Personal and ubiquitous computing vol 9 issue 1 jan 2005

An ex-plicit location model is required for symbolic coordi-nates as they do not provide implicit distance functions.Systems based on geometric coordinates can benefitfrom such a model as

O R I G I N A L A R T I C L E

Andrea Szymkowiak Æ Kenny Morrison Æ Peter Gregor

Prveen Shah Æ Jonathan J Evans Æ Barbara A Wilson

A memory aid with remote communication

using distributed technology

Received: 9 October 2003 / Accepted: 15 February 2004 / Published online: 1 April 2004

Springer-Verlag London Limited 2004

Abstract Electronic memory aids have been used

suc-cessfully to give reminders to individuals with memory

problems These aids usually present short action

reminders that are acknowledged by the user The recent

enhancement of handheld computers with wireless

technology has rendered them multi-functional and

presents an opportunity to be exploited to meet the

de-mands of the user This paper describes the architecture

of an electronic memory aid system we have developed

and are currently evaluating with memory-impaired

participants In addition to providing action prompts,

the developed system allows data entry not only on the

device itself, but also from other stations Hence, the

memory-impaired user and third parties can remotely

enter data into the device, depending on the skills of the

user The system also remotely monitors users’

acknowledgements of reminders and allows third parties

to initiate further actions where appropriate

Keywords Elderly Æ Memory-impaired users Æ Personal

digital assistant Æ Remote communication

1 Introduction: designing for non-average people

Memory problems are often associated with ageing [1]and they are among the most common effects of braininjury Electronic memory aids have been successfullyused to provide action cues to people who haveproblems remembering everyday tasks such as taking ashower or preparing tea [2] The provision of shortcues is usually sufficient rather than a detaileddescription of the action [3] to remind users of thetask The efficiency of electronic devices as memoryaids has been evaluated for numerous devices such as

a pager [4], mobile phone [5], the Voice Organiser(handheld dictaphone) [6], and handheld computers orpersonal digital assistants (PDAs) [7, 8, 9, for a reviewsee 10] The pager, which requires the touch of abutton to acknowledge an action prompt, is effectivefor the memory-impaired user because it is simple touse [11] However, learning to use electronic organisersoften produces great problems for this user group [12]and may require prolonged or repeated training [8] Inconjunction with research that shows that memory-impaired people benefit from errorless training proce-dures [13], it is evident that the training required tolearn how to use the electronic memory aid should beminimal and should produce as few errors as possible

In contrast, current time-management software ning on PDAs requires at least some training for theaverage user Although the ease of use of such soft-ware applications varies across the range of devicesavailable and the platform used (e.g EPOC/PalmOS/PocketPC/Symbian), they are not designed for mem-ory-impaired or elderly people This is crucial giventhe suggestion that memory impairment reduces theability of users to build conceptual models of theworking interface [14] and that older people mightexperience a decline in working memory [15] The needfor a customised interface for specific user groups is,however, highlighted by other researchers [9] Theirstudy showed that customising the interface of PDAs

run-A Szymkowiak (&)

Division of Psychology,

School of Social and Health Sciences,

University of Abertay Dundee, Dundee,

P Shah Æ J J Evans Æ B A Wilson

The Oliver Zangwill Centre,

Princess of Wales Hospital, Lynn Road,

Ely, Cambridgeshire, CB6 1DN, UK

DOI 10.1007/s00779-004-0259-x

for a group of brain-injured users enabled them to

successfully use these devices as memory aids

Based on these reported findings we surmise that the

use of electronic devices as memory aids presents a

challenge for the memory-impaired users Two major

factors to be considered are the ease of the interaction

with the device and, related to this, the degree of

cus-tomisation required for a particular user group

Addi-tional usability issues for the use of small portable

devices are also essential: a report on WAP usability

(Wireless Application Protocol—the technology used to

access the Internet from mobile phones) [16] gives

de-tailed evidence of the problems of creating usable

sys-tems for small screens found on mobile phones and

PDAs for average users Scrolling pages, screen layout

and the use of images and text all contribute to a

difficult usability problem which can only complicate

the use of these technologies as an external memory aid

for the non-average, e.g the elderly or

memory-im-paired user The elderly user group may also experience

declining visual acuity, contrast sensitivity and reduced

sensitivity to colour, particularly blue-green tones (for

a review see [17]), all of which make a small PDA

interface difficult or impossible to see When combined

with difficulties in control of fine movement [18] and

the impact this would have on the ability to manage a

small touch screen device, older, memory-impaired

people present a user group with very specific needs in

this design area

The readiness of users to take up assistive technology

is naturally another component that affects its use

Be-sides insight into the need to use a memory aid [9], the

support of family and professional carers can be of great

importance [19] The possibility of integrating relatives

or professionals in the assistive process may facilitate the

use of new technology It is suggested that this can be

achieved with the implementation of a remote

commu-nications system, the features of which we will discuss in

the next sections

2 A memory aid with remote communication

We have developed a memory aid system that we are

currently evaluating with memory-impaired people in a

rehabilitation clinic The memory aid system consists

of a PDA that the memory-impaired user can operate

but that can also be remotely accessed by third parties

on a PC with access to the Internet In the latter case,

the user would contact the administrator or carer to

enter data into a central database on the Internet for

them and thus is in control of which data are entered

Data are periodically synchronised between the device

and the central Internet database, and thus can be

looked up by the memory-impaired person on the

device as well as by a third person monitoring this

database In the following section we will describe the

rationale used in the design of the PDA as well as that

behind the idea of remote communication

2.1 The design rationale for the PDA as memory aidFor the memory-impaired or elderly user, the factorsdiscussed in the introduction suggest an interface withclearly displayed functionality that minimises the load

on working memory This implies intuitive usability thatresults in minimal training and visibly maintains thestructure of the system at all times, avoiding the use ofdeep menu structures, as these are problematic for el-derly users [20] Given the hardware limitations of smallhandheld devices, such as reduced display size, thispresents a major design challenge The memory aid/PDA that we are currently using, a Siemens SX45, isrunning the Windows CE 3.0 Pocket PC operating sys-tem The PDA is equipped with a 240· 320 pixels (ap-prox 60 · 78 mm) touch screen, a non-reflective TFTLCD with 65,536 colours The PDA weighs about 300 gand its dimensions are 124 · 87 · 26 mm We decidedthat four major functions should be clearly visible on thedefault display A menu structure was not deemed useful

to avoid users becoming disoriented on the small play The device in Fig 1 depicts the interface of thePDA we are currently evaluating with users Four major

dis-Fig 1 Shown is a picture of the prototype interface currently evaluated The device comprises four major functions, 1) looking at today’s entries, 2) looking at a calendar to select a day, 3) accessing diary info (such as birthdays and people), and 4) modifying the diary, which can be accessed by tapping the virtual ‘‘buttons’’ on the interface

2

functions, 1) looking at today’s entries, 2) looking at a

calendar to select a day, 3) accessing diary info (such as

birthdays and people), and 4) modifying the diary have

been selected It was suggested that the user would

mainly be interested in viewing today’s task, as this is the

behaviour that is most relevant for them Thus, the

button to activate this function was placed in a

promi-nent position on the display

The user can go backwards or forwards by a day by

tapping the buttons labelled ‘‘-Day’’ and ‘‘+Day’’

respectively A click on the clock in the right upper

corner displays the current time The user can always go

back to this default display, irrespective of where they

are, by the tapping of, at the most, two buttons

Scrolling through the Entry list in the middle of the

screen allows users to browse through entries The

length of the list depends on the number of entries and

therefore is dependent on the habits of a particular user

For all other displays involving the scroll bar the

num-ber of pages to scroll through has been limited to at most

2 1/2 to keep scrolling to a minimum

It is suggested that the acknowledgment of action

prompts by the user allows easy interaction with the

device, usually by the tap of a button upon receiving a

text or voice message As handheld computers are

usu-ally equipped with touch screens, virtual buttons that are

sufficient in size can be displayed on the screen Thus, to

a certain extent, the design of the interface can actually

circumvent hardware limitations such as tiny buttons

As depicted in Fig 2, the user has two alternatives in

responding to an alarm occurring together with a minder: either to acknowledge it with no furtherreminders or to acknowledge it with the option to bereminded again by tapping one of the two ‘‘buttons’’.This gives the user control over the presentation ofreminders and accounts for situations in which the usercannot comply with an action prompt because of situ-ational factors; for example, the user could be using thebathroom or sitting in a movie theatre when an alarmoccurs Being able to defer a reminder allows the user toselect a place and time to act on the reminder

re-Acknowledging action prompts appears to be a ple task However, interacting with the device to enterevents or data appears to be a greater challenge, as thereliance on menu structures, even if shallow, hides some

sim-of the functions sim-of the device at certain times, whichmight be problematic for users with memory problems.While we strive to obtain maximal usability in the design

of the PDA as a memory aid, depending on the teristics of the user, wireless technology can facilitate theinteraction between the user and the device by creatingflexibility for data entry Generally, memory-impairedusers have carers or relatives who also function as carers.Usually, individuals with severe memory problems reg-ularly visit a clinic in which administrative and/or healthcare personnel can take care of their needs Thus, tasksharing between the user and carers or administrativestaff regarding the interactions with the PDA/memoryaid can be achieved, alleviating the problems users mighthave when learning how to use the device In particular,this task sharing can be achieved by allowing carers oradministrative staff to access the PDA over the Internet,while the user is interacting with the device, therebycreating a remote communication system The systemthat we have developed incorporates this facility and wewill discuss its architecture next

charac-2.2 The architecture of the remote communicationmemory aid

The recent development of PDAs using wireless nology has rendered them multi-functional and presents

tech-an opportunity to be exploited to meet the demtech-ands ofthe user For example, in addition to providing actionprompts, a PDA could allow data entry on the deviceitself but also from other stations (see Fig 3) Hence,users, carers or even administrative staff can enter dataremotely into the device, thus creating the flexibility ofdata entry depending on the characteristics and needs ofthe user

A typical interaction scenario could be as follows: thememory-impaired user has acquired the skill toacknowledge reminders by tapping a button on the de-vice, but data entry presents a challenge to be met at alater stage For the present, a carer or administrativestaff can enter reminders (for example, a reminder fortonight’s dinner) into the user’s PDA using a browser onhis PC to enter the reminder into a database on a server

Fig 2 Shown is a picture of the prototype interface currently

evaluated The user has two alternatives in responding to an alarm

occurring together with a reminder: either to acknowledge it with

no further reminders or to acknowledge it with the option to be

reminded again by tapping one of the two ‘‘buttons’’

In this case the memory-impaired user would contact a

carer to enter the data for them The data entries are

synchronised between the server database and the PDA

at specific times, thus allowing for relatively short

intervals between the time of data entry and the time the

prompt is due Repeated action prompts on a daily or

weekly basis can be entered into the device, as users

often request reminders for routine actions such as

taking medication, and is functionality which has been

implemented in recent memory aids [9, 21]

Once the user has mastered one particular function of

the device, s/he might attempt to learn additional

func-tions For instance, s/he might enter reminders or other

data such as details on individuals, thus reducing the

need for the intervention of the carer or administrative

staff In addition to being able to enter reminders from

various stations, the fact that reminders can be entered

using an ordinary PC circumvents the problems of small

display sizes to enter detailed information For example,

once the user has mastered data entry functions, s/he

could enter contact details for a particular person such

as their name and phone number into the PDA If more

details are required, the carer or even the user can enter

this additional information using a PC with a large

screen size Thus, different data can be entered using

different devices but the user of the PDA can access

them at any time The user controls who enters which

data, depending on their needs and skills

We have discussed issues pertaining to data entry into

the device in the previous section However, having

distributed technology also allows carer or

administra-tive staff to monitor remotely if a user has acknowledged

a reminder and—if that is not the case—to initiate

fur-ther actions such as contacting a carer A typical

sce-nario could be the user failing to acknowledge a

reminder prompting him/her to take medication The

device would then prompt the user again to take

the medication If the user still does not acknowledge the

reminder, the device would automatically update the

server to indicate that this reminder has not beenacknowledged Once this has happened, administrativestaff could take further action such as calling the user orthe carer At present, the acknowledgement of remindershas to be checked by a user (administrative staff) whocan take appropriate action and contact a carer if nec-essary but these tasks could be further automated insuch a way that actions can be initiated from the serverstation, without the intervention of a third party Forexample, once the system has detected that the user hasnot responded to a high priority reminder (e.g ‘‘take theyellow heart pill now’’) a call could be automaticallyinitiated on the server end, contacting a carer with anautomated call to his (mobile) phone to ask him tocontact the user This functionality provides the carerwith the means to monitor the user’s behaviour if nec-essary and may thus alleviate the worry of the carerwhen the user does not acknowledge a particular re-minder It should be noted that the memory aid shares adrawback with most other electronic prompting devices,i.e that acknowledging a reminder does not necessarilyequate to actually executing the action The user couldacknowledge an action reminder, but then, being tem-porarily distracted by other events, still forget to executethe action or even intentionally omit to execute it Thedevelopment of systems that monitor behaviour, such as

in telemonitoring systems [e g 22] and ‘‘smart’’ homes,integrated with the use of prompting devices may be apossible answer to this challenge

