users preferred to hung
the webcam to shelves or to camera tripods. In some cases, the
interactive board was raised, on books, to bring it closer to the
camera.

We always observed d-touch setups on desks, they never appeared


on casual surfaces like sofas, floors or others. The desks were often
cluttered with other things making the available space really small
and the movements to play were often awkward and unnatural.
Sometimes the board was placed in a wrong position respect to the
users, because the camera was at the side of the table or the only
available space were not in the right position. So that the board
resulted diagonally placed or not in front of the user. Obviously
the results in these cases were not perfect, but we saw users mostly
amusing or technically testing the system more than trying to get
a real music out of their playing.

The usage settings were very different, we saw small rooms, bed-
rooms, individual offices, a cubicle and a music studio. The hard-
ware and operating system used was varied: laptops, desktop com-
puter, Windows and OS X. Also audio systems were various, from
laptop speakers to professional hi-fi systems.

Two videos were of particular interest since we observed two cases


of users appropriating the technology in radical ways. The first
example is in a university setting, were a user printed the board
around 8 times bigger (as she herself explained) than the standard
A4 proposed by us. Then she affixed the board vertically on a
magnetic whiteboard using office magnets. The markers were also
glued on office magnets and used on the board as in the original

47
(a) d-touch drum machine put vertical and used col- (b) The tangible interface subverted and made virtual.
laboratively.

Figure 5.6: Two video documented cases of user appropriation.

idea. Then the video shows her explaining the complete setup, how
she was using the classroom hardware to complete the setup and it
finishes with two fellow students using collaboratively the setup to
build a drum sequence.

The other interesting video was, if possible, even a more radical


revolution. The user managed to use d-touch without using pa-
per! He pointed the webcam to his computer screen where, using a
graphic design software (similar to Adobe Illustrator), he displayed
an interactive board with markers drawn on it. He then moved
markers with the standard computer mouse, completely subverting
the original idea of a tangible interface!

48
Chapter 6

Discussion

We have analyzed so far the data separately, now we’ll try to get
a comprehensive view of all the data and get some inspirations for
future works.

From the analysis of reconstructed interaction videos and from


the user sent videos we can say that the simple set-up works. The
initial idea of spreading an experimental technology to a large user
base through a low cost system has been demonstrated successful.
d-touch has been used in different settings with different hardware
setups showing a good degree of versatility. The basic concept of a
board with interactive blocks over it, pointed by a webcam is clear,
we didn’t receive questions or comments about it. Anyway from
the reconstructed video analysis we have seen that the recognition
algorithm is not robust enough in uncontrolled lighting conditions
and that the calibration was not always achieved. Viewing these
common problems, we realized that we gave almost no feedback
about calibration and recognition on the application GUI and the
instructions that we delivered with the system were probably not
enough.

The largest part of the users explored the interface and produced
basic rhythms. The creation of advanced patterns was infrequent

49
as probably the applications were seen as toys more than musical
instruments, as someone pointed out with their comments. The lack
of important features for musicians has been widely communicated
by the users as we reported previously.

The trends in interaction logs, summarized in Table 4.1, support


this supposition. The more playful drum machine has been used
more than the sequencer (65% vs. 35%) and the time spent on the
interfaces spans mostly two days. The 21% of our users continued to
use the instruments for longer times. The signals that more blocks
were used in the sequencer and that sessions were longer on this in-
terface suggest that, thanks to the record and store functions, more
complex sequences were possible, being potentially more interesting
in a longer time span.

From emails and videos sent by our users we encountered several


examples of user appropriation. Two persons wanted to use d-touch
in their classes, to teach music to young children ant to teach game
design in another case. Someone changed the Audio d-touch setup
to better fit their purpose, putting it vertical, to allow collaborative
work and teaching, or scaling it much bigger to artistic purpose
in a festival setting. Someone tried to enhance the block design
building wooden blocks and proposing different weight to represent
different sounds in another case. Another one managed to build a
virtual tangible interface using the computer monitor as the board
and moving markers with the mouse in a graphic application. From
these examples we see that user grasped the idea of tangible interface
and used it to fit their needs in the most different settings.

We observed a high number of people registering to the website


and a significantly smaller number of users successfully trying out
the interface. This may be due in part because of the emotional
interest towards the applications and the lack of motivation to build
it or try it. In other cases users may have encountered problems with
their hardware configuration using d-touch. We suspect that the

50
open source third party libraries that we used to support webcams
and audio cards under different operating systems, together with our
software may create some incompatibility. Moreover we discovered
bugs related to hardware configurations only delivering the software
to our users, that helped us a lot in discovering and fixing them.

Generally the d-touch users are medium to advanced level com-


puter users, DIY lovers and musicians. We heard always good critics
on the audio part of the application and good reports about techni-
cal problems. Instead no one had specific negative comments about
the tangible interface. Overall we received few comments about it,
even if it’s a novel interface for musical instruments and from regis-
tration data we saw that few of them reported previous interactions
with tangible interfaces. Nevertheless from interaction logs, videos
sent by users and comments we observed a rapid learning curve and
no problems approaching the interface.

To sum up, the web distribution and user study of a tangible in-
terface has never been done before the d-touch experiment and at
the end we can say that it has been a success. Even if the applica-
tions were perceived mostly as toys than proper musical instruments
we received a huge media attention and users were really involved in
the project and helped us a lot to improve the applications. A large
number of users managed to use the instruments, even in presence of
technical difficulties, and they explored the possibilities of tangible
interfaces. About the interface we received very few comments and
all of them were positive and encouraging. We interpret this lack of
comments as an evidence that users found the interface “obvious”
to use, even if very few of them reported previous experiences with
TUIs. Maybe this is due to previous informations about TUIs gath-
ered on medias that in the last years reported a lot of works in this
field with a lot of emphasis as they did with us. Finally the cases
of user appropriation are signals of strong interest and advanced
understanding of this novel type of interface.

51
Chapter 7

Future Work

The top priority work for future development is a visual feedback


on the GUI to help users getting a correct calibration and good
recognition. We think that this can help users to get generally a
better experience removing the annoying fact that what you see
in the interface doesn’t correspond to what you hear produced by
the software. We could add on the GUI some very visible message
saying that the system is uncalibrated and the information to put
the four calibration markers visible from the camera to have the
correct audio.

We thought to add also some simple parameter setting of the


computer vision algorithm to improve recognition based on the user
current lighting setting. A possible improvement could be to sim-
plify the current adaptive thresholding method with a fixed thresh-
old method that can be tuned with an on-screen slider by the users,
accordingly to their lighting setup. We observed that the current
algorithm is not perfect with very strong lights, a situation that
may happen on stage for example.

We are also looking into mechanisms that would allow users to


share their compositions with the community of users through the
website, to foster the development of a user base and to encourage

52
them to play more. The compositions could be shared just in form
of images that could be downloaded and printed as a musical score.

The MIDI, OSC or other output signal is not interesting from the
research point of view because it would lead to uncontrollable usage
that could not remotely studied. In any case, it’s already possible
to create this kind of interface with the d-touch recognition library,
freely downloadable from the web in open source1 .

To improve the remote user study we should resolve a bug in the


logging of calibration markers. Now we are not able to understand
if an image is correctly calibrated or not, because we are not logging
correctly the four calibration markers. This could be a useful addi-
tion for next user studies in synergy with the GUIs modifications.

Now with Adobe Alchemy2 it’s possible to port the entire recog-
nition library to Adobe Flash, allowing to reduce problems in hard-
ware compatibility and to make faster the setup of the system. If the
performance loss due to the Flash porting shows to be sustainable
from the performance point of view, this kind of approach would
make d-touch available to an even broader audience, directly from
their Internet browser.

In the future we are interested to mix a visual feedback directly


on the tangible interface to enhance the possibilities of the in-
struments. This could be implemented with small low cost pico-
projectors that could be hung together with the webcam thanks to
their light weight. Another possible solution could be the use of
a LCD monitor as the base of the interface. This could be prob-
lematic for the constant backlight of the screen that might interfere
with the recognition algorithm.

