Eliciting User-deﬁned Zenithal Gestures for Privacy Preferences

Francisco J. Mart

ınez-Ruiz

1 a

and Santiago Villarreal-Narvaez

2 b

Universidad Aut

onoma de Zacatecas, Centro Hist

orico C.P. 98000, Zacatecas, Mexico

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.,

https://orcid.org/0000-0002-8842-7556

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

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

, 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

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

(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,

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

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

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

blurring face

Decline

blurring face

locating

Accept generalAccept sharing Decline

general

Decline

locating

Decline

sharing

Accept tagging Decline

tagging

Average

Agreement rate [0..1]

Referent

[name]

Most preferred (1

, 2

)

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

(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)

2-THD

Thumb(s)

down

3-KAC - Arm forward

raised + vertical flat hand

(Karate chop gesture)

4-SHF

Shaking

finger

5-POU

Pointing

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 (M≥6). Fourteen of them assessed these

gestures with a “good” averaged score (5≤M≤6),

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

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

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

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.

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