3 ConclusionsThe review on the usability of electronic devices asmemory aids in this paper highlights difficulties formemory-impaired users in the learning required to usesuch devices In addition, other researchers have pointedout serious usability challenges with respect to design ofWAP-based interfaces for the average user However,

Fig 3 The architecture of the

memory aid (PDA) with remote

communication Data entry and

monitoring can be achieved

remotely from various stations,

by administrative personnel,

carers or relatives equipped

with PCs with access to the

Internet Both data entry and

monitoring are achieved by

synchronisation between the

PDA and the server

4

the degree to which the device allows task sharing

be-tween the user and a carer or administrative staff can

greatly facilitate the challenges users encounter when

using the device This can take the pressure off the user

to use all functions of the device at once Instead, the

users can slowly get accustomed to the functionality of

the device at their own pace and have at the same time

the assurance that carers can take care of their needs,

while they take advantage of the functionality with

which the device provides them We conclude that there

is great potential in the use of recent technologies by

non-average users such as elderly or memory-impaired

people, if customisation and interface design reflect the

needs and demands of the user Automatically

contact-ing the carer if the user fails to acknowledge reminders

would be a further extension of our memory aid system,

and is to be implemented at a later stage Exploiting

distributed technology and the advantages it provides in

data entry and monitoring can contribute to a usable

system With the advance of multimodal devices that

allow picture/video transmission in addition to text- and

graphics-based information, the future potential of

dis-tributed communication systems for social care is huge

At present, our system is mainly graphics based but new

technologies should also allow the use of speech or audio

based interactions Evaluating our current system with

users is showing the extent to which we have produced a

usable and efficient memory aid system that may

incorporate more sophisticated technology in the future

Given that memory impairments are related to ageing

[1] combined with an increase in the number of older

adults from 11 to 14 million within the next 25 years in

Britain [23], the use of technologies which can maintain

independence for older and memory impaired people

may result in huge savings for the health care system

Acknowledgements This research is funded by Grant 2006/394 from

the Health Foundation, Older People Programme.

References

1 Huppert FA, Johnson T, Nickson J (2000) High prevalence of

prospective memory impairment in the elderly and in

early-stage dementia: findings from a population based study App

Cog Psych 14:S63-S81

2 Wilson BA, Evans JJ, Emslie H, Malinek V (1997) Evaluation

of NeuroPage: a new memory aid J Neurol Neurosur Psych

63:113–115

3 Harris JE (1992) Ways to help memory In: Wilson BA, Moffat

N (eds) Clinical management of memory problems, Chapman

& Hall, London

4 Wilson BA, Emslie HC, Quirk K, Evans JJ (2001) Reducing everyday memory and planning problems by means of a paging system: a randomised control and crossover study J Neurol Neurosur Psych 70:477–482

5 Wade TK, Troy JC (2001) Mobile phones as a new memory aid: a preliminary investigation using case studies Brain Inj 15:305–320

6 van den Broek MD, Downes J, Johnson Z, Dayus B, Hilton N (2000) Evaluation of an electronic memory aid in the neuro- psychological rehabilitation of prospective memory deficits Brain Inj 14:455–462

7 Kim HJ, Burke DT, Dowds MM, George J (1999) Utility of a microcomputer as an external memory aid for a memory-im- paired head injury patient during in-patient rehabilitation Brain Inj 13:147–150

8 Kim HJ, Burke DT, Dowds MM, Robinson Boone KA, Park

GJ (2000) Electronic memory aids for an outpatient brain jury: follow-up findings Brain Inj 14:187–196

in-9 Wright P, Rogers N, Hall C et al (2001) Comparison of pocket-computer memory aids for people with brain injury Brain Inj 15:787–800

10 Inglis EA, Szymkowiak A, Gregor P, Newell AF, Hine N, Shah

P, Evans JJ, Wilson BA (in press) Issues surrounding the centred development of a new interactive memory aid Int J Univ Acc Info Soc

user-11 Wilson BA, Emslie HC, Quirk K, Evans JJ (1999) George: learning to live independently with NeuroPage Rehab Psych 44:284–296

12 Wilson BA, Moffat N (1984) Rehabilitation of memory for everyday life In: Harris JE, Morris PE (eds) Everyday memory, actions and absent-mindedness, Academic Press, London

13 Evans JJ, Wilson BA, Schuri U et al (2000) A comparison of

‘‘errorless’’ and ‘‘trial-and-error’’ learning methods for teaching individuals with acquired memory deficits Neuropsych Rehab 10: 67–101

14 Zajicek M, Morrissey W (2001) Speech output for older ally impaired adults In: Blandford A, Vanderdonckt J, Gray P (eds) Interaction without frontiers, in: Joint Proceedings of HCI 2001 and IHM 2001, Lille, France, September 2001

visu-15 Salthouse TA (1994) The ageing of working memory psychology 8:535–543

Neuro-16 Ramsey M, Nielsen J (2000) WAP usability de´ja` vu: 1994 all over again Nielsen Norman Group, Fremont, CA

17 Hawthorn D (2000) Possible implications of aging for interface designers Interact Comput 12:507–528

18 Vercruyssen M (1996) Movement control and the speed of behaviour In: Fisk AD, Rogers WA (eds) Handbook of human factors and the older adult, Academic Press, San Diego, CA

19 Kapur N (1995) Memory aids in the rehabilitation of memory disordered patients In: Baddeley AD, Wilson BA, Watts FN (eds) Handbook of memory disorders, Wiley, Chichester, UK

20 Freudenthal D (2001) Age differences in the performance of information retrieval tasks Behav Info Technol 20:9-22

21 Wright P, Rogers N, Hall C, Wilson B, Evans J, Emslie H (2001) Enhancing an appointment diary on a pocket computer for use by people after brain injury Int J Rehab Res 24:299– 308

22 Doughty K, Costa J (1997) Continuous automated telecare assessment of the elderly J Telemed Tele 3:23–25

23 HMSO (1998) Population trends Her Majesty’s Stationary Office, London

O R I G I N A L A R T I C L E

James F Knight Æ Anthony Schwirtz Æ Fotis Psomadelis

Chris Baber Æ Huw W Bristow Æ Theodoros N Arvanitis

The design of the SensVest

Received: 29 September 2003 / Accepted: 15 March 2004 / Published online: 26 June 2004

Springer-Verlag London Limited 2004

Abstract The SensVest is an item of wearable technology

that measures, records and transmits aspects of human

physical performance such as heart rate, temperature

and movement The SensVest has been designed for use

by science teachers and students to meet their

require-ments This paper reports the stages undertaken to

design the SensVest, from determining appropriate

methods of assessing human performance, to

con-siderations of mounting the technology on the body

Trials have shown that concessions need to be made

with ease of use and cost to ensure that the data

col-lected is reliable and usable, with an awareness of the

sensors’ limitations By designing the SensVest with the

wearer in mind a system has been developed that is

comfortable, does not inhibit normal performance and is

wearable User trials have shown that meaningful,

reli-able and useful data can be collected using the SensVest

Keywords SensVest Æ Ergonomics Æ Wearable

computer Æ Design Æ Energy expenditure

1 Introduction

This paper presents work undertaken as part of a

European project to investigate science teaching in

schools The paper is split into two main parts The first

part sets up the aims of the project and discusses the

rationale for choices made in deciding what technology

will be incorporated in the wearable so that it meets the

aims of the project The second part discusses the design

and ergonomic considerations in producing the wearableand the evaluations of the prototypes before presentingthe final design

2 Part 1: Requirements of the SensVest

2.1 The SensVestThe SensVest was originally designed for use in the Lab

of Tomorrow project.1 While the Lab of Tomorrowfocussed on the collection of data for education, it isproposed that the resultant design offers a means ofcollecting data in a wide variety of work-related studiesand could be of interest to ergonomists, sports scientistsand members of the medical profession

There are three ways in which data from real worldactivities can be recorded The first is to effectively makethe world the laboratory This requires that the envi-ronment be fitted with sufficient sensors to monitor,measure and record human activities, e.g camera-basedsystems Commercially available vision systems aretypically very expensive, although the Lab of Tomorrowconsortium has successfully developed a low-costtracking system However, these systems require thatactivity be performed within the calibration volume ofspace covered by the vision system, which has implica-tions for the type of activity that can be analysed.Furthermore, the vision system could require post-pro-cessing to be able to derive human performance datafrom the images Alternatively, people could be boughtinto a laboratory and asked to perform examples ofeveryday tasks While such an approach might be morecost-effective, in terms of equipment, it produces prob-lems relating to the ‘‘naturalness’’ of activities that can

J F Knight (&) Æ A Schwirtz Æ C Baber Æ H W Bristow

T N Arvanitis

Department of Electronic, Electrical and Computer Engineering,

School of Engineering, The University of Birmingham,

Pers Ubiquit Comput (2005) 9: 6–19

DOI 10.1007/s00779-004-0269-8

be recorded in the laboratory For example, asking

someone to perform tasks under observation in the

laboratory could lead to a different activity than that

observed in the normal environment Furthermore,

people could be limited as to their range of movement in

that they must remain within the envelope of the sensors

A third approach involves people effectively carrying the

laboratory with them In this scenario the person has the

sensors to measure relevant scientific variables with

them at all times In this approach, the sensors do not

bind the person to a location and thus expands the

envelope of situations to which the sensors can be taken

The aim of this paper is to report the design aspect of

the wearable system developed for the Lab of Tomorrow

project Specifically, the paper will focus on aspects of

mounting the technology on the body and the

ergo-nomic considerations taken into account when doing

this In effect this paper demonstrates the processes

undertaken to get the appropriate technology on the

body in a suitable manner Indeed, a major aspect of this

paper is to determine what the terms ‘‘appropriate’’ and

‘‘suitable’’ mean in the context of developing a wearable

system

2.2 Defining technical requirements

Having the sensors with the person at all times was

se-lected as the approach most appropriate to measure

activity in the real world Within this approach there are

two sub-methods: the first is to have the sensors

embedded into the tools or equipment that the person

uses; the second is to attach the sensors to the people

themselves This second option is beneficial in that it

ensures that the sensors are with the person at all times,

can be used to measure activities and do not require the

manipulation of some artefact Consequently, these

sensors act passively, recording data without requiring

that the users perform some extraneous activity

Attaching technology to the body has received

con-siderable interest in the form of wearable computing

[1, 2] Within the area of wearable computing wearing

sensors is not novel A number of researchers have used

body mounted sensors to try to determine such aspects

as the wearer’s position, posture and activity state with

the aim of developing computer systems that are context

aware [for example 3, 4] Other wearable sensor systems

have been developed to measure a range of physiological

variables For example BODYMEDIA have produced

the SensWear Pro Armband, which monitors skin and

ambient temperature, the galvanic skin response and

movement from a device which is worn on the upper

arm This is marketed as a product for weight

manage-ment, wellness, assisted living, fitness, sleep monitoring

and scientific research

The SensVest differs from these other wearable

sys-tems in that it has been developed specifically as a

teaching tool As such it has been designed with the end

user in mind (i.e a student) and the use of the data

collected for the role of teaching Although the SensVestmay ultimately use the same or similar sensors as thoseused in other systems the context of their use dictatedthat a specific product be designed and developed Many

of the systems discussed above may be wearable but maynot be suitable when the wearer is engaged in strenuousphysical activity The positioning of the sensors is alsoimportant The SensWear records movement from adevice on the upper arm; this would be unsuitable forsituations where the wearer is engaged in considerablephysical activity but with little arm movement, forexample, cycling As such the design of the SensVest had

to consider where, and in what situations, the device will

be worn and also what data would be appropriate

2.2.1 Determining metricsTaking into account that the focus of the project is thedevelopment of a device for use by schoolchildren, dis-cussions with educationalists and teachers were held todetermine what aspects of performance they would like

to record From these discussions two concepts wereproposed The first involves measurements of humanmovement that can be applied to Newtonian physicswith the emphasis on assessing variables such as: force,displacement, velocity and acceleration The second was

a measure of the energy expended by a person whilecarrying out various tasks

2.2.1.1 Measuring movement The intention for ing aspects of movement in this project is to apply thedata to concepts underpinned by Newtonian physics Assuch measurements of displacement, velocity, accelera-tion and force are pertinent Measurements such as these

measur-in human movement applications come under theprovince of biomechanics

An assessment of biomechanical techniques profferednumerous ways of assessing movement However, most

of these methods involve using equipment, which isseparate from the assessed performer and are set up incalibrated areas Examples of these systems may involvedigitising a video or cine image, or they may use auto-matic opto-electronic tracking systems, which use eitherpassive body markers (for example Kinemetrics, Mac-Reflex, ELITE, Motion Analysis, Peak, Vicon andCODA) or active markers (for example, Selspot, IROSand Watsmart) However, as one aim is to be able tomeasure students performing everyday activities it wasdeemed inappropriate to restrict the student to onespecific calibrated location (i.e a laboratory or class-room) As such, these systems were rejected