1 http://sourceforge.net/projects/libdtouch/
2 http://labs.adobe.com/technologies/alchemy/

53
Chapter 8

Conclusions

The work of this thesis focused on the public release of a tangi-


ble interface for musical performance and composition to allow a
large scale user study in the field of human-computer interaction.
We started by creating a logging system to record user movements
through Internet, then we packaged the software and released it
online. We tried to create interest in the Internet community by
publishing videos and content in sites like YouTube and Twitter.
We helped and we were helped by our users to get the better ap-
plications as possible and to get them work on the broadest type
of hardware. Then we developed ways to view the interaction logs
gathered through the months of usage and we collected all the ma-
terial that we have found online regarding d-touch after the public
online launch of the musical applications.

The contribution of this thesis to HCI is on the novel user study


that we have used, even if it has to be fine tuned, has given good re-
sults with a very low cost. We cannot achieve the level of an in-situ
observation, but still a lot of informations can be retrieved. The
time and money spent in these remote studies are dramatically less
than what is needed for an in-situ study, making possible to study
more products even with a small team and with small economical
availability. Probably the best results can be obtained mixing the

54
two different approaches, to gain significant statistical numbers us-
ing the large user base and, on the other side, having the direct
exchange with users might be very useful to pose direct questions
and to see common behaviors. In our case we had to hypothesize
what our statistical analysis meant, while if we had the possibility of
asking questions directly to users we could get the precise answers.
But a positive consequence of this approach is that, being com-
pletely absent from the setting we can assure a low invasiveness and
a greater level of ecological validity of the experiments. With both
approaches would be possible to relate analyzed patterns, especially
on interaction log videos, with in-situ observations explaining much
better the gathered data.

To sum up, despite some technical difficulties experienced by the


users, and the lack of advanced musical features, we received a very
good response. Few comments were directed to the interface per-se
and all of them were strongly positive, emphasizing the novelty of
having “tangible sounds”, improving the creative possibilities. We
argue that these observations show that the time is mature to dis-
tribute tangible user interfaces out of the laboratories or controlled
environments and also in domains different from the audio in an
inexpensive and democratic way. It’s time to bring TUIs to the
masses!

55
Appendix A

UGC texts

A part of the analyzed text content with the open code used.

Table A.1: Some sentence collected with the related open code.
Source Text Category
Mail I think with projects like this the educational
kids will having a lot of fun to per-
form handicraft work & making ex-
periments with music.
Blog in pure style Do-It-Yourself (proba- DIY Love
bly the most appealing part of the
project, from the user perspective)
Blog Bored with mouse pushing and knob low cost ap-
twiddling? The d-touch tangible preciation
sequencer/drum machine makes a
cheap interface (with free download-
able software) for assembling se-
quences.
Blog yes, drums are tangible. We know. tangible
What this is, however, is a tangible interface
interface that is a drum machine. awareness
Continued on next page

56
Table A.1 – continued from previous page
Source Text Category
Blog Global economic recession cutting low cost ap-
into your gear purchases? The team preciation
at d-touch.org may have a solution:
a drum machine that you print with
your computer printer.
Mail All the musicians I showed the unit not good de-
to, commented on how sterile and sign
clean the board looked.
Mail And i’d say you have to be on the net Internet
because the software cost $0. availability
issue
Mail I just tried registering on the d-touch personal data
website and it didn’t like my rather complain
too expansive answer to ”how often
do you play”. It still went ahead
and created the account (username:
Dunx)
Mail Most of my friends used the heavier tangible
objects for the lower sounds instinc- appreciation
tively (metal and glass for kicks and
toms) and the wood for snare and pa-
per for the hats and cymbals.
Forum Had some problem setting the light- lighting prob-
ing (some beat inputs skipped) lem
Forum I will try to improve my setup and Try-to-
experiment a bit with velcro... Will improve
keep you posted. Beesh
Forum Hi my names Gideon and I’m from social sharing
England. I live in a small town called
Chichester which is near Brighton.

57
Appendix B

Open Codes and Axial Coding

The complete open codes produced from collected data.

Table B.1: Open codes and axial coding


Category Entries
Viral sharing 153
Twitter link 103
ReTwit 50
Personal data sharing 15
social sharing 15
First impressions 118
impressed/excited 79
fun 2
skeptical 1
ease of use 2
low cost appreciation 23
like usage 2
fake 9
Applications and real use 10
educational 3
build for others 5
Continued on next page
58
Table B.1 – continued from previous page
Category Entries
commercial issues 2
Improve and extend 40
Try-to-improve 8
DIY love 16
source interest 3
camera under table 2
Try-to-do 11
Physicality of the interface 21
tangible interface awareness 12
TUI comparison 3
tangible appreciation 6
Technical 202
activation problem 63
osx problem 26
camera problem 24
audio problem 20
vista problem 9
registration problem 5
forum problem 1
lighting problem 4
linux problem 1
paper issue 1
linux desire 10
sequencer desire 7
lack of documentation 8
general complain 3
problem fixed 11
calibration 1
augmented reality 7
Cross-platform appreciation 1
Continued on next page

59
Table B.1 – continued from previous page
Category Entries
Audio 25
MIDI 11
use on stage 4
custom sounds 2
music tool 4
tempo selection 1
not good design 1
audio complexity 1
time issuesv1
Field trial related 7
research interest 1
Internet availability issue 5
personal data complain 1
Total 591

60
Appendix C

UGC photos and videos analysis

In this appendix are shown all the sources of our analysis on photo
and video UGCs.

D-touch by cybermolex on 03 July 2009


(http://www.youtube.com/watch?v=tpx9GCBLNhA)

DESCRIPTION:
-What was used for the blocks?
The blocks have been made with the proposed shape.

-What kind of stand?


The stand of the webcam without modifications. It’s a nice long
stand that can be twisted to point where you prefer, it can also
point vertically down.

-Near the computer?


Near two computers! The laptop with Windows is running d-touch,
while on the background we can see a Linux-box.

-What kind of room?


A dark room... probably the user’s bedroom.

61
-Are they on table?
Yes, but in a very small and awkward position. The board is placed
diagonally (probably to fit the webcam position) and the space is
very small, between a monitor, a laptop and a multimeter...

-Audio setup?
Unknown probably laptop speakers.

-Operating system?
Windows on a laptop.

d-touch drum machine on a whiteboard by mutedhar-


mony on 28 July 2009
(http://www.youtube.com/watch?v=D4ApqBnLcEk)

DESCRIPTION:
-What was used
for the blocks?
Plain markers has
been glued on mag-
nets.

-What kind of
stand?
The webcam has
been duct taped
to a music stand
that pointed to the Figure C.1: The vertical setup of d-touch
magnetic whiteboard.

-Near the computer?


No, the computer is far and poorly or not visible from the interface.

-What kind of room?


A big classroom is the setting of this video.

62
-Are they on table?
No, the setup is on a vertical magnetic whiteboard. Also the inter-
active board has been put vertical with magnets.

-Audio setup?
External M-Audio FireWire connection.

-Operating system?
Os X on a MacBook Pro 15”.

Bumm-Tschak by control617 on 28 September 2009


(http://www.youtube.com/watch?v=7-yrn7pyoGk) and
http://gedankenschubla.de/bumm-tschak/ with photos.

DESCRIPTION:
-What was used
for the blocks?
Nothing!

-What kind of
stand?
Cup filled with cof-
fee beans...

-Near the com-


puter?
Yes, on it! Figure C.2: The virtual setup of d-touch and it’s setting

-What kind of room?


Private space.

-Are they on table?


No, just on computer

-Audio setup?

63
Internal audio card, or nothing special

-Operating system?
Windows on a desktop computer.

D-Touch by EATYone on 9 October 2009


(http://www.youtube.com/watch?v=wsHv6P3s6zg)

DESCRIPTION:
-What was used for the blocks?
Wooden blocks with paper glued on it.

