Eliciting User-defined Zenithal Gestures for Privacy Preferences
Francisco J. Mart
´
ınez-Ruiz
1 a
and Santiago Villarreal-Narvaez
2 b
1
Universidad Aut
´
onoma de Zacatecas, Centro Hist
´
orico C.P. 98000, Zacatecas, Mexico
2
Universit
´
e Catholique de Louvain (UCLouvain), LouRIM, B-1348 Louvain-la-Neuve, Belgium
Keywords:
Ambient Intelligence, Gesture-based User Interfaces, Gesture Elicitation Study, Privacy, Public Space,
Zenithal Camera, Zenithal Gestures.
Abstract:
Common spaces are full of cameras recording our pictures purposely or unintentionally, which causes privacy
concerns. Instead of specifying our privacy preferences on one device or sensor at a time, we want to capture
them once for an entire building through zenithal gestures in order to notify all devices in this building. For
this purpose, we present an elicitation study of gestures elicited from thirty participants to notify reactions,
acceptance or refusal of actions, via gestures recognized by a zenithal camera placed on the ceiling at the
entrance. This perspective is different from the tradition frontal or lateral perspective found in other studies.
After classifying the results into forty-six gesture classes, we suggest a consensus set of ten user-defined
zenithal gestures to be used in a common space inside a building.
1 INTRODUCTION
In Ambient Intelligence (AmI) (Cook et al., 2009),
common spaces, like public places, shopping malls,
railway stations, are full of cameras and sensors
where our image and data could be recorded on pur-
pose or unintentionally (Ashok et al., 2014). For ex-
ample, a person can pass by a public display and be
captured for personalizing or customizing the user in-
terface, which is perceived as a good benefit, or for
suggesting some products in a shop close by, which is
perhaps not considered as a benefit, depending on pri-
vacy preferences of the end user. These preferences
can be communicated explicitly from one device to
another (Kray et al., 2010), but this poses an addi-
tional problem of preferences transfer.
We investigate an alternative approach: instead of
informing device by device, privacy preferences of
users (Lopez et al., 2014) are transmitted to an en-
tire building through zenithal gestures so as to notify
all concerned devices. We define zenithal gestures
as end-users body gestures (Kendon, 2004) acquired
by a sensor located on top of the end user, hence its
name borrowing the term “zenith”. Although zenithal
gestures could cover in principle any human limb,
zenithal gestures typically consist of upper-body ges-
tures (e.g., head and shoulders (Vanderdonckt et al.,
a
https://orcid.org/0000-0002-8842-7556
b
https://orcid.org/0000-0001-7195-1637
2019), arms (Liu et al., 2015) until hands (Bostan
et al., 2017)) acquired by a zenithal (usually spheri-
cal) camera hanging from the ceiling.
A zenithal camera offers the following advan-
tages: it enables capturing an entire scene in a snap-
shot instead of a part of the environment, which is par-
ticularly appropriate when this environment dynami-
cally changes (Cook et al., 2009); algorithms for ges-
ture recognition (Coyette et al., 2007) are more sim-
ple than for gestures acquired with a frontal or a lat-
eral camera, such as a MS Kinect Azure because the
physical space should not be scaled, translated, or ro-
tated in pre-processing (Vanderdonckt et al., 2018); it
provides precise support for person detection (Li and
Ma, 2018), localization (Grd et al., 2018), and count-
ing (Anti
´
c et al., 2009) in various environments (e.g.,
counting the customers coming in or out a shop), three
important functions in global human activity recogni-
tion. However, the image captured by a zenithal cam-
era may suffer from irregularities of head movements
and partial limb occlusion (e.g., neck movements can-
not be differentiated as they are hidden by the head).
To support our initial scenario, we conduct a
gesture elicitation study (Wobbrock et al., 2009) of
zenithal gestures that notifies reactions and accep-
tance of gestures recognized from a zenithal perspec-
tive. We collect and propose a set of gestures to be
used in a common space inside a building. For this
purpose, we simulate the environment by equipping
Martínez-Ruiz, F. and Villarreal-Narvaez, S.
Eliciting User-defined Zenithal Gestures for Privacy Preferences.
DOI: 10.5220/0010259802050213
In Proceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2021) - Volume 2: HUCAPP, pages
205-213
ISBN: 978-989-758-488-6
Copyright
c
2021 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
205
it with a collection of sensors capable of recognizing
the presence of human beings and responding to their
presence (Sadri, 2011). This paper aims at exploring
gestures that users associate with privacy preferences
and identifying the reaction of users while defining
privacy settings in common spaces.
The rest of the paper is organized as follows: Sec-
tion 2 presents work related to end user interaction,
concentrating on privacy preferences, ambient intelli-
gence, and gestures. Section 3 describes and conducts
the research method used for eliciting zenithal ges-
tures from participants of an experiment and Section 4
presents its main results and discuss them. Finally,
Section 5 provides conclusion and discusses some fu-
ture avenue to this work.