The requirement was for devices that could be wornand transported into the real world This is achievablewith the use of accelerometers Accelerometers incor-porate a mass mounted on a cantilever beam or springattached to a housing As the housing accelerates,because of its inertia, the mass lags behind, deformingthe beam This deformation is measured either using

strain gauges or piezo-electric devices, which gives a

measure of the acceleration For this project

acceler-ometers were deemed most appropriate, as they are

small, lightweight, wearable and relatively cheap They

can also be used to measure the accelerations of body

segments (e.g arms and legs individually) and if

mounted on the trunk, the acceleration of the body

There are a number of considerations that must be

taken into account when using accelerometers; for

in-stance, they give no indication of a segment’s initial

condition (i.e position, orientation or velocity) They

are gravity sensitive This means that their output

rep-resents the vector sum of the gravity and kinematic

acceleration To derive accurate acceleration additional

information regarding segmental orientation is needed

Without this additional information, the user must

ei-ther assume that any movement was linear so that the

gravity component can be subtracted using a resting

value, or that the gravity component is negligible in

relation to the acceleration measured by the movement

of the body The accelerometer signal tends to drift over

time causing a low-frequency noise effect that increases

with time This means that the accelerometer has to be

regularly calibrated If attached to clothing or the skin

errors can occur due to the relative movement clothing

or skin makes with soft tissue To overcome this

prob-lem the accelerometer must be mounted on the body as

tightly as possible The accelerometer can be delicate

and is easily broken if dropped so some care has to be

taken when handling it These considerations dictate

that the wearer be aware of the limitations Ultimately,

the value of the data depends on the required accuracy

of the user

2.2.1.2 Measuring the energy expenditure The most

common method used in sports and exercise science is to

measure the amount of oxygen consumed This involves

collecting samples of gas breathed into a bag or directly

into a gas analyser This method, though, was deemed

inappropriate for this project, as the equipment is

expensive and usually fixed to one location Portable

devices do exist [5, 6] but these are cumbersome to use,

especially for long-term free movement [7] and as it

in-volves attaching breathing equipment to the face it is

uncomfortable and restricts verbal communication [8]

2.2.1.3 Measuring the heart rate Measuring the heart

rate is a commonly used measure of assessing energy

expenditure and is attractive as a direct linear

relation-ship between the oxygen uptake and the heart rate at

moderate to high intensities of exercise has been found

[9]

During physical activity aerobic respiration uses

oxygen to produce energy Calculating the amount of

oxygen consumed therefore gives an estimation of the

energy expended during physical activity As a linear

relationship has been found between oxygen uptake and

heart rate at moderate to high intensities of exercise a

measure of heart rate can be used to provide this mate More accurate estimates for the prospective users

esti-of the SensVest (i.e adolescents) would involve usingrelationships developed from the same age range Theheart rate max can be estimated using the equation:HRmax= 220 age (in years)

So, for example a 15-year old max heart rate is 205.Estimates of maximum oxygen consumption (VO2max)can be taken from published data Thus, the VO2maxfor15-year old males is 59 ml/min kg [10] With an averagebody weight of 57 kg [11] an estimated VO2max for a15-year-old male is thus 3.36 l/min

Taking an average heart rate during a physicalactivity of, say, 160 bpm, for example, this translatesinto a heart rate of approximately 78% max A heartrate of 78% max relates to an oxygen consumption ofapproximately 70% VO2max [9], which corresponds to2.35 l/min (i.e 70% of 3.36 l/min)

For each litre of oxygen consumed approximately

20 kJ of energy is liberated Therefore, an estimation ofoxygen consumption from heart rate can be used toestimate the energy expenditure For example, our

15 year old male basketball player with a heart rate of

160 bpm was expending energy at a rate of 47 kJ/min

As such, estimates of energy expenditure based on theheart rate can easily be made

2.2.1.4 Methods of heart rate measurement Table 1shows a number of ways of measuring the heart rate thatwere considered for the SensVest In essence there aretwo main methods of measuring heart rate: one is tomeasure the pulse rate from an artery (for example,inserting the microphone, the pressure bulb and theplethysmography in Table 1), and the other is to mea-sure the electrical signal that stimulates the heart tocontract (for example, the 3 lead ECG and the heart ratemonitor in Table 1) Of the two methods, that of mea-suring the pulse was initially the most attractive Thesemethods are generally cheaper and as they do notinvolve attaching technology to the skin of the chest aremore convenient in terms of pupils putting on theSensVest in environments where others may be present.For the initial prototype of the SensVest the heartrate was measured using the insert microphone and thepressure bulb These methods involved attaching sensors

to the wrist or hand and were attractive because theywere cheap and easy to use Unfortunately, being af-fected by hand and finger movements they were found to

be highly susceptible to noise during trials, and as suchthey were deemed unreliable Subsequent prototypesinvestigated the use of plethysmography, but this wasalso found to be unreliable (see Table 1) Therefore, theidea of measuring the pulse had to be discarded for that

of measuring the electrical signal of the heart Using a 3lead ECG method proved promising; however it was notvery practical Setting up the equipment and preparingthe wearer was time-consuming and was uncomfortablefor the wearer, specifically when the adhesive electrodes8

were detached from the skin Therefore, the final version

of the SensVest uses the POLAR heart rate monitor with

the Log-IT receiver Although this method is the most

expensive it is the only method that reliably recorded the

heart during dynamic activity and so met the aims of the

project

2.2.1.5 Estimating energy expenditure using body

accel-erometers Using a body-mounted accelerometer as an

instrument to measure body movement and relating

movement to energy expenditure, another method of

estimating energy expenditure can be achieved using

accelerometers [7, 12, 13] This method relies on

quanti-fying periods of accelerometer data and relating these to

heart rate or oxygen consumption to estimate energy

expenditure By quantifying the accelerometer data (e.g

by integration or the root mean square) a number

relating to an amount of movement can be determined

Regression equations can then be derived correlating

accelerometer readings with energy expenditure For

example, Meijer et al [13], have derived an equation for

energy expenditure (EE) using an integrated 1-min

peri-od of bperi-ody accelerometer output (AO) which reads as:

EE = 1.294· AO + 77.988

A similar equation could be determined using the

body-mounted accelerometer of the SensVest This

equation could then be used to determine the energy

demands brought about by solely by body movement

Heart rate can be affected by factors other than the

physical demands of movement, e.g fatigue, heat and

psychological stress However, the accelerometers are

not affected by these factors; therefore, estimations of

energy expenditure using accelerometers will give a lue purely due to the physiological demands of movingthe body This may be useful for assessing and com-paring different periods during game play (e.g basket-ball, football) The data could be used to determine ifchanges in heart rate are due to changes in bodymovement or if other factors are involved or whetherthe amount of body movement is changing during thegame, which may be a response to factors such as fati-gue As such, comparisons of heart rate and acceler-ometer data, and the levels of energy expenditureestimated by the two sets of data can thus be used togenerate an extensive picture of the physiological factorsinvolved in athletic performance

va-2.2.1.6 Measuring temperature With the finding thatmore than 75% of the energy utilised during physicalwork is converted into heat [9] and that core bodytemperature increases linearly with oxygen uptake inexercise with the arms and the legs [14] it was alsodeemed appropriate that another suitable method ofmeasuring changes in energy expenditure was to mea-sure body temperature A literature review determinedthat the best method of measuring core temperature was

by inserting thermocouples into the body through theear (tympanic), down the throat (oesophageal), into themouth (oral or sublingual) or into the rectum [15].Considering that students under the ages of 16 yearswould use the end product technology these methodswere dismissed due to concerns of safety, hygiene andethics The other option was to measure the temperaturefrom the axilla (armpit) region This method is rarely

Table 1 Measuring heart rate methods

(£)

Attachment point and method

Power consumption (mW)

Comment/findings from trials

Insert microphone Microphone detects pulse

on surface artery

<5 Wrist over the radial artery—strapped in place tightly

<1 Affected by hand and finger

movement, and incorrect pressure applied to the sensor Pressure bulb Pressure bulb detects pulse

on surface artery

<30 Finger tip—strapped firmly

<1 Affected by hand movement,

incorrect pressure applied

to the sensor and vascoconstriction Plethysmography Light sensor detects blood

flow oscillations through non-bony extremity (e.g ear, finger tip)

<30 Ear lobe—clip <110–155 Affected by head movement,

incorrect pressure applied

to the sensor, vascoconstriction and changes in ambient light

3 lead ECG Electrodes detect the

electrical signal that stimulates the heart

to beat

<30 Two electrodes attached

to the chest and one electrode to the hip—held with adhesive pads

<1 Difficult to position correctly,

need for the skin to be prepared, electrodes become loose when body is sweatly, discomfort when removing electrodes Affected by arm movement

POLAR heart rate

monitor with Log-IT

receiver

Electrodes detect the electrical signal that stimulates the heart

to beat

<100 Chest strap (against the skin)

7 Most reliable method, occasional

loss of signal from monitor

to receiver

used in exercise contexts due to the difficulty of holding

the sensor in position and as it is affected by ambient

conditions [15]

However, as the aim was to show relative changes in

body temperature with exercise it was deemed

appro-priate with the proviso that it was understood that it

represented changes in the microclimatic temperature of

the armpit, which although representative, was not the

same as the global body temperature

2.2.1.7 Other physiological measurements Other

meth-ods were considered and although appropriate for the

assessment of energy expenditure were rejected due to

various reasons For example, pulmonary ventilation

using spirometers was rejected as it usually involved

wearing a face-mask and involved non-portable

equip-ment Blood pressure was rejected as it should only be

measured when the subject is at rest and is more

appropriately used in medical examinations [16] Blood

analysis (lactate concentrations) although useful in

assessing the contribution of the anaerobic energy

metabolism pathways [5], was rejected on the grounds of

safety, hygiene and ethics In addition, specialist training

is needed in the handling of blood products and the cost

of equipment to analyse blood products is outside the

realm of the useful expenditure of a school’s budget The

galvanic skin response (GSR) is a method of measuring

the sweat response Although it may have potential uses

for assessing the thermoregulatory response to physical

activity it is more commonly used in situations to record

responses to psychological stress

2.2.2 Conclusion of the SensVest requirements

Through an assessment of the requirements of the

teachers and educationalists and a review of

physiolog-ical and biomechanphysiolog-ical techniques it was proposed that a

wearable system would be developed that recorded heart

rate, body temperature and movement with

accelerom-eters In addition to these sensors a processor,

signal-conditioning unit and display had to be attached to the

body These will not be discussed in any depth here,

as these issues will form the basis of another paper.However, a summary of their specifications is shown inTable 2

Designing the wearable device did not follow theusual methods of user centred design Instead, asthe design of the equipment was heavily dependant onthe technology, a theory of use was applied to developprototypes from which ideas and suggestions fromusers were acquired This theory of use was developedthrough exploration of possible scenarios in which theSensVest could be used, e.g by examining the sort ofexercises that might be expected to be performed usingthe SensVest and the sort of data that might arisefrom these exercises Having developed a few suchscenarios, a simple use-case diagram (based on a uni-fied modelling language) was developed This is shown

in Fig 1

Figure 1 shows the basic structure of the completesystem The aim is also to illustrate how the completeSensVest would be used in the intended applications.From this description, it was possible to produce thebasic technical specification as shown in Table 2 Thus,the ‘‘theory’’ is partly a reflection of the manner in whichthe device would be used and partly a set of technicalrequirements However, the theory should also reflectbasic ergonomic considerations of performing activitywhilst wearing such equipment

3 Part 2: Developing a wearable deviceHaving determined what technology was to comprise theSensVest the next stage was to develop the technology in

a suitable fashion to be attached to the body and todetermine how and where the equipment would be at-tached This second section of the paper describes theprocess undertaken in mounting the technology on thebody and the modifications made both to the design ofthe computational and clothing components of theSensVest to make the SenVest wearable and to ensurethat the data collected is reliable and useful

Table 2 SensVest specifications

Purpose Record physiological data from children playing sports, transmit data to host PC

Inputs Accelerometers, pulse transducer, digital thermometer

Outputs LED to flash with pulse, LCD to show history of values, packets of data for RF transmission Functions Pulse: record, flash LED, show value on LCD