-What kind of stand?


A bracket standing over the table.

-Near the computer?


Yes, the laptop is just at the side of the table.

-What kind of room?


Probably a laboratory.

-Are they on table?


Yes, on a small table with only the board and some blocks.

-Audio setup?
Laptop speakers?

-Operating system?
OS X on a MacBookPro 15”

Lovebytes D-Touch Promo by presenceelectronique on 30


October 2009
(http://www.youtube.com/watch?v=PErY3kmMGv4)

DESCRIPTION:

64
-What was used for the blocks?
Paper blocks and simple paper markers. Mostly the simple markers
for actual playing.

-What kind of stand?


A professional camera stand with a DV camera.

-Near the computer?


Yes, the computer is on the same table as the interface.

-What kind of room?


Different rooms in a public building.

-Are they on table?


Yes.

-Audio setup?
Not seen but probably with an external audio card. We see some
brand of musical instruments (Alesis).

-Operating system?
OS X on a MacBookPro 15”.

Musical Experiment Tangible drum machine - tangible


musical interface by AlienKarma1512 on 24 November 2009
(http://www.youtube.com/watch?v=Y5zF7h4A-JQ)

DESCRIPTION:
-What was used for the blocks?
Plain paper, no block is constructed.

-What kind of stand?


Musical stand with camera branched on it.

-Near the computer?

65
Yes, on the same table.

-What kind of room?


It seems a small room, probably used also to play music, a private
space.

-Are they on table?


Yes, a small one. With a strong light pointing to the system.

-Audio setup?
Computer speakers.

-Operating system?
Windows on a desktop computer.

http://www.flickr.com/photos/85596380@N00/
3689219303/

DESCRIPTION:
-What was used for the blocks?
Plain paper.

-What kind of stand?


A bracket standing over the table.

-Near the computer?


Yes, the computer is visible in the photo at the left of the board.

-What kind of room?


Personal room.

-Are they on table?


Yes, a very small table with a light near the board (probably used
to get good recognition).

66
-Audio setup?
A computer speaker is visible in the top of the photo.

-Operating system?
Windows on a laptop

http://www.forest.impress.co.jp/img/wf/docs/307/274/
html/image2.jpg.html

DESCRIPTION:
-What was used for the blocks?
Simple paper and some blocks. Some marker is glued on some small
object not recognizable.

-What kind of stand?


A professional camera stand with a consumer grade webcam.

-Near the computer?


Yes, just in front of it.

-What kind of room?


A cubicle.

-Are they on table?


Yes but the space is very small. Nevertheless the setup is good and
gives good movement possibilities.

-Audio setup?
Unknown, probably headphones.

-Operating system?
Windows Vista.

http://www.pc-music.com/content/d-touch-paper-drum-
machine-full-hands-review

67
DESCRIPTION:
-What was used for the
blocks?
Paper blocks and simple
paper markers.

-What kind of stand?


A professional camera
stand.

-Near the computer?


Yes, to the left of the
computer, in front of a
big speaker.

-What kind of room?


Probably the personal
studio of the user. Figure C.3: One photo from the hands-on review

-Are they on table?


Yes, in a very clumsy table with little or no space.

-Audio setup?
Probably connected to an external audio card (visible in a photo).

-Operating system?
Windows Vista.

68
Appendix D

Riassunto in Italiano

Questa tesi descrive il mio lavoro svolto al Media and Design Lab-
oratory dell’EPFL del Prof. Jeffrey Huang tra Marzo e Settembre
2009, supervisionato da Enrico Costanza. Lo scopo del progetto è
stato quello di studiare l’interazione di un largo numero di utenti
con un tipo di interfaccia ancora poco studiato, l’interfaccia tangi-
bile. Per fare questo abbiamo usato due strumenti musicali, basati
su interfaccia tangibile, realizzati nel 2003 da Enrico Costanza e
Simon Shelley, d-touch drum machine e d-touch sequencer, li abbi-
amo resi scaricabili da Internet, facilmente riproducibili dagli utenti
e, dopo aver raccolto log di interazione e materiale prodotto dagli
utenti e distribuito su Internet, abbiamo studiato come le persone
si rapportano a questo genere di interfaccia nel loro ambiente e per
un periodo di utilizzo prolungato.

D.1 Introduzione

La tesi è inscritta nell’ambito degli studi sull’Interazione Uomo-


Macchina (HCI). Questo ambito è molto vario ed eterogeneo, fanno
parte di HCI studi di informatica, design, psicologia, e molto al-
tro. HCI si definisce come la disciplina che si interessa della pro-

69
gettazione, valutazione e implementazione di sistemi computazionali
interattivi usati da persone e anche come lo studio di tutti i fenomeni
che gravitano attorno a questo [1].

Negli ultimi anni l’importanza di questi studi è via via cresciuta,


come è cresciuto l’interesse delle grosse aziende al dibattito acca-
demico su HCI. Tutte le principali conferenze mondiali sull’HCI,
come CHI (ACM Conference on Human Factors in Computing Sys-
tems) o UIST (ACM Symposium on User Interface Software and
Technology) sono finanziate da aziende come Microsoft, Google, Au-
todesk, IBM and Nokia1 . L’interesse è portato dal fatto che sistemi
interattivi che si adattano meglio alle esigenze dell’utente sono per-
cepiti meglio e venduti di più, oltre ad essere usati con maggior
piacere.

Un particolare campo di ricerca di HCI è basato sulle Interfacce


Utente Tangibili (TUI). Il concetto principale delle TUI è che la rap-
presentazione fisica di un oggetto (come un oggetto fisico manipo-
labile) è associata ad una rappresentazione digitale (audio o video),
che porta alla creazione di interfacce mediate dall’informatica ma
non identificabili come “computer” [3].

L’interesse accademico verso le TUI è nato a metà degli anni ’90,


quando Wellner, Fitzmaurice, Ishii and Ullmer [5, 6, 3, 7] iniziarono
a sviluppare queste nuove interfacce. Da allora il dibattito attorno a
queste interfacce è cresciuto costantemente, fino a quando, nel 2007
è stata organizzata una conferenza specifica su queste tematiche,
TEI2 , che da allora si tiene con cadenza annuale.

A partire da questo interesse nelle TUI sono nati svariati altri


esperimenti nei campi più disparati, come nell’istruzione, nella mu-
sica, nell’intrattenimento e in molto altro. L’ambito musicale ha
avuto molta crescita e rilevanza mediatica con progetti come Au-
diopad e Reactable [13, 11].
1 CHI 2010 web: http://www.chi2010.org/ e UIST 2009 web: http://www.acm.org/uist/uist2009/
2 Tangible and Embedded Interaction Conference. http://tei-conf.org/

70
Le TUI sono sempre state sperimentate e studiate in ambienti
controllati come laboratori e musei e sempre in numeri ristretti per
le difficoltà tecniche di riproduzione. Con d-touch abbiamo cercato
di risolvere questo problema creando e distribuendo su Internet una
TUI a bassissimo costo. Per studiare l’utilizzo di questa interfaccia
abbiamo registrato i movimenti prodotti dagli utenti con le appli-
cazioni e abbiamo raccolto materiale trovato su Internet a proposito
di d-touch, ispirati da un modo di studio già sperimentato in HCI
negli ultimi anni [15, 16, 17].

Alla fine del periodo di osservazione, abbiamo raccolto una quan-


tità di dati rilevante, molto superiore alla media degli altri studi
su TUI che abbiamo analizzato [28, 12, 8, 10, 14, 29] con un costo
sicuramente inferiore, in quanto il nostro team era composto da
tre persone, non dedicate a tempo pieno sul progetto, e il materi-
ale di supporto è stato solo un server web fornito dall’università.
Dall’analisi di questi dati si può osservare che l’interfaccia tangibile
è stata accettata bene dagli utenti che hanno criticato esclusiva-
mente le possibilità musicali dell’applicazione, come se fosse ovvio a
tutti il modo di interagire che però pochissimi avevano sperimentato
prima, come hanno affermato gli stessi utenti in fase di registrazione
sul sito.