2 RELATED WORK
This section presents an overview of the concepts and
work about privacy and gesture elicitation studies in
general, then specifically in common spaces.
2.1 Privacy Preferences and Ambient
Intelligence
Privacy is a particularly important issue for users and
governments (Lopez et al., 2014). In a survey docu-
mented in the RFID Position Statement of Consumer
Privacy and Civil Liberties Organizations
1
, privacy is
considered more important than the benefits of AmI
itself. The phantom of the “Big Brother” is present
for users (Cook et al., 2009). Although an always
watching system could be useful at home, museums,
business, health care facilities, and in public build-
ings (Sadri, 2011), it is also useful in a personal space.
For instance, St. Pierre et al. (2014) propose a record-
ing system that helps users to recover information
from previous sessions to make it available only in
an individual context. Nevertheless, end users toler-
ate a certain reduction in privacy in the social network
realms. A photo or a tagging is acceptable if shared
with friends (Besmer and Richter Lipford, 2010).
However, this is a vastly different story in public
spaces. How to provide users with specific configura-
tions and services in a public area? Grd et al. (2018)
proposed a system to encode biometric information
and deliver a hash id, so identity and biometric data
of users remain safe. Similarly, Ashok et al. (2014)
introduced a way to inform cameras privacy prefer-
ences of users with IR LEDs an approach that can be
used also in buildings. Finally, the myriad of cameras
1
See https://www.aclu.org/other/rfid-position-statement
Figure 1: Gesture variations for a “X” symbol to express a
Decline action (Erazo et al., 2017).
available nowadays are informed with gestures about
privacy concerns (Koelle et al., 2018), but with frontal
cameras and hand gestures.
A zenithal camera can be effectively used as a in-
door GPS whose data can be merged with additional
data (Valera et al., 2007). In our exploratory study,
end users have an active role as they are engaged in
participatory design (Bergold and Thomas, 2012) be-
cause privacy preferences are explicitly expressed and
then broadcasted to cameras and sensors in the public
space. Contrarily to the “display blindness” of pub-
lic interfaces where users avoid participating because
information is considered not relevant (M
¨
uller et al.,
2009), this scenario foster end users to participate by
communicating their preferences.
2.2 The Design and Elicitation of
Gestures
Gestures can be described using multiple notations
and classifications (Grijincu et al., 2014). More tech-
nology and more reasonably priced becomes avail-
able now to identify and register gestures, such as a
Microsoft Kinect Azure or radar-based devices (Ma-
grofuoco et al., 2019). Also, there is an effort from
open source communities to make available libraries
of these sensors to a wider audience to implement an
AmI environment and delivering a series of services.
The research method consists of applying a gesture
elicitation study (GES) (Wobbrock et al., 2009; Mor-
ris et al., 2010): a set of voluntary participants is pre-
sented with a series of referents, which materializes a
task or an action to be executed by system according
to the end user’s gesture. Then, gestures are elicited
from participants and analysed to identify a consensus
set of gestures. This kind of study allows users to pro-
pose gestures that they feel more acceptable for a spe-
cific task (Rodriguez and Marquardt, 2017). A Sys-
tematic Literature Review (SLR) of existing GES re-
veals that they are primarily conducted depending on
several conditions (Villarreal-Narvaez et al., 2020),
depending on the type of users (e.g., able-bodied peo-
ple vs. people with disabilities), depending on their
tasks (e.g., actions performed in an AmI environ-
ment (Grijincu et al., 2014), sketching actions (Ki-
effer et al., 2010; Sangiorgi et al., 2012)), depending
on the device (e.g., a tabletop (Morris et al., 2010),
a radar-based device (Magrofuoco et al., 2019), or
HUCAPP 2021 - 5th International Conference on Human Computer Interaction Theory and Applications
206
a combination of devices (Kray et al., 2010)), de-
pending on which human limbs are concerned (e.g.,
hands (Bostan et al., 2017), arms (Liu et al., 2015)),
and depending on the physical environment where
end users are interacting (e.g., in public settings (Rico
and Brewster, 2010)).
Since this study elicits gestures from participants
in order to express their privacy settings, the closest
studies are related to opt-in/out actions (Erazo et al.,
2017; Koelle et al., 2018; Rodriguez and Marquardt,
2017), which are different from general privacy con-
cerns. For example, different gesture patterns can be
used to express a same action, like variations for a
“X” symbol to express a Decline action (Fig. 1).