Accelerometer: record Temperature: record, show value on LCD Data: packet for RF

LCD: display pulse and temperature, allow review of previous values Performance Continuous monitoring for lesson, packets sent at 10-s intervals, no data loss

Power Runs from 9 V battery for duration of lesson

Physical size To be integrated into shirt Should not impede movement, and should not impose risk on wearer 10

3.1 Locating the technology on the body

Gemperle et al [17] developed guidelines for the

wear-ability of wearable computers In this they coined the

term ‘‘dynamic wearability’’ highlighting that the devices

have to be wearable when the body is in motion For the

development of the SensVest similar considerations had

to made In terms of locating the technology on the

body, as the device had to be worn while the wearer was

engaged in physical activity, the concept of dynamic

wearability was of paramount importance In addition

though, locating the technology on the body was

somewhat bound by the restraints imposed by using

sensors that had to generate reliable and useful data As

such there was not a total free reign in selecting

appro-priate attachment points

The sensors were to be attached to the upper body;

therefore, the wearable system was designed to be

mounted on a shirt To discuss where to position the

devices a shirt was laid on a table Then, post-it notes

were attached to the shirt in locations where the sensorsand devices could be positioned (Fig 2) These positionswere discussed using three criteria:

1 The data from the sensors must be meaningful

2 Wearing the devices should not alter performance(e.g movement)

3 The shirt should be comfortable to wear

The location of the sensors was fairly well determined

by the variables they were to measure Initially, the heartrate was measured using a microphone and a pressurebulb, which requires strapping the sensor around apalpable artery There are a number of such arteries, themost commonly used being the ones in the neck on ei-ther side of the larynx (the carotid artery) and in thewrist (the radial artery) In this case the radial artery wasselected as it was easy to strap the sensors around thewrist and strapping something tight around the neck wasconsidered to be highly dangerous The temperaturesensor, as discussed previously, was to be located in the

Fig 2 Identifying the sensor

and the device locations

ud Use Case Model

Teacher / Lesson Plan

Pupil#1

Pupil#2

SensVest

Transmitter Module

PC

Define Experiment

Define Data Collection Parameters

Transmit Data

Configure Collect Data

Display Data

Send Packets of Data

Fig 1 A use case diagram of

the SensVest

armpit The location for the accelerometers was partly

based on the activities the teachers wished to assess We

were given the scenario that they would like to assess

limb movements such as throwing and general whole

body movements such as walking and running It was

decided therefore that an accelerometer would be

mounted at the wrist This would give the acceleration

data from a point as close to any object held in the hand

without being attached to the hand, which might affect

hand manipulation A second biaxial accelerometer was

mounted on the hip This would be used to measure

whole body movements in the vertical and horizontal

directions

The size and weight of the other technologies initially

determined that they should be located on the trunk as

they would inhibit arm movement or fatigue the

mus-culoskeleteal system of the arm and shoulder [18] To

mount the devices on the body a fairly flat and relatively

large surface area was needed The area also had to have

minimal flexibility that didn’t change shape when

bending or moving, as rigid articles mounted on these

areas would inhibit movement These conditions are also

highlighted by Gemperle et al [17] as being criteria for

the placement of wearable devices The locations

iden-tified as being appropriate were the upper chest, the

upper back-shoulder regions and the hips Areas

iden-tified as being inappropriate were the abdominal region

and lower chest region These areas differ slightly from

those that Gemperle et al [17] identified as being

unobtrusive For example, they suggest that the area

around the waist and hip, including the abdominal

re-gion is unobtrusive; however, for the size of the device

that was produced for the SensVest, these may inhibit

movement, especially trunk flexion They may also

ob-struct hand and arm movement across and in front of

the body

3.1.1 Initial prototype

To test the devices and the design of the shirt the first

prototype was developed (Fig 3) The devices were

mounted onto foam backing for cushioning and held in

position on the shirt with elastic straps In addition they

were held in position by being housed inside a

zip-fas-tened pouch attached to the shirt over the chest,

shoul-der and upper back Strips from the pouch were feddown the arm and side of the body to hold the wires inplace for the wrist accelerometer and pulse sensors andthe body accelerometer mounted on the hip The tem-perature sensor was attached to the armpit region of theshirt using tape Elastic drawstrings around the wristand waist of the shirt held the accelerometers andmicrophone pulse sensor in position

To test the shirt laboratory trials were carried outwhere data was recorded from students performingthrowing and whole body movement activities (forexample walking, running, jumping) In addition, thecomfort of the shirt was assessed using specially de-signed comfort rating scales [19]

Trials with the shirt showed that the locations of thedevices on the body were not inappropriate in terms ofwearability However, changes to the design of the shirthad to be made The shirt design was thick, big andbulky and even in the cooler English climate, hot towear The pouch design although adding security to thedevices on the body increased the rigidity of the shirtand with the devices in pulled the shirt out of shapemaking it feel uncomfortable Although there weredrawstrings on the wrist and waist there was nothingholding the devices mounted on the chest and back,which meant that the devices moved uncomfortablyagainst the body, especially when carrying out a dy-namic activity such as running In addition putting onand taking off the shirt was cumbersome, as wasremoving the devices from the shirt so that it could bewashed

3.2 The SensVestFrom testing with the first prototype and discussionswith potential users at workshops a number of designmodifications were made The first major alteration was

to change from a shirt design to a vest design The designwas based on a modified running vest, which was muchlighter, less bulky and cooler to wear than the originalprototype To aid in putting on and taking off the vest acut up the front was made so that it did not have to beput on over the head Velcro straps across the frontclosed the shirt and as they were adjustable they could

Fig 3 Initial prototype with

the display mounted on the

chest and processor and the

signal-conditioning unit

mounted on the upper-back/

shoulder region

12

be pulled tight to hold the shirt firmly against the body.

The smaller vest design was such that it could be worn

under or over a normal shirt

Modifications made to the devices meant that the

display could be detached from the processor after the

equipment had been calibrated and recording started

This meant that it was not necessary to wear the display

and had the added benefit of removing the technology

mounted on the upper chest, which had the potential to

be uncomfortable especially for female users To attach

the processor and signal-conditioning unit the pouch

design was replaced with two pockets mounted on the

upper-back/shoulder region similar to the initial design

Holding the leads for the pulse sensors and arm

accelerometer in place was initially done using surgical

tape both at the site of sensor attachment (i.e the wrist)

and further up the arm to stop the leads flapping and

inhibiting movement Modifications to this design

in-volved attaching the sensors to wrist sweatbands The

body accelerometer was attached to the side of the vest

with tape such that it measured acceleration in the

sag-ittal plane (i.e horizontally in forward and backwards

directions, as opposed to side-to-side and vertically)

The temperature sensor was again attached to the

arm-pit region of the vest using tape

3.2.1 Evaluating the SensVest

Testing of the SensVest included carrying out laboratory

trials on a treadmill and cycle ergometer (Fig 4) and

also demonstrating it to teachers and educationalists

associated with the project during a teachers workshop

held in Athens in 2002 (Fig 5) The evaluation of the

SensVest also included a comfort assessment using the

comfort scales reported in [13]

The trials showed that the modifications made to the

design of the SensVest were favourable The reduction in

size, weight and bulk resulted in improvements in wearer

comfort including perceptions that the new design

inhibited movement less than the shirt design The

adjustability of the size of the vest meant that the

components worn on the upper-back/shoulder region

could be held firmly in place and did not collide with thebody causing discomfort when running

The teachers were satisfied with the modificationsmade to the vest especially in terms of reductions of sizeand bulk and aesthetics Indeed, one concern with theoriginal prototype shirt for them was that it would betoo hot to be worn, especially in Greece and Italy, and as

a result students would not wear it The teachers lieved that making the vest smaller and lighter and thusmore wearable in terms of comfort increased the likeli-hood that the students would accept the SensVest, andthus it would be more usable as a teaching tool

be-3.2.2 Technological modificationsDuring the trials the flaws with the insert microphoneand pressure bulb to measure the heart rate were iden-tified (see Table 1) At this stage trials with the ECGmethod began The technical development of the ECGsystem involved adding an extra unit to be worn on theupper-back/shoulder region, which connected to theprocessor and three extra leads that fed from the ECGmodule to the chest and hip regions

Further technical developments included the nation of the display, processor and signal-conditionerinto one unit (Fig 6) In addition the SensVest was alsorequired to accommodate a communications unit,developed by partners on the project from ANCO, S.A.Greece, which would transmit the data collected to aremote base station

combi-The design of the all-in-one-unit was shaped asshown in Fig 6 This shape is similar to that of the rightscapula (shoulder blade), over which it is positioned.The intention was for the unit to be worn snugly againstthe shoulder and the shape was chosen to support acomfortable fit The locations of the attachments of thesensors and cables for the other units were positioned tominimise the cabling required to connect the SensVestnetwork ‘‘A’’ is the arm accelerometer, which feeds upover the shoulder and down the arm to be attached to

Fig 5 The SensVest demonstrated at a teacher’s workshop in Athens, in 2002

Fig 4 Cycle and treadmill trials with the SensVest

the wrist It can be attached to either the right or left

wrist depending on arm dominance but is initially set up

for right-handed users ‘‘B’’ is the body accelerometer,

which it attached to the trunk of the body to assess

movement in the sagittal plane The LED is a light

emitting diode, which flashes with every heart beat to

give an indication that the heat rate monitoring

equip-ment is operational ‘‘Temp’’ is the temperature sensor,

which is positioned under the right armpit ‘‘ECG’’ is the

connector the ECG unit worn on the left upper-back/

shoulder region ‘‘Comms’’ is the connector to the

communications unit, which is worn in a pocket located

on the left side of the lower back

3.2.3 Developments and modifications

As mentioned above the development of the technology

and user requirements led to considerable changes of the

design of the SensVest, specifically the addition of

pockets mounted on the lower back Other

consider-ations though, also accounted for modificconsider-ations in

de-sign These were mainly aesthetic but also served some

functional purpose

The vest was made longer so as to cover the

abdominal region This would ultimately be required so

as to accommodate the addition of lower back pockets,

but was also a requirement to enable wearers to feel

socially comfortable when wearing the vest with no shirt

over or under the vest Although the Velcro fastenings at

the front of the shirt enabled adjustments to be made to

different sizes of wearer there was some concern as to the

stability of the fastening Therefore, a zip was added to

the front of the shirt, adjoining both sides This had the

added benefit of being aesthetically pleasing as it

closed the shirt, removing gaps left between the Velcro

fastenings

Fixing the wrist-mounted accelerometer and the leads

down the arm with surgical tape was considered

inap-propriate and disconcerting to wearers considering the

discomfort experienced when the tape was removed.Therefore, adjustable Velcro straps were incorporated inthe design to hold these components in place Usingthese straps has the virtue that as they are adjustablethey can be worn by anyone They are reusable and soare not wasteful of resources, and they relinquish littlediscomfort when they are attached or removed

Figure 7 shows the modified SensVest being used in anumber of sporting activities Testing the modified de-sign in this fashion showed that the vest holds the de-vices firmly in place and enables the full range of arm,leg and body movement

3.3 Final SensVest designThe initial SensVest and the modified version showedthat a vest with four mounted pockets on the back was asuitable design to house the components necessary tomeasure, record and transmit variables associated withthe assessment of energy expenditure and movement asdictated by the requirements of the project’s aims Theprevious designs had been based on modifying normalcommercially available shirts The final stage was toproduce a purpose built vest

Figure 8 shows the latest design of the SensVest Thevest was made from 100% polyester twill chosen for itsstrength, durability and ease of washing The patternwas based on a Butterick 6843 paper pattern for a man’ssleeveless zipped jacket Modifications to the patternincluded making the cut shorter to obtain a closer fit atthe waist and a higher ‘‘V’’ neckline and armholes.The size of the vest was loose fitting for most wearersand could be pulled tight around the body usingadjustable straps To determine the size to which the vestwas cut, anthropometric data for the upper end age

Fig 7 Modified SensVest used in a number of sporting acitivites Fig 6 Processor, signal-conditioner and display in one unit