D.2 Audio d-touch

Audio d-touch è una coppia di applicazioni, d-touch drum machine


e d-touch sequencer, basate su un’interfaccia tangibile per comporre
musica e suonare dal vivo.

71
D.2.1 Come funziona

La configurazione di Audio d-touch è semplice, come in altre inter-


facce tangibili basate su tavoli, è presente una webcam collegata ad
un computer che riprende dall’alto il tavolo sul quale è posto un
foglio A4 che sarà l’area interattiva dell’applicazione. Per il trac-
ciamento e il riconoscimento dell’area interattiva e dei blocchi usati
per suonare, è stato usato un algoritmo di computer vision, d-touch,
sviluppato da Costanza et al. [25, 27]. L’algoritmo è in grado di trac-
ciare in real-time (a 25/30 Frame Per Secondo) l’area interattiva e
i marker che vengono posizionati su questa. Al momento, con un
foglio A4 come area di utilizzo, abbiamo sperimentato l’applicazione
con 25-30 marker e non abbiamo notato cali di prestazioni.

Dal punto di vista musicale, le applicazioni di Audio d-touch sono


basate sul concetto di loop, cioè che una sequenza di suoni viene
ripetuta all’infinito mentre possono essere modificate delle parti.
Come se ci fosse un cursore virtuale che ripercorre l’area interattiva e
quando incrocia un blocco questo produce il suono che rappresenta.
Questo modello di strumento è molto popolare nell’industria della
musica elettronica alla quale ci siamo ispirati.

Le applicazioni sono scritte in C++ con il supporto di alcune


librerie. Per la gestione dell’audio è stata usata STK3 , sviluppata
al CCRMA di Stanford. Per la parte di computer vision è stata
utilizzata libdtouch, liberamente scaricabile da SourceForge4 che
permette anche la gestione della camera. Per l’interfaccia grafica
e la gestione della rete è stata utilizzata la libreria C++ Qt5 . Tutte
queste librerie sono liberamente utilizzabili e cross-platform, cosa
che ha permesso lo sviluppo e la distribuzione dell’applicazione sotto
Windows e Os X, più una versione di sviluppo per Linux.
3 http://ccrma.stanford.edu/software/stk/
4 http://sourceforge.net/projects/libdtouch/
5 http://qt.nokia.com/

72
D.2.2 d-touch Drum Machine

La drum machine è il primo strumento realizzato e distribuito on-


line. L’obiettivo di questo strumento è di essere facile e divertente,
come una prima introduzione alle possibilità date dalle interfacce
tangibili. L’area interattiva è molto semplice, divisa in undici righe
e sedici colonne: ogni riga rappresenta un suono diverso, identificato
da un testo scritto sulla riga stessa, mentre ogni colonna rappresenta
lo spostamento nel tempo relativo al loop. Per i suoni sono stati us-
ati due marker differenti, uno che indica un volume di riproduzione
normale e un altro per un volume più forte utile per porre l’accento
su alcuni suoni.

D.2.3 d-touch Sequencer

d-touch Sequencer è un’evoluzione della drum machine. Il concetto


alla base, di riproduzione di campioni nel loop continuo, è lo stesso.
La differenza sta nel fatto che con il sequencer è possibile registrare
e riprodurre i propri campioni durante l’utilizzo. In questo stru-
mento sono presenti 18 diversi marker che indicano i diversi suoni
possibili. Più marker identici possono essere presenti nell’interfaccia
ma riprodurranno lo stesso suono. L’area interattiva è composta da
una zona dove i marker attivano la registrazione dal microfono, e
una dove attivano la registrazione di quello che viene riprodotto,
in modo che si possano creare sequenze sempre più complesse. In
questa interfaccia l’asse orizzontale ha lo stesso significato che nella
drum machine, mentre quello verticale indica un diverso volume di
riproduzione. Inoltre la rotazione dei marker fa sı̀ che i campioni
vengano riprodotti ad una velocità diversa, da metà fino al doppio.

73
D.2.4 Sistema di Registrazione Remota

Per permettere l’osservazione da remoto è stato implementato un


sistema di registrazione dei movimenti e di invio ad un server remoto
per l’archiviazione dei dati su database.

Per avere dei dati precisi è stato sviluppato anche un sistema


di attivazione del software. Gli utenti devono infatti usare i dati
di registrazione sul sito anche per attivare il software, in modo
che sia possibile il tracciamento dell’utilizzo attraverso più sessioni
dello stesso utente, permettendo la visualizzazione di comporta-
menti o per capire la curva di apprendimento. Alcuni accorgimenti
sono stati presi per evitare il più possibile la condivisione di ac-
count e di software, come il tracciamento del MAC address della
macchina o dell’indirizzo IP. I dati di utilizzo che registriamo sono
completamente anonimi e riguardano solo gli spostamenti dei bloc-
chi nell’interfaccia. Nessuna immagine è stata presa dal software.
Durante la registrazione è stato specificato chiaramente agli utenti
che il software avrebbe registrato l’utilizzo per fini scientifici e solo
una persona ha chiesto la rimozione dei propri dati dal database.
Altri hanno lamentato l’invasività delle domande in fase di regis-
trazione o la necessità di avere una connessione ad Internet per
usare il software, ma il fatto che il tutto fosse gratuito spesso faceva
accettare questi limiti.

Dal punto di vista server, la registrazione dati è stata sviluppata


modificando un sistema esistente basato su TikiWiki6 , un gestore
di contenuti basato su PHP. È stato sviluppato uno script PHP
che ogni volta che riceve una chiamata POST HTTP raccoglie i
dati (generalmente di 25 frame di utilizzo) e li immagazzina in un
database MySql. I dati contengono un timestamp, un riferimento
temporale dall’inizio della sessione, l’utente che ha prodotto quei
dati, la posizione e l’angolo di ogni marker presente nell’interfaccia.
Il sistema si è rivelato abbastanza robusto, dato che ha sopportato
6 http://tikiwiki.org/

74
l’uso di centinaia di utenti, talvolta anche più utenti simultanea-
mente, apparentemente senza perdere dati.

D.2.5 Distribuzione

La natura semplice e a basso costo del progetto è stata da sempre


uno degli obiettivi principali del progetto. Nessun componente elet-
tronico da saldare, nessuna particolare abilità manuale necessaria,
solo una webcam, un computer con una scheda audio e una stam-
pante. Il pacchetto completo di software, immagini dei blocchi e
dell’area interattiva è scaricabile in meno di 10 Mega Byte.

Il fatto di avere un software cross-platform che abbia a che fare


con l’hardware su cosı̀ tanti livelli, ha reso necessario del lavoro non
scontato. La distribuzione su computer non conosciuti di questo
genere di software è complicata ed è molto facile incontrare problemi
non previsti.

Per avere un’installazione semplice abbiamo preparato le immag-


ini dei blocchi da tagliare, le istruzioni per costruire il supporto per
la webcam e delle fotografie che spiegano tutti i passaggi per arrivare
al prodotto finale. Per promuovere meglio gli strumenti abbiamo an-
che girato e montato due video che sono stati pubblicati su Internet7
e che hanno raccolto un certo numero di visite. L’obiettivo di questo
materiale era quello di spiegare la natura semplice del progetto e di
catturare un po’ di attenzione mediatica, come altri progetti simili
prima di noi avevano fatto.
7 http://www.youtube.com/watch?v=lCv0TvnVUHg http://www.youtube.com/watch?v=AmGP3eCOF5A

75
D.2.6 Il Lancio Online

Fin dall’inizio del 2003 Audio d-touch ha ricevuto commenti in-


formali ed è stato provato informalmente da un ristretto numero
di persone, inclusi alcuni musicisti. I commenti sono sempre stati
positivi e ci hanno motivato a pubblicare online gli strumenti cre-
dendo che a una certa nicchia potesse interessare questo genere di
strumenti.