While the elicitation considers multiple factors
from participants, such as naturalness, intuitiveness,
memorability, etc., the proposal can be affected by
the legacy bias (Morris et al., 2010): most partici-
pants, who are often users of WIMP interfaces (win-
dows, icons, menus and pointing mechanisms), are
prone to an unintentional reuse of familiar metaphors
and tend to propose the same gestures they already
know, even if the conditions are quite different. We
apply different techniques to minimize this bias, such
as kinesthetic priming (Hoff et al., 2016) and soft con-
straints (Ruiz and Vogel, 2015). Since our study is
specifically tailored to privacy concerns, social ac-
ceptability plays an important role (Rico and Brew-
ster, 2010): people may elicit different gestures when
they are in a public environment, such as discrete ges-
tures, and less constrained gestures when they are in
a private environment.
3 ZENITHAL GESTURE
ELICATION STUDY
We conducted a GES following the methodology
originally defined from the literature (Wobbrock
et al., 2009; Vatavu and Wobbrock, 2015) to collect
users’ preferences for zenithal gestures associated to
privacy actions.
3.1 Participants
Thirty voluntary participants (15 Females, 15 Males;
aged from 18 to 55 years, M=26.50, SD=11.43) were
recruited for the study via a contact list broadcasted
in our organization. Most participants were students
(25), the the rest being employees (4 professors and
1 administrative assistant). The age groups were in-
tended to be as representative as possible for adopters
of wearable computing, a technology that the per-
centage of individuals using it is the highest for the
Figure 2: Apparatus used in the setup.
age group 25-34 years. All the participants were fre-
quent users of computers and smartphones. Most of
the users have experience with videogames (25 users,
with 5 among them reporting to be frequent users) and
gestural interface knowledge was extremely limited.
3.2 Apparatus
The controlled experiment took place in the main en-
trance of the building, where some space was devoted
to dispose the experiment settings and materials. A
space of two square meters space was marked with
an “X” to indicate the interaction position to partici-
pants captured by a SriHome 3 MegaPixel AI IP cam-
era
2
(Fig. 2). This location was selected to simulate
a crowded space (the space was used by other people
and passersby) to test gestures under real conditions.
3.3 Procedure
During the introduction to the experimental set-
ting, each session was initiated by an explanation of
the privacy settings to be tested, the position of the
camera, the signing of a consent, form and the filling
of a socio-demographic questionnaire about their ex-
perience with different devices and demographic in-
formation (based on a 7-point Likert scale (Likert,
1932) ranging from 1=strongly disagree to 7=strongly
agree). The experimenter explained to participants
the following tasks that they had to perform and the
allowed types of zenithal gestures they could propose
to be compliant with our definition.
During the test phase, each session applied the
original GES protocol (Wobbrock et al., 2009). Each
participant was verbally submitted to 10 referents
(Table 1) in a sequential order: S1=Accept Sharing,
S2=Decline Sharing, T1=Accept Tagging, T2=Decline
Tagging, L1=Accept Locating, L2=Decline Locat-
ing, B1=Accept Blurring, B2=Decline Blurring, and
OkAll=Accept all general actions, NotOkAll=Decline
all general actions. For each referent, the partici-
pant proposed one full-body gesture (they could use
the full range of gestures such as head, hands, arms,
2
See http://www.sricam.com/
Eliciting User-defined Zenithal Gestures for Privacy Preferences
207
chest, legs, and feet) with the following conditions:
each gesture should fit well with the referent, it should
be easy to produce and to remember, it should not
involve any vocal or touch interaction. Participants
were instructed to remain as natural at all times as
possible. They operated with the belief that no tech-
nological constraint (e.g., no constraint due to gesture
recognition) was imposed to preserve the natural and
intuitive character of the elicitation. Proposed ges-
tures were captured by the zenithal camera in a video
sequence, one per participant, and saved in a file for
analysis.
During the post-test phase, participants filled in
the IBM Post-Study System Usability Questionnaire
(PSSUQ) (Lewis, 2002). While this questionnaire
is traditionally used for testing software interaction,
it was selected here among other questionnaires to
enable participants to express their level of satisfac-
tion with the perceived setup usability, the conve-
nience of the zenithal camera, and the testing pro-
cess. This questionnaire has been empirically vali-
dated with a large number of participants on a signif-
icant set of stimuli (Lewis, 2006), it is widely appli-
cable for any system, and it benefits from a proved
α=.89 reliability coefficient between its results and
the perceived system usability (Lewis, 1995). Each
IBM PSSUQ closed question is measured using a
7-point Likert scale (1=strongly disagree, 2=largely
disagree, 3=disagree, 4=neutral, 5=agree, 6=largely
agree, 7=strongly agree) and four measures are com-
puted: system usefulness (SysUse: Items 1-5), quality
of the information (InfoQual: Items 6-11), quality of
the interaction (InterQual: Items 12-15), and system
quality (Overall: Item 16). A whole session in aver-
age took 12 minutes: 5 minutes for each questionnaire
and 2 minutes for eliciting zenithal gestures.