14

range of the prospective users was employed This data

was used to generate benchmarks for the dimensions of

the vest, such that the vest should be larger than these

dimensions to ensure a loose fit The prospective users of

the SensVest are those engaged in the latter stages of

secondary education and beginning higher education

(i.e 16-year-olds) Therefore, anthropometric data for

the 95th percentile of this age group was used as the

benchmark, as this would ensure that the majority of

16-year-olds would fit inside the vest The body dimension

data used to determine the benchmark for the cut of the

vest, and the cut used for the SensVest is shown in

Table 3

The front of the vest is similar to the previous

Sens-Vest design with an open-end zipper down the middle

Three adjustable clip straps across the front pull the shirt

tight to the body ensuring stability for the devices on the

body The straps are constructed from 1-in nylon

webbing, chosen for its strength and non-slip

charac-teristic Push-fit buckles on the straps allow easy size

adjustment An additional feature to the previous design

is a fold back clip on the loose end of the strap, which

secures the excess strap against the body to stop it

flapping

Four pockets are mounted on the back The upper

right side pocket was designed to house the processor

The upper left pocket was originally designed to house

the communications unit and the ECG unit Later trials

showed that the ECG was unreliable during dynamic

activity and due to other problems (see Table 1) wasreplaced with the POLAR heart rate monitor and LogITreceiver (the heart rate receiver (HRR) in Fig 11) Tomaintain as consistent an uninterrupted signal as pos-sible the wire connecting the LogIT receiver to theprocessor was fed through the holes in the vest originallydesigned for the ECG leads, such that the receiver could

be attached to the strap that holds the body mountedaccelerometer in place in a position close to the heartrate monitor Further technical modifications meantthat a rechargeable battery pack (PS in Fig 11) for theSensVest was added This then replaced the ECGmodule in the upper left pocket

The lower pockets were designed to house futureadditions of a leg accelerometer unit (this would attach

Fig 8 The latest SensVest

design

Table 3 95th percentile body dimensions for 16–17.5-year-olds for the SensVest

Male Female SensVest

Shoulder breadth (biacromial) 42 33 44 Chest circumference (at armpit) 99.9 a 110 Waist circumference (at navel) 83.4 75.7 110 Torso length (from shoulder to hip) 52.5 52 55

NB Data adapted from Pheasant (1998) for 16-year-olds, except

a Synder (1977) for 16.5–17.5-year-olds (average of both sexes) All data cited in [11]

to L in Fig 11), which could fit into either pocket

depending on leg dominance The size of these pockets

was determined by using actual size foam replicas of the

equipment Tucks were incorporated into the pockets

design to produce a shape that accommodates the

de-vices securely A Velcro fastening flap across the top

edge of the top pockets closes the pockets The lower

pockets are also closed with Velcro (Fig 9)

Previous vest designs had the wiring outside the shirt

This posed potential risks of users catching the vests on

extraneous objects A network of channels, which opens

into each pocket, was sewn into the vest through which

the leads could be passed (Fig 10) The channels were

constructed from 1.5-in cotton tape Strengthened exit

holes are located around the network, through which the

ends of the sensor leads can be passed

The design of the vest is such that all the technology

is hidden from view, except the accelerometer sensors,

which pass down the limbs (Fig 11)

As suggested above some changes were made to theSensVest devices subsequent to the design and con-struction of the vest In addition, with respect to thelayout of the vest, the attachment placement of thesensors and other devices to the main unit were modified(Fig 11) This included moving the body accelerometer(B) to the underside as this would be fed under the arm

to be strapped around the body The LogIT HRR, whichreplaced the ECG unit was also attached to the under-side as it would also be strapped to the chest, under thearm The Comms attachment and the rechargeablepower supply (PS) were fed from the left side of the unit

so that they could be placed into the left pocket

3.4 User trialsUser trials with the SensVest have taken place at aschool in Wolverhampton, England In these trials,

Fig 9 Design of the back panel

16

students of physical education have carried out a

num-ber of sporting activities while having data recorded

from them Initial trials were carried out to demonstrate

the technology to potential end users and to get

feed-back from them In addition these trials enabled the

designers to study the performance of the devices and

the data collected in sporting activities such as team

game play (e.g basketball), individual game play (e.g

tennis) and for athletic disciplines (e.g sprinting, javelin,

discus and shot) These initial trials proved promising;the students were able to carry out their activities with

no apparent inhibition to movement and little fort and the data were reliably collected Based on theseinitial trials a second trial was carried out whereby theSensVest was used in a teaching context The SensVestwas used to record body movement using the body-mounted accelerometer and heart rate while a studentplayed football (Fig 12)

discom-By considering the body accelerometer and heart ratedata together (Fig 13) the responses of the heart toexercise were successfully demonstrated This includeddemonstrating such phenomena as follows: an increase

in heart rate above resting values before exercise isstarted as a result of an early release of adrenalin, which

is known as the anticipatory rise; that heart rate tuates during game play with an increase as exerciseintensity increases and a decrease as exercise intensitydecreases; that heart rate decreases rapidly immediatelyafter exercise stops; and that the heart rate continues todecrease but at a slower rate and remains elevated forsome time as the body recovers and pays off the oxygendebt

fluc-4 Conclusions

Developing wearable technology requires continuoustradeoffs to be made between technical and ergonomicrequirements, as well as considerations relating to tex-tiles, the manufacture and the context of use Inreporting a complete design process, this paper hasdemonstrated some of the salient ergonomic factors thathave affected design

The first set of factors relate to the practical andtheoretical issues surrounding the recording and use ofdata pertaining to human movement, specifically interms of energy expenditure An assessment of physio-logical and biomechanical recording techniques deter-mined that the most appropriate methods that could beused for the SensVest were to use accelerometry, tem-perature and heart rate However, the use of thesemethods must be considered with respect to their limi-tations Accelerometers are affected by gravity and assuch their orientation must be taken into account or thegravity effect must be assumed to be negligible Using atemperature sensor under the armpit requires that theuser be aware that the temperature value is affected byambient conditions and so may best be used for assess-ing relative changes and not as an accurate measure ofbody temperature per se User trials with these systemsdetermined that for the measurement of the heart rateusing a heart rate monitor, though the most expensiveoption, was the only method reliable enough to meat thedictates of the project

The second set of factors relate to the mounting oftechnology on the person Of particular concern werethe potential problems associated with impairment ofmovement and risks associated with having electronicsFig 10 The latest SensVest being worn

Fig 11 Redesign of the SensVest electronics

on the person Through refinement and development of

the technology, and by using assessments with such tools

as the comfort rating scales [19, 20], we believe that we

have dealt with these problems satisfactorily

Future work will involve further trials with the

Sens-Vest and assessments of the value of such technology from

a pedagogical perspective This assessment will be carried

out by partners on the Lab of Tomorrow project

Acknowledgements The work reported in this paper was supported

by a EU Grant reference IST-2000-25076 ‘‘Lab of Tomorrow’’.

The authors would like to acknowledge the work and help of S.

Baber who constructed the first SensVest and C M Knight who

developed the final version of the vest for the SensVest.

References

1 Bass L (1995) Is there a wearable computer in your future? In:

Proceedings of the IFIP TC/WG2.7 Working Conference on

Engineering for Human Computer Interaction, Yellowstone

Park, USA, August 1995, pp 3–14

2 Mann S (1998) Wearable computing as means for personal

empowerment In: Proceedings of the International Conference

on Wearable Computing (ICWC-98), Fairfax, VA 12–13 May

1998

3 Bristow HW, Baber C, Cross J, Wooley S (2002) Evaluating contextual information for wearable computing In: The Sixth International Symposium of Wearable Computers, Seattle,

WA, 7–10 October 2002, pp 179–185

4 Van Laerhoven K, Schmidt A, Gellersen HW (2002) sensor context aware clothing In: The Sixth International Symposium of Wearable Computers Seattle, WA, 7–10 Octo- ber 2002, pp 49–56

Multi-5 Hausswirth C, Bigard AX, Chevalier JM Le (1997) The med K4 telemetry system as an accurate device for oxygen uptake measurements during exercise Int J Sports Med 18:449– 453

Cos-6 Schultz H, Helle S, Heck H (1997) The validity of the telemetric system CORTEX X1 in the ventilatory and gas exchange measurement during exercise Int J Sports Med 18:454–457

7 Sirard JR, Pate RR (2001) Physical activity assessment in children and adolescents Sports Med 31(6):439–454

8 Mahon AD, Stolen KQ, Gay JA (1998) Using a facemask and sealant to measure respiratory gas exchange in children during exercise Pediatr Exerc Sci 10:347–355

9 A˚strand P, Rodahl K (1986) Textbook of work physiology physiological basis of exercise, 3rd edn McGraw-Hill, London

10 Kemper HC, Verschuur R (1987) Longitudinal study of imal aerobic power in teenagers Ann Hum Biol 14(5):435–444

max-11 Norris B, Wilson JR (1995) CHILDATA: the handbook of child measurements and capabilities—data for design safety Department of Trade and Industry, Consumer Safety Unit, London

-100 -80 -60 -40 -20 0 20 40 60 80 100 120 140 160 180 200

Time (minutes)

HR Body movement Fig 13 Data collected from a

student playing football

wearing the SensVest The

student went from rest to

playing football to rest

Fig 12 A student playing

football wearing the SensVest

18

12 Fehling PC, Smith DL, Warner SE, Dalsky GP (1999)

Com-parison of accelerometers with oxygen consumption in older

adults during exercise Med Sci Sports Exerc 31(1):171–175

13 Meijer GA, Westerterp KR, Koper H, Hoor FT (1989)

Assessment of energy expenditure by recording heart rate and

body acceleration Med Sci Sports Exerc 21(3):343–347

14 Nielsen M (1938) Cited: Astrand P, Rodahl K (1986) Textbook

of work physiology physiological basis of exercise, 3rd edn.

McGraw-Hill, London

15 Eston R, Reilly T (1996) Kinanthropometry and exercise

physiology laboratory manual: tests, procedures and data E &

FN Spon, London

16 Beevers G, Lip GYH, O’Brien E (2001) Blood pressure

mea-surement Part IV—Automated sphygmomanometry: self

blood pressure measurement BMJ 322:1167–1170

17 Gemperle F, Kasabach C, Stivoric J, Bauer M, Martin R (1998) Design for wearability In: Proceedings of the Second Inter- national Symposium on Wearable Computers, Pittsburgh, PA, 19–20 October 1998, pp 116–122

18 Knight JF (2002) The ergonomics of wearable computers: implications for musculoskeletal loading PhD thesis, School of Manufacturing and Mechanical Engineering, University of Birmingham

19 Knight JF, Baber C, Schwirtz A, Bristow HW (2002) The comfort assessment of wearable computers In: Proceedings of the Sixth Iinternational Symposium of Wearable Computers, Seattle, WA, 7–10 October 2002, pp 65–72

20 Knight JF, Baber C A tool to assess the comfort of wearable computers (in press)

O R I G I N A L A R T I C L E

Christian Becker Æ Frank Du¨rr

On location models for ubiquitous computing

Received: 20 May 2003 / Accepted: 20 March 2004 / Published online: 25 August 2004

Springer-Verlag London Limited 2004

Abstract Common queries regarding information

pro-cessing in ubiquitous computing are based on the

loca-tion of physical objects No matter whether it is the next

printer, next restaurant, or a friend is searched for, a

notion of distances between objects is required A search

for all objects in a certain geographic area requires the

possibility to define spatial ranges and spatial inclusion

of locations In this paper, we discuss general properties

of symbolic and geometric coordinates Based on that,

we present an overview of existing location models

allowing for position, range, and nearest neighbor

que-ries The location models are classified according to their

suitability with respect to the query processing and the

involved modeling effort along with other requirements

Besides an overview of existing location models and

approaches, the classification of location models with

respect to application requirements can assist developers

in their design decisions

Keywords Location model Æ Ubiquitous computing Æ

Context-awareness Æ Location-based services

1 Introduction

Location plays an important role in the domain of

location-aware and context-aware systems Especially in

the ubiquitous computing domain, location is commonly

considered to be an important source of context [1], but

not the only one [2] However, whenever applications or

users are interested in objects depending on their

loca-tion or spatial relaloca-tionship, localoca-tion models are required

in order to provide notions about distances or ranges

This paper presents an overview of possible approaches,

discusses existing work, and classifies the approachesand existing work according to their suitability to allowfor range and nearest neighbor queries

Information about locations is presented in differentformats Geometric coordinates, as used by GPS, refer

to a point or geometric figure in a multi-dimensionalspace, typically, a plane or a three-dimensional space.The topological properties of such a space allow thecalculation of distances between locations and theirinclusion in other locations

Symbolic coordinates, on the other hand, do notprovide any reasoning about their spatial properties(distance and inclusion) without any additional infor-mation Such coordinates are available via cell-IDs incellular networks, such as GSM or wireless LAN, as well

as via other positioning technologies, such as radio quency tags (RFIDs) or infrared (IR) beacons

fre-Examples for the use of location information inapplications are navigation services or location-basedinformation systems, which select services based on theirspatial proximity, e.g., the nearest printer, or notifywhen some events occur in the vicinity, e.g., a friendappears or an accident happens

In order to allow such applications based on symboliccoordinates, a notion of spatial relations such as dis-tance and inclusion is required This information has to

be modeled explicitly in a location model

In this paper, we will discuss general requirements onlocation management and derive three types of que-ries—position, nearest neighbor, and range—whichshould be supported by location models The properties

of symbolic coordinates are discussed in general Based

on these properties, different kinds of location modelsare discussed and classified along with their suitability tosupport the queries

2 System modelOur system model consists of three kinds of components(see Fig 1):

C Becker (&) Æ F Du¨rr

Institute of Parallel and Distributed Systems (IPVS),

University of Stuttgart, Universita¨tsstraße 38,

The location model Central part of our system model.