Cosı̀, in seguito ai successi di altri progetti simili, come Re-


actable [11] e alla pubblicazione di ricerche online condotte con
l’ausilio di informazioni raccolte attraverso Internet, anche in modo
informale [15, 16, 17], abbiamo deciso di pubblicare online Audio
d-touch e di raccogliere i dati di utilizzo.

In seguito alla creazione del sito web d-touch.org abbiamo pub-


blicato materiale fotografico e video sulle interfacce tangibili e il
software necessario per riprodurre gli strumenti. Il giorno del rilas-
cio del software, il 28 Giugno 2009, abbiamo pubblicato materiale su
altri siti per pubblicizzarci e abbiamo usato la piattaforma Twitter8
per comunicare con i nostri futuri utenti. Nell’arco di una settimana
abbiamo ricevuto più di 20000 visite al video, 671 utenti registrati al
sito, 208 persone hanno attivato l’applicazione e 112 l’hanno usata
per più di un minuto. L’attenzione mediatica sul progetto è stata
notevole e abbiamo raccolto gli articoli principali e più interessanti
sul nostro sito9 .

Il 17 Agosto abbiamo lanciato ufficialmente il sequencer ed una


nuova versione della drum machine, con del nuovo materiale pro-
mozionale di supporto. Un kit per la stampa è stato spedito ad
alcune interessanti testate online e abbiamo cercato di essere pro-
mossi dai siti che già ci avevano pubblicato durante il primo lan-
cio. Probabilmente a causa della minor novità, l’entusiasmo è stato
8 http://www.twitter.com
9 http://d-touch.org/coverage/

76
meno di quello del primo lancio, ma dati i tempi più lunghi, cioè
fino al 15 Dicembre 2009, gli utenti registrati sono stati 1252, di cui
389 hanno provato l’interfaccia e 273 hanno suonato per almeno un
minuto, risultando in 199 ore di uso registrato.

D.3 Raccolta e Analisi dei Log

I dati analizzati in questa ricerca sono stati raccolti dagli utenti


che si sono registrati tra il 17 Agosto e il 15 Dicembre. Non abbi-
amo potuto utilizzare anche gli altri dati a causa di un errore nella
registrazione dei dati, che li ha resi inservibili.

L’analisi quantitativa dei log è iniziata con un’indagine demografica


compiuta sui dati inseriti volontariamente alla registrazione. I dati
riguardano i 273 utenti che sono stati presi in considerazione in tutte
le analisi dei log, tranne quelli di 7 persone che ne hanno inseriti di
insensati. Il 27% degli utenti ha meno di 20 anni, il 73% meno di 30
e il 94% meno di 40 anni. Le femmine sono state molto poche, solo
il 2% degli utenti. Pochi utenti hanno riportato conoscenza prece-
dente di interfacce tangibili e ancora meno hanno detto di averle
usate.

I dati statistici principali che abbiamo raccolto sull’uso delle in-


terfacce sono raccolti nella Tabella 4.1, che contiene la lunghezza
media delle sessioni, il numero di sessioni per utente e il numero di
blocchi per utente. Dai boxplot alla Pag. 32 si può vedere che, an-
che se la drum machine è stata usata più del sequencer, nel periodo
analizzato, il sequencer ha avuto sessioni più lunghe nella maggior
parte dei casi.

In generale l’uso di d-touch è stato infrequente: il 21% degli utenti


l’ha usato per un periodo superiore ai 2 giorni e l’11% per un periodo
superiore ad una settimana e dall’istogramma alla Fig. 4.3 si può

77
vedere che l’uso è stato compresso nella maggioranza dei casi in
meno di due ore.

Questi dati saranno utili per interpretare meglio i commenti degli


utenti che vedremo più avanti e saranno ripresi nella discussione
finale per trarre delle conclusioni sull’esperimento.

Un altro tipo di analisi che abbiamo percorso è stato quello in-


trapreso analizzando i video ricavati dai log. Abbiamo creato un
riproduttore di log, in modo da poter visualizzare a video i bloc-
chi riconosciuti dall’applicazione e affiancare questo video all’audio
riprodotto dalle applicazioni alle quali veniva dato come input il log
registrato. In questo modo è stato possibile rivedere tutte le azioni
fatte dagli utenti e riascoltare i loro risultati.

Solo una parte dei video è stata presa in considerazione, dopo


un’analisi statistica delle sessioni, in modo da visualizzare solo quelle
più interessanti e non tutte, che avrebbero impiegato troppo tempo.
Dato che non abbiamo visto tutte le sessioni daremo solo indicazioni
di massima su comportamenti comuni e non percentuali precise che
comunque riteniamo utili come linee guida per l’interpretazione dei
dati e come suggerimenti per futuri lavori sulle interfacce tangibili.

Tra le persone che sono riuscite ad usare d-touch abbiamo no-


tato che tutti sono riusciti a produrre un semplice ritmo con pochi
blocchi o hanno passato un po’ di tempo a giocare con gli stru-
menti, con un paio di blocchi, cercando di capirne il funzionamento.
Ci riferiremo a questo secondo comportamento chiamandolo esplo-
razione dell’interfaccia. Quasi tutti hanno iniziato ad usare gli stru-
menti con una fase esplorativa muovendo dei blocchi davanti alla
telecamera o cercando di calibrare l’area interattiva mostrando i
quattro marker agli angoli. Solo una piccola parte degli utenti ha
cominciato subito producendo dei ritmi, probabilmente quelli che
avevano un background di conoscenze musicali o sulle interfacce
tangibili. L’esplorazione talvolta è stata fatta usando il foglio con

78
i marker non tagliati, come se avessero stampato il PDF pubbli-
cato sul sito, senza ulteriore lavoro. Questo comportamento porta
ad un alto numero di marker riconosciuti (circa 30) per un breve
tempo avendo come risultato un rumore indistinto. Questo compor-
tamento è stato individuato più di una volta, in quanto è il modo
più rapido per controllare il funzionamento dell’applicazione pro-
ducendo qualche suono. Abbiamo anche osservato che dopo questo
test spesso si nota un certo tempo l’interfaccia vuota e poi che poco
a poco, lentamente, si riempie, come se l’utente stesse tagliando i
marker e ponendoli sotto la webcam uno alla volta. Questi pat-
tern esplorativi sono stati visti solo nella prima sessione di ciascun
utente. Le sessioni successive spesso iniziavano già calibrate o con
dei marker già presenti, come se il setup fosse stato lasciato intatto
dall’ultimo utilizzo.

A parte questi pattern esplorativi e la generazione basilare di


ritmi, poche sessioni contengono un numero significativo di blocchi,
permettendo la creazione di sequenze più complesse, come si può
vedere nella Tabella 4.1. Anche se queste poche sessioni lunghe di-
mostrano la possibilità di creare sequenze complesse e musicalmente
interessanti.

Un altro fenomeno interessante è stato osservato, i marker spesso


sono stati riconosciuti in modo intermittente, tradendo difficili con-
dizioni di illuminazione. Questo fenomeno chiaramente peggiora
l’esperienza dell’utente perchè la disposizione fisica dei blocchi non
corrisponde alla disposizione vista dall’applicazione, portando ad
una generazione musicale sbagliata. Nonostante l’algoritmo sia stato
sperimentato a lungo per cinque anni e nonostante la presenza di un
semplice algoritmo di tracciamento, questo fenomeno si è presentato
più spesso del previsto. Crediamo quindi che le condizioni luminose
fossero veramente pessime per dare un cosı̀ difficile riconoscimento
e che probabilmente avremmo dovuto dare maggiori indicazioni su
come illuminare l’area e qualche segnale in più sull’interfaccia grafica
dell’applicazione.

79
In generale abbiamo osservato interessanti sessioni ritmiche quando
si aveva la calibrazione e buone condizioni luminose. Negli altri
casi gli utenti erano in grado soltanto di esplorare l’interfaccia rap-
idamente e quindi dopo breve tempo abbandonarla senza ottenere
niente di interessante.