3.4 Design
Our study was within-subjects with two groups (fe-
male vs. male) and with one independent variable:
REFERENT, a nominal variable with 10 conditions,
representing privacy preferences, i.e., Accept or De-
cline actions (Table 1). Most systems reacted late
at preventing the sharing or tagging photos and in-
cluded no tagging, no taking photos, and blurring op-
tions (Besmer and Richter Lipford, 2010). Sharing
and tagging have been treated in social networks. So,
we proposed a list of privacy preferences to be in-
formed a priori at the entrance of the public space
or building. In contrast to (Koelle et al., 2018) who
elicited frontal gestures for the opt-in and opt-out ac-
tions, we considered the Accept (Ok) and DECLINE
(NOK) symmetric options. Lastly, we decided to in-
Table 1: List of referents by action on various objects.
Referents Accept (Ok) Decline (NOk)
Sharing S1 S2
Tagging T1 T2
Locating L1 L2
Blurring face B1 B2
General OkAll NotOkAll
clude global preferences that should be valid in any
case. These selectors will inform the public space
that we accept or decline all preferences in a single
command. Two groups were considered to investi-
gate whether the gestures elicited by participants are
sensitive to gender: would male and female partic-
ipants prefer zenithal gestures differently or consis-
tently since their respective preference is subject to
social acceptance?
4 RESULTS AND DISCUSSION
A total amount of 300 gestures were elicited from 2
groups × 15 participants × 10 referents, which we
clustered into groups of similar types according to the
pattern criteria devised in (Erazo et al., 2017) (see
Fig. 1 for some examples of variations clustered in
the same class/pattern). For this purpose, all ges-
tures were labeled with the terms described in Ges-
tureML
3
to maintain a consistent vocabulary, thus re-
sulting into 46 classes of zenithal gestures (Fig. 4).
Fig. 3 shows the agreement rates (Vatavu and
Wobbrock, 2016) computed by AGATe (Vatavu and
Wobbrock, 2015) for each REFERENT condition
sorted by decreasing order of the agreement rate. The
values are aggregated for both groups. Below are the
two most preferred gestures elicited by participants
or each referent. No GES exists today that covers
the same referents. The closest GES elicited gestures
for opt-in/out in both commercial/private setups (Ro-
driguez and Marquardt, 2017) (Fig. 3).
Overall, these values belong to the range of
medium agreement (M=.129, SD=.085) according to
the magnitude table (Vatavu and Wobbrock, 2015).
This is aligned with classical results of GES but can
also be explained by the size of the classification:
the more classes are included in the classification,
the more choices participants invented, and the less
the agreement rates become. The Accept blurring
face referent stands out as it is the only one bene-
fiting from a high agreement (M=.363), thus sug-
gesting that the most frequent gesture class elicited
3
See http://www.gestureml.org
HUCAPP 2021 - 5th International Conference on Human Computer Interaction Theory and Applications
208
0.363 0.172 0.138 0.138 0.101 0.09 0.085 0.078 0.069 0.053 0.129
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Accept
blurring face
Decline
blurring face
Accept
locating
Accept generalAccept sharing Decline
general
Decline
locating
Decline
sharing
Accept tagging Decline
tagging
Average
Agreement rate [0..1]
Referent
[name]
Most preferred (1
st
, 2
nd
)
gestures [id] in this study
Low agreementMedium agreement
High agreement
6-FSF - Flat swipe
in front of face
7-CHS
Chest swipe
6-FSF - Flat
swipe in front
of face
35-CFG- Cover
face with hands
20-IPD-Index
pointing
down
5-POU
Pointing up
1-THU
Thumb(s) up
13-CIG - Arm flexed +
index pointing up
and rotating 360°
1-THU
Thumb(s) up
8-CLF
”C” like
finger
7-CHS
Chest swipe
9-BIX- Big X
with arms
20-IPD-Index
pointing
down
11-HAW
Grabbing
arm
9-BIX- Big X
with arms
8-CLF
”C” like
finger
7-CHS
Chest swipe
1-THU
Thumb(s) up
7-CHS
Chest swipe
16-HAG- Arm raised
forward + Closed fist
(hammer gesture)
Most preferred gestures
[name] in (Rodriguez &
Marquardt, 2017)
Opt-In
Face and head: Smile
Fingers and arms: Waving
Arms: Swipe
Torso & posture: Standing in front of it
Opt-out
Face and head: Shake head
Fingers and arms: Close fist
Arms: Pull down arms
Torso & posture: Turning away
Figure 3: Agreement rates among gestures for all referents, sorted in decreasing order of their value with the most most
preferred gestures and (below) comparison with gestures elicited for Opt-in and Opt-out in (Rodriguez and Marquardt, 2017).
for this referent should be selected. Surprisingly, its
opposed referent, i.e., Decline blurring face, received
the second highest agreement rate among participants
(M=.172), although it belongs to the medium agree-
ment range. This is the case for the next four ref-
erents until Accept sharing (M=.101). The paired
referents Accept tagging (M=.069) and Decline tag-
ging (M=.053) received the lowest agreement rates,
perhaps because participants were not familiar with
any metaphor to represent these actions, as opposed
to other ones which are more familiar for them.