It stores representations of static and mobile real-world

objects like representations of buildings and people,

respectively It is not the focus of this paper to describe

how these objects are managed by an infrastructure, but

we concentrate on the typical properties of the different

kinds of location models Examples of such location

models are the Nexus platform [3, 4], the context

information server [5], or the guide project [6]

Applications Query the location model in order to

carry out different tasks like navigation (see next

sec-tion) They also update the location model, e.g., by

inserting new objects into the model, deleting old

ob-jects, or by altering existing objects whose state has

changed For the context of this paper, we are interested

in the different kinds of queries and tasks that are

car-ried out by these applications because they determine the

internal structure and organization of a location model

As will be shown later in this paper, the suitability of a

location model for distinct queries depends on its

internal organization This is especially of interest when

a location model is not tailored towards a single

appli-cation or domain, but should manage information for a

variety of applications and their potentially diverging

requirements

Positioning systems Update position information of

mobile objects like people or cars The output of these

systems also influences the location model, as we will see

in the next section However, the multitude and variety

of positioning systems and its discussion is beyond the

scope of this paper For the remaining part of this paper,

we will assume that a positioning system allows a mobile

object or tracking system to issue a position update with

a coordinate identifying a location to the location

model This is sufficient for the discussion of the

prop-erties of location models However, the interested reader

can find an overview of different positioning systems in[7] Fusing aspects of different positioning systems into acommon location framework are presented in [8] In thefollowing, a brief overview of the properties of coordi-nates as they are provided by current positioning sys-tems is presented

3 Basic properties of coordinates

A coordinate x is an identifier which specifies the tion of an object with respect to a given coordinate sys-tem A coordinate system is a set X of coordinates Someexamples for different kinds of coordinates and coordi-nate systems are:

posi-– Geographic coordinates in the WGS84, used by GPS,are expressed as triples containing the geographiclongitude, latitude, and the elevation above main sealevel

– The active bat system [9] is a high-resolution indoorpositioning system providing three-dimensionalcoordinates—i.e., x, y, z value—with respect to a localCartesian reference system

– The active badge system [10] provides symbolic tifiers for locations via IR Coordinates are the sym-bolic identifiers of the fixed IR sensors registering theusers‘ active badges that transmit a unique identifier.Two basic classes of coordinates can be identifiedfrom these examples: geometric and symbolic coordi-nates

iden-3.1 Geometric coordinatesGeometric coordinates define positions in the form ofcoordinate tuples relative to a reference coordinate sys-tem We further distinguish global and local geometriccoordinate systems The World Geodetic System 1984(WGS84) is a global reference system and, thus, can beused to define coordinates anywhere on this planet,whereas the Cartesian coordinates of the active batsystem are typically only valid locally, e.g., in one roomequipped with such a system

Geometric coordinates can be used to calculate thedistance between two geometrically defined positions.Through geometric operations, it can also be determined

if two areas overlap, touch each other, or one areacontains the other, i.e., topological relations like spatialcontainment can be derived from the geometry of ob-jects Hence, geometric coordinates already allow simplespatial reasoning

3.2 Symbolic coordinatesSymbolic coordinates define positions in the form ofabstract symbols, e.g., the sensor identifiers of the activebadge system, or room and street names, etc In contrast

Location Model -Positions of mobile objects

to geometric coordinates, the distance between two

symbolic coordinates is not implicitly defined Also,

topological relations like spatial containment cannot be

determined without further information about the

rela-tionship between symbolic coordinates Symbolic

loca-tion models provide this addiloca-tional informaloca-tion on

symbolic coordinates

4 Requirements for location models

In order to derive requirements on location models and

discuss their properties with respect to the

organiza-tion, we will motivate queries to location models from

the perspective of users and applications Besides

po-sition queries, which are obviously needed in

location-based applications, the necessity of nearest neighbor

and range queries is motivated This will serve as the

foundation of the later classification of location

mod-els The choice of a distinct location model will be

dependent on the queries required by applications

Therefore, we have to consider these queries and tasks

in order to assess the functional requirements for

location models

4.1 Position queries

The determination of the positions of mobile and static

objects like users, buildings, bus stops, etc is a common

building block of location-based and context-aware

systems The tasks described below cannot be carried

out without the known positions of objects Therefore,

all location models contain this information, but they

differ in the way it is represented

The definition of a position requires some form of

coordinates Based on an object’s position, actions can

be carried out, such as teleporting the user’s interface [9],

controlling the input and output of applications to

arbitrary spaces in the physical environment via

pro-jection techniques [11], or in industrial settings, such as a

smart factory [12], the positions of resources and tools

can be monitored in a production planning system Such

systems require a common interpretation of the

coor-dinates in a specific global coordinate system Within

moving objects, such as trains, local reference systems

can help to address objects, such as travelers with respect

to their compartment in the train and not their absolute

position to the ground

This shows that a general location model has to

support different coordinate reference systems, global and

localones

Beside well-known geometric coordinates, some

positioning systems provide symbolic coordinates, e.g,

the cell ID in a cell-phone network or identifiers of IR

beacons, and often, these symbolic coordinates can be

interpreted more intuitively by users than geometric

coordinates Later, we will show how simple symbolic

location models can be set up allowing for spatial

rea-soning with low modeling effort Therefore, this kind ofcoordinate has to be supported as well

4.2 Nearest neighbor queries

A nearest neighbor query is the search for the n objectsclosest to a certain position For instance, a user cansearch for the nearest restaurant with respect to hiscurrent position, or the next printer Beside known ob-ject positions, the definition of a distance function on thecoordinates is required for this type of query For geo-metric coordinates, the direct physical distance betweentwo positions can be calculated using well-known for-mulas like Pythagoras in Cartesian systems If onlysymbolic coordinates are modeled, then the model mustcontain explicit definitions of distances between thesecoordinates, e.g., to define the distance between roomnumber X and the printers in the rooms number Y and

Z, since symbolic coordinates do not contain a naturalembedment into a metric space

There are other notions of distance that are oftenmore relevant than the direct physical distance For in-stance, for a pedestrian, it might be impossible to cross ahighway Therefore, a restaurant across the highwaywith a direct physical distance of 100 m might be fartheraway than a restaurant with 200 m direct physical dis-tance not located across this highway In these cases,additional model information like the road network auser uses to get from location A to B has to be taken intoaccount For such more complex nearest neighbor que-ries, this leads to similar requirements as for naviga-tional tasks described in the next subsection, because

‘‘paths’’ between locations have to be found and their

‘‘lengths’’ have to be compared

To sum up, a notion of ‘‘distance’’ is required inmany context-aware or location-based systems An ex-plicit location model is required for symbolic coordi-nates as they do not provide implicit distance functions.Systems based on geometric coordinates can benefitfrom such a model as well, as spatial restrictions can bemodeled, e.g., road networks

4.3 NavigationNavigation systems have become standard equipment inmany modern cars Such systems require a locationmodel to find paths between locations Possible pathsare defined by the transportation network (roads, train

or bus routes, etc.) and consist of several interconnectedlocations This means that it does not suffice to know thegeometry, e.g., of roads, but it is also important to knowhow to get from one location to neighboring locations,e.g., from one road segment to another road segment at

a junction, and finally, to the destination Therefore, thetopological relation ‘‘connected to’’ has to be modeled,which describes these interconnections between neigh-boring locations (Fig 2)

22

There are different kinds of navigational tasks, e.g.,

finding the shortest path or the fastest path Finding, for

instance, a suitable path for a person in a wheelchair

requires additional information about locations, e.g.,

staircases or elevators Therefore, different attributes

need to be modeled to implement these variants, e.g., the

distance that has to be traveled to get from one location

to another location, the maximum allowed speed on a

road segment, the presence of stairs, which cannot be

used with a wheelchair, etc Even highly dynamic

information like the current traffic situation on a road

can be part of the model In general, this means

mod-eling some kind of weight on path segments The

‘‘length’’ of a path is then calculated by summing up the

weights of each path segment

4.4 Range queries

A range query returns all objects within a certain

geo-graphic area It can be used, for instance, to query the

occupancy of a room as well as to check whether an

evacuation plan is processed correctly, i.e., if a room is

empty before the fire doors are closed and sealed Also,

simple algorithms for new types of communication can

be implemented on the basis of range queries, e.g.,

geocast [13], i.e., the sending of messages to receivers in a

certain geographic area First, a range query can be used

to determine all receivers in the target area of the

mes-sage Second, the message is sent to these receivers, e.g.,

using multiple unicast messages

First of all, object positions have to be known to

answer a range query Additionally, the topological

relation ‘‘contains’’ has to be modeled, i.e., it has to be

defined whether a coordinate lies within a spatial area

For geometric coordinates, this information can be

de-rived from the known geometry But for symbolic

coordinates, this relation has to be defined explicitly

For instance, a model can define that the room 2.062 is

on (‘‘within’’) the second floor that, in turn, is part of

(‘‘within’’) a certain building, etc Thus, querying for a

larger area automatically includes all objects from

locations that lie within that area

4.5 Visualization

Drawing maps is one of the most obvious applications

of location models Maps can be used for many different

tasks like positioning, navigation, etc., which we have

already described in the subsections above A map helps

the user to execute these tasks manually or it can be used

to display the results of these tasks if they were carriedout automatically All model information introducedabove can be visualized, but usually, a map is drawn,which requires a more or less detailed geometric repre-sentationof these objects, depending on the desired level

of detail (see below)

4.6 RequirementsFrom the use cases presented above, the followingrequirements for location models can be derived Notethat not all of these requirements have to be fulfilled atthe same time However, being aware of the applicationrequirements is crucial in order to choose the appro-priate location model organization

Based on position, nearest neighbor, and range ries, it can be concluded that a location model shouldprovide:

que-1 Object positions: Positions of objects have to bemodeled in the form of coordinates Supportedcoordinates and reference systems are:

• – Geometric and symbolic coordinates

• – Multiple local and global coordinate referencesystems

2 Distance function: Distances between spatial objectshave to be modeled This can also be the ‘‘size’’ of alocation, e.g., the length of a road segment, whichrepresents the distance one has to travel whencrossing this location in order to reach anotherlocation

3 Topological relations: The following topologicalrelations between spatial objects have to be modeled:

• – Spatial containment in order to allow range queries

• – Spatially connected to for navigation servicesFurthermore, the position of objects alone is notsufficient for some applications that also require thedirection of a moving object or the orientation of a user,e.g., in order to provide information about the building

a tourist looks at

4 Orientation: In addition to the positions of mobileobjects, the orientation in the horizontal and/or ver-tical dimensions can be supported

These requirements have to be regarded in tion with the requirement of minimal modeling effort.There are different factors that influence the modelingeffort:

length=80m length=

200m Fig 2 Road geometry (left)

and road topology (right)

– Accuracy: The model should describe the real world as

accurately as possible, i.e., the stored information

should be consistent with the real world Accuracy is

not a question of the model type, but of how the

model is created and updated, and of the dynamics of

the modeled objects: highly dynamic objects require

high update rates, e.g., highly mobile objects will have

to update their position frequently to get accurate

position information These issues are not the focus of

this paper, and, therefore, accuracy will not be

con-sidered any further

– Level of detail: The level of detail describes the

pre-cision or granularity of the model Fine-grained

models describe locations down to room level or

be-low; coarse-grained models stop at buildings or larger

A flexible model allows both ends of the scale

– Scope: The scope is the area covered by the model

Local models may only describe one single room,

whereas global models, at the other end of the scale,

describe locations all over the world

The two last items are intimately connected Highly

detailed models usually only describe small parts of the

world because they require high modeling effort;

coarse-grained models may have a larger scope [14]

Also, the architecture used to manage the model plays

an important role for the level of detail and scope A

federation of highly detailed partial models with limited

scope can be used to extend the scope of the (federated)

model and make highly detailed global models feasible

[4] In this paper, we do not consider how location

models are implemented, but we concentrate on the

general properties of the different kinds of location

models

The following discussion first addresses location

models for geometric and symbolic coordinates Then,

the integration of geometric coordinates into symbolic

location models leading to hybrid location models is

discussed Based on this discussion, a classification of

the general approaches is presented and existing work is

classified

5 Geometric location models

Geometric models describe locations by geometric

fig-ures If not only global coordinate systems are to be used

but also local ones, the position and orientation of local

systems with respect to other local systems or the globalsystem has to be defined in order to translate coordi-nates of one system to other systems