D.4 Raccolta e Analisi di Contenuti Creati dagli Utenti

L’obiettivo principale della ricerca è stato quello di osservare gli


utenti di interfacce tangibili nei propri ambienti. Attraverso i log
di interazione abbiamo raccolto informazioni sul loro utilizzo, ma
probabilmente la parte più interessante delle informazioni arriva dai
loro commenti informali e dalle foto e i video che hanno condiviso
con noi.

Attorno a d-touch si è levato molto interesse mediatico, anche


grazie alla promozione che abbiamo fatto insieme al lancio delle
applicazioni. Abbiamo osservato e raccolto tutto ciò che abbiamo
trovato in Rete tra il periodo dell’anteprima, cioè dal 28 Giugno
2009, fino al 9 Settembre 2009, più alcuni video in date successive.
In questo tempo abbiamo raccolto più di 120 email, più di 330 post
sul forum di d-touch.org, più di 50 post su blog, centinaia di com-
menti su siti sociali e più di 220 post su Twitter. Attraverso Twitter
e le email abbiamo domandato ai nostri utenti di condividere foto e
video delle loro istallazioni, per capire meglio come il sistema fosse
usato in ambienti non controllati. Sei video sono stati pubblicati su
YouTube10 , una foto su Flickr11 e una foto ci è stata mandata via
mail. Abbiamo anche trovato due articoli di descrizione approfon-
dita del sistema dopo una prova sul campo, con foto di descrizione.
I contenuti trovati arrivano da luoghi sparsi in tutto il mondo, in
particolare da Stati Uniti, Germania, Gran Bretagna, Cile, Brasile,
10 http://www.youtube.com/channel/dtouchshow
11 http://www.flickr.com/photos/85596380@N00/3689219303/

80
Giappone. I più interessanti possono essere trovati sul nostro sito12 .

Tutti i dati raccolti, sono stati analizzati con un approccio ispirato


alla grounded theory [31, 15], specialmente per i contenuti basati su
testi, mentre video e foto sono stati analizzati per capire meglio
in che condizioni viene usato d-touch e se alcuni comportamenti
ricorrenti emergono. In due casi abbiamo trovato degli interessanti
episodi di appropriazione da parte degli utenti.

D.4.1 Analisi dei Testi

Per analizzare i contenuti testuali li abbiamo divisi a livello di frase e


abbiamo categorizzato queste ultime con un metodo chiamato open
coding. La fase di open coding consiste nel categorizzare ogni frase
in un gruppo appropriato e creare un nuovo gruppo se la frase non
fa parte di nessun gruppo creato precedentemente. In questo modo
una grande quantità di dati può essere significativamente ridotta ad
un numero inferiore di categorie che descrivano adeguatamente la
complessità. Il passo successivo consiste nell’axial coding, che con-
siste nel raggruppare le categorie create precedentemente in gruppi
più piccoli che possono essere descritti.

Usando questo approccio, fortemente fondato sui dati, abbiamo


raccolto 591 frasi e raggruppate in 50 categorie (open code), mentre
con l’axial coding abbiamo creato 9 categorie più ampie: fisicità
dell’interfaccia, audio, tecnica, migliora ed estendi, relative alla
prova sul campo, applciazioni e uso reale, prime impressioni, condi-
visione di dati personali, condivisione virale. Per presentare meglio
queste categorie abbiamo creato due gruppi, uno relativo all’utilizzo
reale degli strumenti e uno che si riferisce a commenti generici,
spesso composto da prime impressioni che gli utenti hanno del sis-
tema, anche senza averlo provato. Un riepilogo delle categorie può
essere trovato alla Tabella 5.1 mentre i contenuti più interessanti
12 http://d-touch.org/coverage

81
possono essere trovati alla Sezione 5.2.1 e altri esempi nelle Appen-
dici A e B.

D.4.2 Analisi delle Foto e dei Video

Le foto e i video che abbiamo ricevuto sono stati probabilmente


la parte più informativa del nostro studio, dato che è stato possi-
bile vedere le istallazioni nel loro ambiente e come gli utenti hanno
usato d-touch. Per analizzare i materiali ci siamo posti metodica-
mente delle domande: Cosa è stato usato per i blocchi? Che tipo di
supporto per la webcam? Vicino al computer? Che tipo di camera?
Sono sul tavolo? Che configurazione audio? Che sistema opera-
tivo? Un sommario delle analisi e altro materiale può essere trovato
nell’Appendice C.

Sul lato della fabbricazione, abbiamo osservato che alcune per-


sone hanno preferito restare più semplici possibili, senza piegare i
blocchi di carta, come suggerito, ma usando i marker ritagliando
soltanto i fogli, a scapito dell’usabilità. A volte sono state osser-
vate delle strisce di carta non tagliata piene di marker per costruire
ritmi ripetitivi o per sperimentare rapidamente l’interfaccia, come
abbiamo già osservato dai log. Altre persone invece hanno prefer-
ito costruire dei blocchi di legno sui quali hanno incollato i marker,
per avere una bella configurazione, comoda da usare. Alcuni hanno
usato dei treppiedi professionali per la webcam, altri hanno usato
mobili e scaffali, aiutandosi con dei libri sotto la superficie di utilizzo
per raggiungere l’inquadratura ottimale.

Abbiamo sempre osservato d-touch su tavoli, non è mai stato


costruito su divani, pavimenti o altre superfici improvvisate. I tavoli
erano spesso disordinati e con poco spazio disponibile, portando
a movimenti scomodi e innaturali. Alcune volte il foglio di base
è stato posizionato non di fronte all’utente perché la webcam si
trovava in posizioni laterali o storte. Ovviamente in questi casi non

82
ottimali i risultati musicali erano poco interessanti e neanche troppo
ricercati dagli utenti, che puntavano più al lato tecnico o divertente
dell’applicazione.

Gli scenari di utilizzo erano molto diversi, piccole stanze, camere


da letto, uffici e studi musicali. L’hardware e i sistemi operativi
erano svariati: laptop, desktop, Windows, OS X. Anche dal punto
di vista della configurazione audio i sistemi erano diversi a partire
dalle casse del portatile per finire con sistemi hi-fi professionali.

Due video erano di particolare interesse, dato che sono stati os-
servati casi di appropriazione, da parte degli utenti, della tecnologia
in modo radicale. Il primo esempio era in un’università, dove un
utente ha realizzato d-touch 8 volte più grande del normale e l’ha
attaccato in verticale su una lavagna magnetica con dei magneti
da ufficio, come ha fatto con i marker, attaccandoli su piccoli mag-
neti. In questo modo d-touch è stato usato in verticale, in maniera
collaborativa da più persone, come si può vedere nel video.

L’altro video interessante era, se possibile, ancora più rivoluzionario.


L’utente è riuscito ad usare d-touch senza la carta! Ha puntato la
webcam sullo schermo dove, usando un programma di grafica, ha
mostrato l’area interattiva con i marker e spostava questi ultimi
usando il mouse, sovvertendo completamente l’idea di interfaccia
tangibile.

D.5 Discussione

Dopo l’analisi separata dei diversi dati, procediamo ad un’analisi


complessiva per capire le direzioni di lavoro futuro.

Dall’analisi complessiva dei dati possiamo desumere che il sem-


plice format di interfaccia tangibile a basso costo, stampabile, fun-

83
ziona. Un alto numero di persone è riuscita a farlo funzionare, senza
il nostro intervento fisico. d-touch è stato usato in diversi ambienti e
ha dimostrato un elevato grado di versatilità. Il concetto di webcam
e di area interattiva ripresa dall’alto con blocchi mobili funziona ed
è ben compreso. Un punto negativo è stato l’algoritmo di riconosci-
mento che non si è dimostrato sufficientemente robusto. Nonostante
gli anni di sperimentazione, essere usato in ambienti non controllati
è qualcosa che non può essere sperimentato prima. Anche la cali-
brazione dell’area interattiva non ha dato ottimi risultati e spesso
il sistema produceva suoni che non rispecchiavano la configurazione
fisica a causa di questi due problemi. Con il presentarsi di questi
problemi ci siamo resi conto che il sistema forniva piccoli indizi sul
fatto che l’illuminazione fosse corretta o il sistema calibrato, questo
perché per noi erano fatti scontati dei quali non ci siamo preoccu-
pati.