AGATe (Vatavu and Wobbrock, 2015) also com-
putes a statistical test to determine whether these
rates are correlated. Since our sampling contains five
pairs of related referents, each pair with a Accept
and a Decline action, we discovered four correlations:
a statistically significant effect (p.050
) between
Accept sharing and Decline sharing (V
rd(1,N=60)
=
5.188) and between Accept general and Decline gen-
eral (V
rd(1,N=60)
= 5.188), a highly significant one
(p.010
∗∗
) between Accept locating and Decline lo-
cating (V
rd(1,N=60)
= 9.981), and a very highly sig-
nificant one (p.001
∗∗∗
) between Accept blurring
face and Decline blurring face (V
rd(1,N=60)
= 49.561),
which confirms the two high rates for this pair of
referents. All Accept (Ok) referents (V
rd(4,N=150)
=
204.373) taken together as well as all Decline (NOk)
referents (V
rd(4,N=150)
= 31.680) were also related in a
very highly significant manner (p
∗∗∗
=.001), thus sug-
gesting that the gesture classes respectively proposed
by participants exhibit a high internal cohesion and
are consistently chosen. We did not find any statis-
tically significant difference between the two groups
(female vs. male): V
bg
(2, N=30) always returned a
value p.239, n.s..
In order to establish a basic mapping between cat-
egories, we opened the pool of options to the second
most frequent gestures selected by participants. The
two most preferred gesture classes for Accept Sharing
are classes coded as THD/1 and CLF/8 respectively.
We remark that most of the users avoid a mirror ges-
ture (e.g., right hand and then left hand). The Accept
tagging referent was related to CHS/7, a gesture that
includes the chest (upper body to indicate that privacy
setting affected the whole person). For the Decline
tagging, the strong gesture coded HAG/16 represents
a hammer gesture, which could be also suitable. The
Accept locating referent used a space gesture IPD/20
(Index Pointing Down) and for the Decline locating,
hand(s) waving coded HAW/11 in order to alert the
system to avoid record user location (this is an am-
biguous use of this gesture that could be used for opt-
in commands). The Accept Blurring face referent is
marked with FSF/6 (Face Swipe). Participants repro-
duced this legacy behaviour, probably because of their
previous experience with drawing GUI applications.
Next, for the Decline Blurring face, covering the face
was the second option but it is a good fit because,
instead of a motion gesture, users have proposed a
static one CFG/35. Finally, the Accept General ref-
Eliciting User-defined Zenithal Gestures for Privacy Preferences
209
7-CHS
Chest
swipe
8-CLF
”C” like
finger
9-BIX
Big X with
arms
10-POH
Pointing
Head
11-HAW
Hand(s)
waving
12-GRA
Grabbing arm
13-CIG - Arm flexed + index
pointing up and rotating 360°
(circle gesture)
14-SWG - left arm horizontal
adduction + right arm horizontal
abduction and vice versa
(swirl gesture)
15-SPH
Sprayed hand(s)
16-HAG
Arm raised forward
+ Closed fist
(hammer gesture)
1-THU
Thumb(s)
up
2-THD
Thumb(s)
down
3-KAC - Arm forward
raised + vertical flat hand
(Karate chop gesture)
4-SHF
Shaking
finger
5-POU
Pointing
up
6-FSF - Flat
swipe in
front of
face
17-SHG
Arm forward folded
+ lateral index trigger
(shooter gesture)
18-BOI
Both indexes
pointing
body
19-ARS
Arm
raised
side
20-IPD
Index
pointing
down
21-AIG
Both arms raised
lateral + flat hands
(airplane gesture)
22-SPG
Both arms raised forward
+ flat hands + full body
spinning (spinning gesture)
23-STF
Step
forward
24-BUF
Bumping
fists
25-SLG
Both arms raised forward
+ with palms down
(sleepwalking gesture)
26-HEN
Head
nodding
27-HES
Head
shaking
28-WBA
Wave both arms
horizontally
29-SMX
Small X with
fingers
30-LMF
Leg moved
forward
31-GRL
Grabbing Leg
32-LSR
Leg side
raised
33-AOH
Arm(s)
on hip
34-BAW
Both arms forward
flexed waving with
palms down
35-CFG
Cover face with
hands
36-CLG
Clapping hands
37-SNG
Snapping
38-STB
Step
backward
39-LAS
Lateral
step
40-WAS
Move
Shoulders
41-COF
Counting
fingers
42-PRG
Both flexed arms +
flat palms together
(prayer gesture)
43-IRP
Index marking random
points
44-FIN
Finger
nodding
45-STG
Arm lateral flexed +
flat hand raised
(Stop gesture)
46-GUG
Arm horizontal flexed +
right closed fist
(Fist in front of chest gesture)
Figure 4: Classification of zenithal gestures resulting from the study.
erent was associated by participants in second term
with the circle gesture CIG/13, a motion gesture that
involves users and their environment. For the Decline
General, the big “X” was selected as the first term.