On the basis of geometric coordinates, the cal relation ‘‘contained in’’ can be derived In contrast tothe containment relation, the ‘‘connected to’’ relationmodeling e.g, doors connecting rooms, cannot be de-rived from location geometries This relation has to bemodeled explicitly If this information is modeled, itcan be used to improve the notion of distances, e.g.,

topologi-by incorporating the distance a user has to travel

in contrast to the direct distance reflected by theunderlying geometry However, it is also reasonable for

a geometric location model to store the spatial tainment relation explicitly since geometric operationsare costly

con-6 Symbolic location models

In this section, we describe different types of symboliclocation models and discuss their suitability for thedifferent types of queries described in the requirementssection of this paper Set-based, hierarchical, and graph-based models are presented

6.1 Set-based model

A set L of symbolic coordinates forms the basis for theset-based approach Locations comprising several sym-bolic coordinates are defined by subsets of the set L As

a simple example, consider a building with several floors.The set L consists of all the room numbers of thisbuilding The second floor as shown in Fig 3 can bemodeled by the set Lfloor2={2.002, 2.003, ., 2.067}.Further arbitrary locations may be defined, e.g., thelocations A={2.002, 2.003} and B={2.003, 2.005} inFig 3

This model can be used to determine overlappinglocations and, as a special case of overlapping locations,the containment relation by calculating the intersection

of two sets L1 and L2 If L1\L2„ [, then L1 and L2overlap If L1\L2=L1, then L2 contains L1 Thus, thismodel can be used for range queries where the range isdefined by one set R of symbolic coordinates, and allsubsets of R define locations within R

Fig 3 Set-based location model

24

This model can also be used to express a simple

qualitative notion of distance between symbolic

coor-dinates by modeling sets of ‘‘neighboring’’ symbolic

coordinates, which we call neighborhoods (by Lcon, we

denote the set of neighborhoods) For instance, the sets

Aand B in Fig 3, as well as the set Lfloor2defined above,

are such neighborhoods in Lcon Distances between the

symbolic coordinates x, y and x, z are compared as

follows:

dðx;yÞ\dðx;zÞ

, 9L18L22 Lconðx 2 L1^ y 2 L1^ x 2 L2^ z 2 L2

! L1
L2

That means, the two smallest neighborhoods containing

x, y and x, z, respectively, define the distance from x to y

and x to z Consider, for instance, the three symbolic

coordinates 2.002, 2.003, and 2.006 d(2.002,

2.003)<d(2.002, 2.006) because A (the smallest

neigh-borhood that contains 2.002 and 2.003) is a proper

subset of Lfloor2 (the smallest neighborhood that

con-tains 2.002 and 2.006 in our example) To achieve a fine

distance granularity, neighborhoods can be defined for

each pair of directly connected locations, e.g., rooms

which are connected by a door For instance, the

loca-tions A and B introduced above are such localoca-tions

Larger neighborhoods are defined recursively by joining

smaller neighborhoods which have non-empty

intersec-tions, e.g., the neighborhood C=A[B By modeling

pairs of connected locations, possible paths can also be

derived A negative effect of this approach is the huge

number of resulting sets and the involved modeling

ef-fort

Beside this qualitative notion of distance, this

ap-proach does not permit to define a quantitative notion of

distance, e.g., to make statements like ‘‘the distance

between a and b is as long as the distance between c and

d.’’ Therefore, the support for queries related to spatial

distances (e.g., nearest neighbor queries and navigation)

is limited

In contrast to set-based location models, which do

not contain explicit relations between locations, the

following two models, i.e., hierarchical and graph-based,model the relations between locations

6.2 Hierarchical modelsHierarchical models consist of a set of locations L Thelocations are ordered according to the spatial contain-ment relation, i.e., a location l1 is an ancestor of alocation l2(l1>l2), if l2is spatially contained in l1 If thelocations do not overlap each other, this leads to a tree-based model [15] If overlapping locations are to bemodeled, the more general lattice-based model is appli-cable where intersections of locations are modeled byseparate locations with more than one parent location[13, 16] Figure 4 shows an example of such a lattice-based model The set of locations L consists of thebuilding B, the floors F1, ., Fm, two wings W1and W2,and several rooms R1, ., Rn The locations FiWjdenoteintersections of the floor Fiand the wing Wj Figure 4balso shows the relationship of the hierarchical models tothe set-based approach Locations in the hierarchy canalso be interpreted as sets of symbolic coordinates.Overlapping locations are defined by the intersection ofsets Therefore, hierarchical models can be seen as aspecial case of set-based models

Because the hierarchical models are based on thecontainment relation, they support range queries natu-rally A range is defined by a location in the hierarchyand the descendants of this location denote locationswithin this range

A simple notion of distance comparable to the onediscussed in the previous subsection can also be applied

in the same wing but not on the same floors

R ,R ,3 4

W = {R ,R }

1

W = {R ,R ,R }

2

R = {R }

5

F W = {R , }

Fig 4 Hierarchical

lattice-based location model

(F1W2=sup({R1, R2})<sup({R2, R5})=W2) In some

situations, this interpretation of distance may be

coun-ter-intuitive If, for instance, a short connection exists

between R2and R5, e.g., stairs, then R2could be closer

to R5than to some room located on the same floor and

wing as R2 Hierarchical models provide no means to

model interconnections between locations, and,

there-fore, this situation can not be handled adequately As

for the set-based approach, this notion of distance is also

only qualitative

6.3 Graph-based model

In the graph-based approach, symbolic coordinates

de-fine the vertices V of a graph G=(V, E) An edge is

added between two vertices if a direct connection

be-tween corresponding locations exists Edges or vertices

can be weighted to model distances between locations

Figure 5 shows an example of a graph-based model for

the already presented second floor of a building In this

example, the distance between two coordinates is just the

number of hops, but with additional information, a

higher accuracy could be achieved Drosdol [17] gives a

deeper discussion of this aspect of graph-based models

From the construction of the graph, it is already clear

that a graph-based model supports the definition of the

topological relation ‘‘connected to’’ as well as the explicit

definition of distances between symbolic coordinates It

is, therefore, well-suited for nearest neighbor queries as

well as navigation For the latter, the edges or nodes can

be further attributed to model e.g., speed limits, vehicle

restrictions, etc [18]

For range queries, first the range itself has to be

de-fined, i.e., an area has to be described within which we

want to search for included objects The only locations

which are explicitly defined in the graph-based model are

the nodes of the graph, e.g., the rooms shown in the

example above This is surely a very limited set of

ran-ges Because the graph-based model allows to define a

distance between symbolic coordinates, this distance can

be used to define ranges That means an object is in the

area if the distance between its position and a reference

location is at most the radius of the area In Fig 5 for

instance, the unfilled symbol locations are within the

range defined by a reference location marked by the

dark symbol and the radius 2, thus, all objects at these

locations are within this range What we are missing is

the possibility to explicitly define bigger locations

com-prising several smaller locations, e.g., a whole floor,

building, or even parts of a city In the next section, wewill show how this limitation can be overcome bycombining the different types of symbolic locationmodels

6.4 Combination of graph-based and set-basedsymbolic models

Our discussion of the different location models hasshown that, for symbolic coordinates, the graph-basedapproach supports queries based on distance and thedefinition of connected locations well, whereas the set-based approach can be used for range queries withexplicitly defined locations like floors, building, etc.,representing ranges Therefore, a combination of graph-based and set-based symbolic locations models can beused to combine the benefits of both types of models.The set-based part of the combined symbolic locationmodel consists of a set of symbolic coordinates Loca-tions are subsets of this set of locations, e.g., repre-senting rooms, floors, buildings, etc This part of themodel is used for range queries as described in the sec-tion about set-based models

In the graph-based part of the combined model,locations are connected by edges if a connection betweenthese locations exists in the real world For instance, tworooms will be connected in the graph if there is a doorbetween them; two floors will be connected if stairs leadfrom one floor to the other, etc As mentioned in theprevious section, edges can also be weighted to modeldifferent distances Figure 6 shows an example of theresulting combined model

Besides the already mentioned support for differenttopological relations and distances and the range andnearest neighbor queries based on this information, thismodel shows another interesting feature It allows togenerate views with different levels of detail Figure 7

Fig 5 Graph-based model

floor A.2

floor A.1 floor B.1

floor B.2

floor B.3

room A.2.1

building A building B

Fig 6 Combined symbolic location model 26

shows three examples The first example shows the

rooms on one particular floor and their connections

This view will be used if a very fine granularity is

re-quired, e.g., if we are searching for the next printer

Figure 7b shows only the floors of building A Floor A.1

and A.2 are connected because elements of floor A.1 and

A.2 have a connection—e.g., two hallways connected by

an elevator Finally, Fig 7c depicts only buildings and

the paths between them The latter could be used in a

scenario where only coarse-grained location information

suffices, and so, it allows to generate small models that

cover large areas, e.g., a whole city district

6.5 Summary

We now summarize the properties of the different types

of symbolic location models presented in this section

(Table 1)

We see that the graph-based approach, as well as the

hierarchical models, support the containment relation

well, making them suitable for range queries The

graph-based approach is well-suited for all kinds of queries

where distance plays an important role, e.g., nearest

neighbor queries and navigation The combined symbolic

location model combines the benefits of all other symbolic

model types at the cost of higher modeling effort

Still, the accuracy of the combined model can be

further improved by adding geometric information The

next section presents different hybrid models, which

integrate symbolic and geometric information

7 Hybrid location models

The combined symbolic location model presented in the

previous section shows how the benefits of set-based and

graph-based models can be integrated into a common

symbolic model There are two major arguments for

additionally adding geometric information to such a

symbolic model First, geometric information can beused to achieve higher accuracy and precision for allkinds of distance-related queries Secondly, arbitrarygeometric figures can be used, for instance, to defineranges for nearest neighbor queries, whereas symboli-cally defined locations are always restricted to a givenstructure

We distinguish between two types of hybrid locationmodels The first approach, which we call the subspaceapproach, stores geometric information for every mod-eled location The second approach only stores geo-metric information for some locations, leading to partialsubspaces

7.1 SubspacesThe basis for this hybrid location model is a symbolicmodel like the combined symbolic model presented inthe previous section Additionally, the geometric extent

of locations is stored in the location model The metric extent can be either defined using a global refer-ence system like WGS84 or local reference systemswhere coordinates are only valid within a certain scope,e.g., in one building or room Subspaces are formed byembedding coordinate systems into other coordinatesystems by defining the position and orientation ofembedded systems (a detailed description of thisembedding of subspaces can be found in [15]) With thisinformation, coordinates can be translated from onesystem to other systems, and, thus, coordinates of dif-ferent systems can be compared

geo-Figure 8 shows a simple example of a hybrid locationmodel using subspaces The symbolic part of this model

is based on a graph defining the interconnections tween the rooms on a certain floor The extent of everyroom is also modeled geometrically using the coordinatesystem SB of the building B Within room 2.1, a localcoordinate system S2.1is defined that is embedded intothe system of building B The system of building B in

be-Table 1 Properties of symbolic location models

Symbolic model type Supported coordinate

types

Modeling effort a Distance support ‘‘Connected to’’

relation support

‘‘Containment’’ relation support

Combined (set-based

and graph-based)

a Modeling effort is always dependent on the granularity and scope of location information, as stated in the requirements section Therefore, we give a range here.

floor A.1

building A building B

Fig 7 Levels of detail

turn may be embedded into a global coordinate system.