I dati nella Tabella 4.1 supportano l’idea che gli strumenti siano
stati percepiti più come dei giochi che come dei veri strumenti musi-
cali. Si può vedere ciò dal fatto che la drum machine, più semplice è
stata preferita al sequencer (65% contro il 35%) e il tempo trascorso
tra il primo e l’ultimo utilizzo raramente supera i due giorni, solo il
21% degli utenti ha usato le interfacce più a lungo. Inoltre il fatto
che sul sequencer siano stati utilizzati più blocchi spinge a pensare
che lo strumento più complesso è più interessante per usi musicali,
anche se meno interessante dal punto di vista mediatico.

Dal materiale trovato su Internet o ricevuto via email, abbiamo


notato diversi casi di appropriazione da parte degli utenti. Due
persone volevano usare d-touch per insegnare, musica ai bambini e
design all’università. Alcuni hanno cambiato il setup di d-touch,
ingrandendolo per un festival, mettendolo in verticale per insegnare
e rendere il lavoro collaborativo. Alcuni hanno migliorato i bloc-
chi facendoli di legno o proponendo soluzioni migliori per sfruttare
il peso degli oggetti. Un utente è addirittura riuscito a creare
un’interfaccia tangibile virtuale usando la webcam per riprendere

84
lo schermo sul quale visualizzava i marker tramite un programma
di grafica (simile ad Adobe Illustrator). Da questi casi si può capire
che gli utenti hanno capito bene il concetto di interfaccia tangibile
e l’hanno adattato ai loro interessi negli scenari più diversi.

Come riflessione finale, possiamo sostenere che la distribuzione


online e lo studio di un’interfaccia tangibile da remoto è stato un
successo, anche se è stato il primo caso di sperimentazione in questo
senso. Anche se le applicazioni sono state percepite come giochi più
che strumenti musicali, abbiamo ricevuto un’enorme attenzione me-
diatica e gli utenti coinvolti nell’esperimento ci sono stati di grande
aiuto nel migliorare le applicazioni. Molti commenti sono stati rac-
colti, ma pochissimi sull’interfaccia e i pochi erano molto positivi.
L’appropriazione da parte degli utenti è stato un segnale incorag-
giante per capire che l’interfaccia è stata percepita e capita a fondo.
Si può dire che l’interfaccia era “ovvia” da usare anche per chi non
aveva mai provato niente di simile prima.

D.6 Lavoro Futuro

La maggiore priorità al momento, per prossimi sviluppi, è di aggiun-


gere dei segnali video nella GUI per aiutare gli utenti a calibrare
meglio il sistema e per avere un’illuminazione migliore. Crediamo
che questo possa essere di grande aiuto per ottenere un’esperienza
migliore facendo sı̀ che l’impostazione fisica dei blocchi corrisponda
con la riproduzione sonora. Due migliorie in questa direzione potreb-
bero essere un chiaro segnale che indichi che l’applicazione non è
calibrata al momento e una semplice spiegazione su come calibrare.
Sul versante riconoscimento si potrebbe semplificare l’algoritmo di
soglia, che ora è poco funzionale in caso di luce forte e diretta, e sos-
tituirlo con una soglia più semplice calibrabile dall’utente in base
alla luce di cui dispone. Crediamo che queste due migliorie pos-
sano incrementare notevolmente le possibilità di Audio d-touch nel

85
prossimo futuro.

Siamo interessati anche a creare le condizioni per rendere questo


genere di strumenti più interessanti sul lungo periodo. Un possibile
spunto potrebbe essere quello di permettere agli utenti di condi-
videre le loro composizioni sul sito e permettere agli altri di scari-
carle, stamparle e risuonarle, come se si trattasse di uno spartito
musicale.

Sempre nell’ottica Web ci sarebbe la possibilità di fare il porting


dell’applicazione su Adobe Flash grazie ad Alchemy13 , che permet-
terebbe di ampliare ulteriormente l’audience possibile, rimuovendo
anche la necessità di installare il software, usando semplicemente un
sito web dal proprio browser.

Siamo anche interessati ad aggiungere un feedback video diretta-


mente sull’interfaccia per incrementare le possibilità degli strumenti.
Per restare nel basso costo potrebbero rivelarsi utili i pico-projector
che potrebbero essere affiancati alla webcam oppure, se si risolve
il problema del costante controluce, si potrebbe usare un monitor
LCD come area interattiva al posto del foglio A4.

D.7 Conclusione

Il lavoro di questa tesi si compone nel rilascio pubblico di un’interfaccia


tangibile per la produzione musicale e in uno studio di utilizzo da
parte di un grande numero di utenti nei loro ambienti naturali. Il la-
voro è cominciato con la creazione di un sistema di registrazione dei
movimenti degli utenti, quindi abbiamo pubblicato il software on-
line e, dopo qualche mese di osservazione e di raccolta dati, siamo
passati all’analisi, tenendo in considerazione anche tutte le infor-
mazioni che abbiamo trovato su Internet condivise dagli utenti.
13 http://labs.adobe.com/technologies/alchemy/

86
Il contributo di questa tesi allo studio dell’interazione uomo-
macchina è dato dal nuovo metodo di osservazione utilizzato, che
nonostante sia ancora da rivedere e migliorare, ha dato buoni risul-
tati tenendo conto dei ridottissimi costi che ha comportato se lo
compariamo ai metodi tradizionali di studio sulle interfacce tangi-
bili. Probabilmente il metodo potrebbe essere migliorato notevol-
mente se fosse affiancato ad un’osservazione più tradizionale degli
utenti, che a noi non è stata possibile per la dimensione ridotta
del nostro gruppo di ricerca. Con l’uso di entrambi gli approcci si
sarebbe potuta eliminare una parte di speculazione sui dati che ha
portato a risultati non sicuri, ad esempio quando i dati statistici
erano contraddittori. Con alcune interviste a campione avremmo
capito meglio dei comportamenti ricorrenti che avrebbero semplifi-
cato di molto l’analisi e che avrebbero portato a risultati più affid-
abili.

In ogni caso, a parte i problemi tecnici e la mancanza della


complessità necessaria agli strumenti musicali, abbiamo ricevuto
un’ottima risposta dal pubblico di Audio d-touch e dai media. Una
parte ridottissima dei commenti erano diretti all’interfaccia, tutti
positivi e che enfatizzavano l’aspetto del “suono tangibile” o delle
nuove possibilità creative date dallo strumento. Crediamo che questo
tipo di osservazioni mostri che il tempo è ormai maturo per far us-
cire le interfacce tangibili dai laboratori e dagli ambienti controllati,
anche in ambiti diversi da quello musicale in modo economico e più
democratico. È tempo di portare le interfacce tangibili alle masse!

87
Bibliography

[1] T. T. Hewett, R. Baecker, S. Card, T. Carey, J. Gasen, M. Man-


tei, G. Perlman, G. Strong, and W. Verplank. http://old.
sigchi.org/cdg/cdg2.html, 1996.
[2] M. Weiser, “Ubiquitous computing.” http://sandbox.xerox.
com/ubicomp/, 1996.
[3] B. Ullmer and H. Ishii, “Emerging frameworks for tangible user
interfaces,” IBM Syst. J., vol. 39, no. 3-4, pp. 915–931, 2000.
[4] E. Hornecker, “Tangible interaction.” http://www.
interaction-design.org/encyclopedia/tangible_
interaction.html, 2009.
[5] P. Wellner, “The digitaldesk calculator: tangible manipulation
on a desk top display,” in UIST ’91: Proceedings of the 4th
annual ACM symposium on User interface software and tech-
nology, (New York, NY, USA), pp. 27–33, ACM, 1991.
[6] G. W. Fitzmaurice, H. Ishii, and W. A. S. Buxton, “Bricks:
laying the foundations for graspable user interfaces,” in CHI
’95: Proceedings of the SIGCHI conference on Human factors in
computing systems, (New York, NY, USA), pp. 442–449, ACM
Press/Addison-Wesley Publishing Co., 1995.
[7] H. Ishii and B. Ullmer, “Tangible bits: towards seamless inter-
faces between people, bits and atoms,” in CHI ’97: Proceedings
of the SIGCHI conference on Human factors in computing sys-
tems, (New York, NY, USA), pp. 234–241, ACM, 1997.