Again, it is a good fit to express the decline of all op-
tions because the gesture is clear in a zenithal view
and also can be understood and remembered by par-
ticipants in subsequent interactions with the system.
The big “X” was also elicited for Opt-out referent in a
commercial setup (Rodriguez and Marquardt, 2017).
Fig. 5 graphically depicts the mean PSSUQ score
per participant. Nine participants in the 30 sampling
assessed zenithal gestures as they elicited them in the
setup as “excellent” proposals suitable for expressing
the referents (M6). Fourteen of them assessed these
gestures with a “good” averaged score (5M6),
which represents the vast majority of them. This is
also depicted in Fig. 6, where the first bag of the
Pareto distribution covers the interval [5, 6[, while the
second corresponds to the “excellent” region with the
interval [6, 7]. These two bags already cover 80% of
the population sampling, which means a good repre-
sentativeness. Only three participants out of thirty
were not confident in the gestures they proposed by
reporting they were not suitable. P22 may be an out-
lier: the lowest score was answered for all questions.
Fig. 7 reports the results of the IBM PSSUQ ques-
tionnaire. On the one hand, most questions were an-
5.81
6.19
5.81
5.56
6.00
5.31
4.25
5.38
5.06
5.88
6.25
5.44
6.38
3.25
5.19
3.88
6.19
4.69
4.75
6.06
6.00
4.75
5.69
5.88
6.00 6.00
5.63
5.69
5.81
3.0
3.5
4.0
4.5
5.0
5.5
6.0
6.5
7.0
P1
P2
P3
P4
P5
P6
P7
P8
P19
P10
P11
P12
P13
P14
P15
P16
P17
P18
P19
P20
P21
P22
P23
P24
P25
P26
P27
P28
P29
P30
Score [Value]
Participant [ID]
Average
Good
Excellent
Bad
Figure 5: Results of the IBM PSSUQ.
Figure 6: Pareto distribution of IBM PSSUQ.
swered positively, which directly impacted the first
metric SysUse (M=5.54, SD=1.25). PSSUQ metrics
having a value greater or equal than five are consid-
ered assessed positively, even if 5 is not the median
value of the 7-point Likert scale. This happens for
SysUse, InterQual (M=5.58, S D=1.28) and Overall
satisfaction (M=5.33, SD=1.51), but not the infor-
HUCAPP 2021 - 5th International Conference on Human Computer Interaction Theory and Applications
210
0 5 10 15 20 25 30
Q1
Q2
Q3
Q4
Q5
Q6
Q7
Q8
Q9
Q10
Q11
Q12
Q13
Q14
Q15
Q16
Distribution of scores per question
I strongly disagree I disagree I am so so I agree I strongly agree
5.54
4.93
5.58
5.33
0 1 2 3 4 5 6 7
System Usefulness (SysUse: 1-6)
Information Quality (InfoQual: 7-
12)
Interface Quality (InterQual: 13-15)
Overall Satisfaction (Overall: 1-16)
IBM PSSUQ Measures
Figure 7: Results of the IBM PSSUQ.
mation quality InfoQual metric (M=4.93, SD=1.80).
This is mainly due to questions Q7 and Q8 which
were assessed rather negatively and with a wide dis-
tribution. Q7 assesses the quality of error messages,
which is an inappropriate criteria since gestures, as
they were presented, do not include any feedback in
case of error. All elicited gestures were accepted.
Similarly, Q8 assesses the error recovery, which is not
really appropriate in this case.
Regarding the location, the selected space was a
real entrance of the Computer Engineering faculty
building. The conditions of the interaction were real.
Students and passersby at any time. However, the for-
tuity audience was influenced in the time of response
of participants and in the thinking time. In future
studies, we must avoid a space with the public un-
til we have a more developed set of gestures. Most
of the participants were students of the computer sci-
ence engineering faculty. So, they have some expe-
rience with GUI interactions, but could be biased by
their background. The consequence was that some
motion hand gestures were part of the gestures pro-
posed as participants were familiar with these ges-
tures in video games. However, other types of back-
ground influenced the results. For instance, one par-
ticipant is a folkloric dancer, so the swirl gesture nat-
urally emerged and another participant, who was part
of the university soccer team, proposed leg and feet
gestures.
5 CONCLUSION AND FUTURE
WORK
In this paper, we gathered and studied the gestures
that a collective of users has suggested in order to se-
lect different privacy settings. Most of the gestures
have been reduced to simpler versions and loosely
gathered in 46 categories. Since we have avoided any
limitation in the gesture proposal. Participants have
proposed complex gestures including arms, legs, and
hand motions. Participants started to create a vocabu-
lary, especially with the use of gesture number 9 (Big
“X”) as prefix to indicate declination of specific pri-
vacy options. The Legacy bias was present in some
of the gestures and interaction suggestions. An inter-
esting discovery was that instead of producing mirror
gestures to express the dichotomy of allow/decline an
option, users changed to a different suggestion. Fi-
nally, this is an exploratory exercise that gave us a
repertory of gestures that can be used in forthcoming
studies. Also, we can introduce some extra guide-
lines (Vanderdonckt, 1995) for the participants taking
into account the current results and avoid the occlu-
sion in the zenithal plane and small gestures. Besides
that, testing Welcome and Ending gestures to open
and close the privacy preferences configuration ses-
sion should be considered. Since this was a first ex-
ploration of the zenithal gestures, we must ask par-
ticipants to take more time. Also, we must avoid the
rush times in the entrance of the building to see if we
get an alternative group of categories. In addition, in
this study most of the participants have a computer
science background. So, we should include a more di-
verse pool of users. Finally, we should include feed-
back in the mock-up interface. For instance, adding
lights in future iterations of the elicitation study to
help participants to test their proposals.
REFERENCES
Anti
´
c, B., Leti
´
c, D.,
´
Culibrk, D., and Crnojevi
´
c, V. (2009).
K-means based segmentation for real-time zenithal
people counting. In Proc. ICIP ’09, pages 2565–2568.
Ashok, A., Nguyen, V., Gruteser, M., Mandayam, N., Yuan,
W., and Dana, K. (2014). Do Not Share! Invisible
Eliciting User-defined Zenithal Gestures for Privacy Preferences
211
Light Beacons for Signaling Preferences to Privacy-
Respecting Cameras. In Proc. of VLCS ’14, page
39–44, New York. ACM.
Bergold, J. and Thomas, S. (2012). Participatory research
methods: A methodological approach in motion. His-
torical Social Research/Historische Sozialforschung,
pages 191–222.
Besmer, A. and Richter Lipford, H. (2010). Moving beyond
untagging: Photo privacy in a tagged world. In Proc.
of CHI ’10, page 1563–1572, New York. ACM.
Bostan, I., Buruk, O. T., Canat, M., Tezcan, M. O., Yur-
dakul, C., G
¨
oksun, T., and
¨
Ozcan, O. (2017). Hands
as a controller: User preferences for hand specific on-
skin gestures. In Proc. of DIS ’17, pages 1123–1134.
Cook, D. J., Augusto, J. C., and Jakkula, V. R. (2009).
Ambient intelligence: Technologies, applications, and
opportunities. Pervasive and Mobile Computing,
5(4):277 – 298.
Coyette, A., Schimke, S., Vanderdonckt, J., and Vielhauer,
C. (2007). Trainable sketch recognizer for graphical
user interface design. In Proc. of INTERACT ’07, vol-
ume 4662, pages 124–135. Springer.
Erazo, O., Rekik, Y., Grisoni, L., and Pino, J. A. (2017).
Understanding gesture articulations variability. In
Proc. of INTERACT ’17, pages 293–314, Cham.
Springer.
Grd, P., Tomi
ˇ
ci
´
c, I., and Baca, M. (2018). Privacy improve-
ment model for biometric person recognition in ambi-
ent intelligence using perceptual hashing. In Proc. of
CECC ’18, New York. ACM.
Grijincu, D., Nacenta, M. A., and Kristensson, P. O. (2014).
User-defined interface gestures: Dataset and analysis.
In Proc. of ITS ’14, page 25–34, New York. ACM.
Hoff, L., Hornecker, E., and Bertel, S. (2016). Modifying
gesture elicitation: Do kinaesthetic priming and in-
creased production reduce legacy bias? In Proc. of
TEI ’16, page 86–91, New York. ACM.
Kendon, A. (2004). Gesture: Visible action as utterance.
Kieffer, S., Coyette, A., and Vanderdonckt, J. (2010). User
interface design by sketching: A complexity analysis
of widget representations. In Proc. of EICS ’10, page
57–66, New York. ACM.
Koelle, M., Ananthanarayan, S., Czupalla, S., Heuten, W.,
and Boll, S. (2018). Your smart glasses’ camera both-
ers me! exploring opt-in and opt-out gestures for pri-
vacy mediation. In Proc. of the NordiCHI ’18, page
473–481, New York. ACM.
Kray, C., Nesbitt, D., Dawson, J., and Rohs, M. (2010).
User-defined gestures for connecting mobile phones,
public displays, and tabletops. In Proc. of MobileHCI
’10, pages 239–248, New York. ACM.
Lewis, J. R. (1995). IBM computer usability satisfaction
questionnaires: Psychometric evaluation and instruc-
tions for use. Int. J. of Human-Computer Interaction,
7(1):57–78.
Lewis, J. R. (2002). Psychometric evaluation of the PSSUQ
using data from five years of usability studies. Int. J.
of Human-Comp. Int., 14(3-4):463–488.