88
[8] D. Xu, J. C. Read, E. Mazzone, and M. Brown, “Designing and
testing a tangible interface prototype,” in IDC ’07: Proceedings
of the 6th international conference on Interaction design and
children, (New York, NY, USA), pp. 25–28, ACM, 2007.
[9] P. Marshall, “Do tangible interfaces enhance learning?,” in TEI
’07: Proceedings of the 1st international conference on Tangible
and embedded interaction, (New York, NY, USA), pp. 163–170,
ACM, 2007.
[10] T. Drori and M. Rinott, “Pixel materiali: a system for creating
and understanding pixel animations,” in IDC ’07: Proceedings
of the 6th international conference on Interaction design and
children, (New York, NY, USA), pp. 157–160, ACM, 2007.
[11] S. Jordà, G. Geiger, M. Alonso, and M. Kaltenbrunner, “The re-
actable: exploring the synergy between live music performance
and tabletop tangible interfaces,” in TEI ’07: Proceedings of
the 1st international conference on Tangible and embedded in-
teraction, (New York, NY, USA), pp. 139–146, ACM, 2007.
[12] K. Ryokai, S. Marti, and H. Ishii, “I/o brush: drawing with ev-
eryday objects as ink,” in CHI ’04: Proceedings of the SIGCHI
conference on Human factors in computing systems, (New York,
NY, USA), pp. 303–310, ACM, 2004.
[13] J. Patten, B. Recht, and H. Ishii, “Audiopad: a tag-based inter-
face for musical performance,” in NIME ’02: Proceedings of the
2002 conference on New interfaces for musical expression, (Sin-
gapore, Singapore), pp. 1–6, National University of Singapore,
2002.
[14] G. Zufferey, P. Jermann, A. Lucchi, and P. Dillenbourg, “Tin-
kersheets: using paper forms to control and visualize tangible
simulations,” in TEI ’09: Proceedings of the 3rd International
Conference on Tangible and Embedded Interaction, (New York,
NY, USA), pp. 377–384, ACM, 2009.
[15] M. Blythe and P. Cairns, “Critical methods and user generated
content: the iphone on youtube,” in CHI ’09: Proceedings of the

89
27th international conference on Human factors in computing
systems, (New York, NY, USA), pp. 1467–1476, ACM, 2009.
[16] A. Kittur, E. H. Chi, and B. Suh, “Crowdsourcing user studies
with mechanical turk,” in CHI ’08: Proceeding of the twenty-
sixth annual SIGCHI conference on Human factors in comput-
ing systems, (New York, NY, USA), pp. 453–456, ACM, 2008.
[17] M. V. Kleek, M. Bernstein, K. Panovich, G. Vargas, D. Karger,
and M. C. Schraefel, “Note to self: Examining personal infor-
mation keeping in a lightweight note-taking tool,” in ACM CHI
2009, 2009. Nominated, best in CHI Award, ACM CHI 2009.
[18] M. Weiser, “The computer for the 21st century,” Scientific
American, September 1991.
[19] M. Weiser and J. S. Brown, “Designing calm technol-
ogy.” http://www.ubiq.com/hypertext/weiser/calmtech/
calmtech.htm, 1995.
[20] J. Underkoffler and H. Ishii, “Urp: a luminous-tangible work-
bench for urban planning and design,” in CHI ’99: Proceedings
of the SIGCHI conference on Human factors in computing sys-
tems, (New York, NY, USA), pp. 386–393, ACM, 1999.
[21] J. Y. Han, “Low-cost multi-touch sensing through frustrated
total internal reflection,” in UIST ’05: Proceedings of the 18th
annual ACM symposium on User interface software and tech-
nology, (New York, NY, USA), pp. 115–118, ACM, 2005.
[22] P. Bennett and S. O’Modhrain, “The beatbearing: a tangi-
ble rhythm sequencer,” in Proceedings of NordiCHI 2008: 5th
Nordic Conference on Computer-Human Interaction (electronic
proceedings), 2008.
[23] H. Kato and M. Billinghurst, “Marker tracking and hmd cal-
ibration for a video-based augmented reality conferencing sys-
tem,” in IWAR ’99: Proceedings of the 2nd IEEE and ACM
International Workshop on Augmented Reality, (Washington,
DC, USA), p. 85, IEEE Computer Society, 1999.

90
[24] M. Gandy, B. Jones, S. Robertson, T. O‘Quinn, and A. John-
son, “Rapidly prototyping marker based tangible user inter-
faces,” in VMR ’09: Proceedings of the 3rd International Con-
ference on Virtual and Mixed Reality, (Berlin, Heidelberg),
pp. 159–168, Springer-Verlag, 2009.
[25] E. Costanza and J. Robinson, “A region adjacency tree ap-
proach to the detection and design of fiducials,” in VVG,
pp. 63–69, 2003.
[26] R. Bencina, M. Kaltenbrunner, and S. Jorda, “Improved topo-
logical fiducial tracking in the reactivision system,” in CVPR
’05: Proceedings of the 2005 IEEE Computer Society Confer-
ence on Computer Vision and Pattern Recognition (CVPR’05)
- Workshops, (Washington, DC, USA), p. 99, IEEE Computer
Society, 2005.
[27] E. Costanza and J. Huang, “Designable visual markers,” in
CHI ’09: Proceedings of the 27th international conference on
Human factors in computing systems, (New York, NY, USA),
pp. 1879–1888, ACM, 2009.
[28] M. Fjeld, J. Fredriksson, M. Ejdestig, F. Duca, K. Bötschi,
B. Voegtli, and P. Juchli, “Tangible user interface for chem-
istry education: comparative evaluation and re-design,” in CHI
’07: Proceedings of the SIGCHI conference on Human factors in
computing systems, (New York, NY, USA), pp. 805–808, ACM,
2007.
[29] M. Waldner, J. Hauber, J. Zauner, M. Haller, and
M. Billinghurst, “Tangible tiles: design and evaluation of a tan-
gible user interface in a collaborative tabletop setup,” in OZCHI
’06: Proceedings of the 18th Australia conference on Computer-
Human Interaction, (New York, NY, USA), pp. 151–158, ACM,
2006.
[30] Y. Jung, J. Blom, and P. Persson, “Scent field trial: under-
standing emerging social interaction,” in MobileHCI ’06: Pro-
ceedings of the 8th conference on Human-computer interaction

91
with mobile devices and services, (New York, NY, USA), pp. 69–
76, ACM, 2006.
[31] H. Sharp, Y. Rogers, and J. Preece, Interaction Design: Beyond
Human-Computer Interaction. Wiley, 2 ed., March 2007.
[32] E. Costanza, S. Shelley, and J. Robinson, “D-touch: A
consumer-grade tangible interface module and musical appli-
cations,” in Proceedings of Conference on HumanComputer In-
teraction (HCI03), 2003.
[33] E. Costanza, S. Shelley, and J. Robinson, “Introducing audio
d-touch: A tangible user interface for music composition and
performance,” in Proc. of the 6th Int. Conference on Digital
Audio Effects (DAFX-03), 2003.
[34] J. Raskin, The humane interface: new directions for designing
interactive systems. New York, NY, USA: ACM Press/Addison-
Wesley Publishing Co., 2000.

92

S-ar putea să vă placă și