Lewis, J. R. (2006). Sample sizes for usability tests: Mostly
math, not magic. interactions, 13(6):29–33.
Li, L. and Ma, J. (2018). Zenithal people detection based
on improved faster r-cnn. In IEEE 4th Int. Conf. on
Computer and Com., ICCC ’18, pages 1503–1508.
Likert, R. (1932). A technique for the measurement of atti-
tudes. Archives of Psychology, 22(140):55–.
Liu, M., Nancel, M., and Vogel, D. (2015). Gunslinger:
Subtle arms-down mid-air interaction. In Proc. of
UIST ’15, pages 63–71. ACM.
Lopez, M., Pedraza, J. P., Carb
´
o, J., and Molina, J. M.
(2014). Ambient intelligence: Applications and pri-
vacy policies. In Proc. of PAAMS ’14, volume 430
of Communications in Computer and Information Sci-
ence, pages 191–201. Springer.
Magrofuoco, N., P
´
erez-Medina, J. L., Roselli, P., Vander-
donckt, J., and Villarreal, S. (2019). Eliciting contact-
based and contactless gestures with radar-based sen-
sors. IEEE Access, 7:176982–176997.
Morris, M. R., Wobbrock, J. O., and Wilson, A. D. (2010).
Understanding users’ preferences for surface gestures.
In Proc. of Graphics Interface 2010, GI ’10, pages
261–268. Canadian Information Processing Society.
M
¨
uller, J., Wilmsmann, D., Exeler, J., Buzeck, M., Schmidt,
A., Jay, T., and Kr
¨
uger, A. (2009). Display blind-
ness: The effect of expectations on attention towards
digital signage. In Pervasive Computing, pages 1–8.
Springer.
Rico, J. and Brewster, S. (2010). Usable gestures for mobile
interfaces: evaluating social acceptability. In Proc. of
CHI ’10, pages 887–896, New York. ACM.
Rodriguez, I. B. and Marquardt, N. (2017). Gesture elici-
tation study on how to opt-in & opt-out from interac-
tions with public displays. In Proc. of ISS ’17, pages
32–41, New York. ACM.
Ruiz, J. and Vogel, D. (2015). Soft-constraints to reduce
legacy and performance bias to elicit whole-body ges-
tures with low arm fatigue. In Proc. of CHI ’15, pages
3347–3350, New York. ACM.
Sadri, F. (2011). Ambient intelligence: A survey. ACM
Comput. Surv., 43(4).
Sangiorgi, U. B., Beuvens, F., and Vanderdonckt, J. (2012).
User interface design by collaborative sketching. In
Proc. of DIS ’12, pages 378–387. ACM.
Valera, A., Weiss, M., Vall
´
es, M., and D
´
ıez, J.
(2007). Bluetooth-networked trajectory control of au-
tonomous vehicles. IFAC Proc. Vol., 40(1):198–203.
8th IFAC Symposium on Cost Oriented Automation.
Vanderdonckt, J. (1995). Accessing guidelines information
with sierra. In Proc. of INTERACT ’95, pages 311–
316. Chapman & Hall.
Vanderdonckt, J., Magrofuoco, N., Kieffer, S., P
´
erez, J.,
Rase, Y., Roselli, P., and Villarreal, S. (2019). Head
and shoulders gestures: Exploring user-defined ges-
tures with upper body. In Design, User Experi-
ence, and Usability. User Experience in Advanced
Technological Environments, pages 192–213, Cham.
Springer.
Vanderdonckt, J., Roselli, P., and P
´
erez-Medina, J. L.
(2018). !FTL, an articulation-invariant stroke gesture
recognizer with controllable position, scale, and rota-
HUCAPP 2021 - 5th International Conference on Human Computer Interaction Theory and Applications
212
tion invariances. In Proc. of ICMI ’18, page 125–134,
New York. ACM.
Vatavu, R.-D. and Wobbrock, J. O. (2015). Formalizing
agreement analysis for elicitation studies: new mea-
sures, significance test, and toolkit. In Proc. of CHI
’15, pages 1325–1334, New York. ACM.
Vatavu, R.-D. and Wobbrock, J. O. (2016). Between-
subjects elicitation studies: Formalization and tool
support. In Proc. of CHI ’16, pages 3390–3402, New
York. ACM.
Villarreal-Narvaez, S., Vanderdonckt, J., Vatavu, R., and
Wobbrock, J. O. (2020). A systematic review of ges-
ture elicitation studies: What can we learn from 216
studies? In Proc. of DIS ’20, pages 855–872. ACM.
Wobbrock, J. O., Morris, M. R., and Wilson, A. D. (2009).
User-defined gestures for surface computing. In Proc.
of CHI ’09, pages 1083–1092, New York. ACM.
Eliciting User-defined Zenithal Gestures for Privacy Preferences
213