The known geometry can be used to define precise

dis-tances between rooms

7.2 Partial subspaces

In contrast to the subspaces approach, the partial

sub-spaces approach does not assume that the geometric

extent for every location is modeled, but only for some

locations Figure 9 shows an example where a geometric

location model exists for the outdoor domain, but within

buildings, symbolic models are used By linking

geo-metric information to symbolic locations, the symbolic

building models can be embedded into the global

geo-metric model The benefit of this integration becomes

clear when we consider a range query with a

geometri-cally defined range, e.g., a polygon drawn on a city plan

Users within a building may only know a symbolic

po-sition like room 2.1 in building B Through the known

geometric extent of the building, the user’s position can

be approximated geometrically with the geometry of the

whole building This approximated geometric position

can be compared to the geometrically defined range of

the query, and, thus, the query can be answered Of

course, the approximation has its limitations For

in-stance, using geometric areas within a building that is

only modeled symbolically makes no sense But it

mains an interesting alternative that can be used to

re-duce modeling effort

8 Discussion

A summary of the properties of the presented location

models is shown in Table 2 In contrast to the purely

symbolic models presented in the previous section, all

hybrid models support geometric coordinates as well as

symbolic coordinates By using geometric information,

distances can be modeled more accurately and precisely

The spatial containment relationship does not need to

be modeled manually if the geometry of locations isknown This information can be derived by using geo-metric operations Still, it makes sense to have a modelthat stores the containment relation explicitly to allowfor efficient queries

Geometric information can also be used to find outwhether two locations lie next to each other, but con-nections like doors or junctions can not be derived fromgeometric information and, therefore, have to be mod-eled explicitly as for the symbolic approaches

Compared to the subspaces approach, the modelingeffort can be reduced by using a partial subspace modelwhere not every location is modeled geometrically Still,

a geometry can be associated with location by usingapproximation

9 Summary and classification of existing approaches

This section briefly summarizes the properties of thedifferent location models presented so far Table 3classifies the location models with respect to the sup-ported coordinate types (sym=symbolic, geom=geo-metric), the supported queries (P=position, R=range,N=nearest neighbor), and the modeling effort Exam-ples for projects using the location model class are listed

as well, and are discussed in the following subsections.Since the discussion so far has shown that there is nolocation model serving all requirements at the same timewith similar modeling effort, designers of locationmanagement systems have to chose an appropriatestructure for the underlying location model Especially,the trade-off between supported queries and the involvedcomplexity of the location models has to be taken intoconsideration

Table 2 Properties of hybrid location models

Model type Supported coordinate

Subspaces Symbolic, geometric High to very high Very good Yes (if modeled explicitly) Yes

Partial subspaces Symbolic, geometric High Good to very good Yes (if modeled explicitly) Yes

floor 2 floor 1 building B

S B

S 2.1

Fig 8 Hybrid location model with subspaces

28

9.1 Set-based location models

Modeling symbolic locations as identifiers and mapping

object IDs to location IDs in location services has been

widely adopted The Guide project identifies the

loca-tions of interest to tourists by the WaveLAN access

point ID [6] The Active Badge system [10] stores the

identifier of a user’s badge with the symbolic location

where the badge has been observed Without defining

further locations as ranges, only position queries can be

processed with minimum modeling effort However, an

extension of such systems allowing for overlapping sets

of locations and, thus, range queries has been used in the

Open Distributed Office project [19] The modeling effort

increases with the number of locations introduced to the

system QoSDREAM [20] relies on a mapping of

loca-tion identifiers and object IDs By applying observers to

sets of locations, applications can be notified when a

mobile object has been observed in a set of locations

This provides means for range queries but causes

con-siderable effort, since the overlay of observers modeling

spatial inclusion has to be set up based on the basic sets

9.2 Graph-based location models

This class of location models naturally provides means

to model distance, making them suitable for all

navi-gation-oriented tasks Applications can be found in the

domain of smart environments [21, 22] Spatially scoped

areas are modeled by the location users populate, e.g.,

floors and rooms, and a connection model defines

con-nectivity and distance Navigation services

incorporat-ing the positions of individual objects can be

implemented that way There is no direct notion of

ranges Either a combined approach is taken modeling

ranges as an overlay structure—in the simplest case,

ranges are specified as sets of locations themselves—or

ranges can be defined based on their extension, i.e., by areference location and the distance to this location

9.3 Hierarchical location models

In contrast to graph-based models, which reflect tance well but require additional overhead to expressranges, hierarchical models are designed to reflect theinclusion of locations This allows to structure locationsinto a hierarchy It is noteworthy that, although ap-proaches such as EasyLiving [23] or MOOsburg [24]only model the spatial inclusion between locations, otherkinds of hierarchical relations can be modeled, such as

dis-an orgdis-anizational structure A compdis-any may structureits location into development, marketing, research, andproduction A distributed systems development tea-m—and its offices—may be organized to be nearer to thedistributed systems research team in a hierarchy than thetheoretical computer science research group in the officesnearby

Ranges and their relations—spatially, or with respect

to other criteria such as organizational relations—arewell reflected in a hierarchy Distances do not come with

a direct concept in such location models One way to use

a hierarchy to compare distances between positions is toconsider the smallest locations in the hierarchy thatcontain these positions That means positions grouped

by smaller locations are considered to be closer to eachother than positions grouped by larger locations, e.g.,two rooms on the same floor can be said to be closerthan two rooms where the smallest common range is thebuilding

9.4 Combined symbolic location models

An obvious approach combining the benefits of based and hierarchical location models are combined

graph-Table 3 Properties of location models and overview of existing implementations

Supported coordinates

Supported queries Modeling effort Projects

Open Distributed Office [21]

MavHome [23]

Semantic Spaces [25]

Subspaces (hybrid model) Yes Yes Good Good Good b Very high Jiang [15]

Leonhardt [26]

Partial subspaces (hybrid model) Yes Yes Good Good Good High Nexus [13, 27]

Semantic Location Model [28]

a ‘‘Range’’ defined by distance to reference location

b If the ‘‘connected to’’ relation is modeled

symbolic location models, such as those used in the

Active Map [1] Either a common data structure is

ap-plied that allows to reflect the inclusion relation as well

as the ‘‘connected to’’ relation between locations, such as

in [25], or two different location models are maintained

where one reflects the distances and the other the ranges

Clearly, the expressiveness of such models combines the

benefits of both models, but with a trade-off with respect

to the modeling effort, which basically consists of the

effort of creating two location models This effort is only

justified when applications require range and nearest

neighbor queries This will likely be the case when a

location model is set up to serve a number of

applica-tions, e.g., by providing an application-spanning context

model

9.5 Hybrid location models

Hybrid location models provide information about

locations based on symbolic and geometric coordinates,

which are used to define the spatial extent of locations

Basically, all of the symbolic models above can be

extended to a hybrid model by annotating location with

their spatial extent A graph-based model may use this

information to calculate the weight of connections or

rooms in order to provide more accurate distances Since

the effort of obtaining spatial extensions of locations is

rather high, some projects consider a combined model as

basis, e.g., [25] and [26] The effort of annotating all

locations in a location model with geometric

informa-tion can be used to map the symbolic coordinates into a

global, geometric reference systems realizing a subspaces

approach [15] If this is not necessary, a partial subspace

approach can be taken Such approaches can be realized

either top-down or bottom-up In [13], a top-down

ap-proach is taken that allows approximating the spatial

extents of children in a location hierarchy by the extents

of their father nodes A bottom-up approach would

annotate the leaves in a location model and approximate

the spatial extents of a father node by the extents of its

child nodes The top-down approach allows the

inte-gration of an area that is modeled by a hierarchy of

symbolic locations into a geometric model The root of

the integrated hierarchy is exact with respect to the

annotated spatial extent, whereas the approximation

leads to some errors in the spatial extents along the

hierarchy In contrast to that, the bottom-up approach

provides the highest accuracy at the leafs The modeling

effort is great for this approach if the hierarchy has

many leaves Clearly, it is application dependent which

approach should be taken under given requirements

10 Conclusion

Modeling locations is crucial for most location-based or

context-aware applications Location models provide

means for spatial reasoning based on coordinates, e.g.,

the determination whether a coordinate is within a givenrange or which coordinates are nearby Although geo-metric coordinates already provide an implicit notion ofdistance and ranges, location models allow to model theconstraints of the physical world, e.g., road networks orfloor plans For symbolic coordinates like room or floornumbers, a location model with explicitly modeledrelations between locations is essential to support que-ries beyond simple position queries

The requirements of applications can be manifold.Since the structure of a location model determines whichkinds of spatial reasoning can be processed, a number oflocation models may be appropriate Besides the rele-vant queries a location model has to support, especiallythe modeling effort has to be taken into considerationwhen choosing a location model for an application or aplatform serving a number of applications A hybridmodel managing geometric and symbolic coordinatessupports all kinds of location-based queries very well,but is, at the same time, the most complex type oflocation model Location models managing only sym-bolic locations can be set up more easily If, besidesobject positions, distance is the only relevant informa-tion, a graph-based symbolic model can be used,whereas range queries are supported very well by hier-archical symbolic models If higher accuracy is requiredonly partially within limited areas, a partial subspacesmodel, which augments a symbolic model partially withgeometric information, might be the right choice.The discussion of location models in this paper showsthat there is no location model which satisfies all iden-tified requirements at the same time with a low modelingeffort Designers of context-aware applications andsystems, thus, have to choose location models carefullywith respect to the required spatial reasoning and theinvolved modeling effort

Acknowledgements The second author gratefully acknowledges the financial support by the Deutsche Forschungsgemeinschaft (DFG) within the Center of Excellence 627 ‘‘Spatial World Models for Mobile Context-Aware Systems.’’

4 Nicklas D, Großmann M, Schwarz T, Volz S (2001) A based, open architecture for mobile, spatially aware applica- tions In: Proceedings of the 7th international symposium on spatial and temporal databases (SSTD 2001), Redondo Beach, California, July 2001

model-5 Judd G, Steenkiste P (2003) Providing contextual information

to pervasive computing applications In: Proceedings of the 1st IEEE international conference on pervasive computing and 30

communications (PerCom 2003), Fort Worth, Texas, March

2003, pp 133–142

6 Cheverst K, Davies N, Mitchell K, Friday A (2000) Experiences

of developing and deploying a context-aware tourist guide: the

GUIDE project In: Proceedings of the 6th annual international

conference on mobile computing and networking (MobiCom

2000), Boston, Massachusetts, August 2000, pp 20–31

7 Hightower J, Borriello G (2001) Location systems for

ubiqui-tous computing Computer 34(8):57–66

8 Hightower J, Brumitt B, Borriello G (2002) The location stack:

a layered model for location in ubiquitous computing In:

Proceedings of the 4th IEEE workshop on mobile computing

systems and applications (WMCSA 2002), Callicoon, New

York, June 2002, pp 22–28

9 Ward A, Jones A, Hopper A (1997) A new location technique

for the active office IEEE Personal Commun 4(5):42–47

10 Want R, Hopper A, Gibbons J (1992) The active badge

loca-tion system ACM Trans Inf Syst 10:91–102

11 Pingali G, Pinhanez C, Levas A, Kjeldsen R, Podlaseck M,

Chen H, Sukaviriya N (2003) Steerable interfaces for pervasive

computing spaces In: Proceedings of 1st IEEE international

conference on pervasive computing and communications

(Per-Com 2003), Fort Worth, Texas, March 2003, pp 315–322

12 Bauer M, Jendoubi L, Rothermel K, Westka¨mper E (2003)

Grundlagen ubiquita¨rer Systeme und deren Anwendung in der

’Smart Factory’’ Industrie Management—Zeitschrift fu¨r

industrielle Gescha¨ftsprozesse 19(6)

13 Du¨rr F, Rothermel K (2003) On a location model for

fine-grained geocast In: Proceedings of the 5th international

con-ference on ubiquitous computing (UbiComp 2003), Seattle,

Washington, October 2003, pp 18–35

14 Rothermel K, Dudkowski D, Du¨rr F, Bauer M, Becker C

(2003) Ubiquitous computing—more than computing anytime

anyplace? In: Proceedings of the 49th photogrammetric week,

Stuttgart, Germany, 1–5 September 2003

15 Jiang C, Steenkiste P (2002) A hybrid location model with a

computable location identifier for ubiquitous computing In:

Proceedings of the 4th international conference on ubiquitous

computing (UbiComp 2002), Go¨teborg, Sweden, September/

October 2002, pp 246–263

16 Kainz W, Egenhofer MJ, Greasley I (1993) Modeling spatial

relations and operations with partially ordered sets Int J Geogr

Inf Syst 7(3):215–229

17 Drosdol T (2003) Unterstu¨tzung symbolischer Koordinaten im

Lokationsmanagement Diploma thesis, University of

Stutt-gart, Germany

18 Volz S, Großmann M, Ho¨nle N, Nicklas D, Schwarz T (2002) Integration mehrfach repra¨sentierter Straßenverkehrsdaten fu¨r eine fo¨derierte Navigation Informationstechnik und Techni- sche Informatik 44(5):260–267

19 Marmasse N, Schmandt C (2001) Location modeling In: Proceedings of the workshop on location modeling for ubiq- uitous computing, Atlanta, Georgia, September 2001

20 Naguib H, Coulouris G (2001) Location information agement In: Proceedings of the 3rd international conference on ubiquitous computing (UbiComp 2001), Atlanta, Georgia, September/October 2001

man-21 Rizzo M, Linington P, Utting I (1994) Integration of location services in the open distributed office Technical report 14/94, University of Kent, Canterbury, UK

22 O’Connel T, Jensen P, Dey A, Abowd G (2001) Location in the aware home In: Proceedings of the workshop on location modeling for ubiquitous computing, Atlanta, Georgia, Sep- tember 2001

23 Roy A, Bhaumik SKD, Bhattacharya A, Basu K, Cook DJ, Das SK (2003) Location aware resource management in smart homes In: Proceeding of the 1st IEEE international conference

on pervasive computing and communications (PerCom 2003), Fort Worth, Texas, March 2003

24 Ganoe CH, Schafer WA., Farooq U, Carroll JM (2001) An analysis of location models for MOOsburg In: Proceedings of the workshop on location modeling for ubiquitous computing, Atlanta, Georgia, September 2001

25 Brumitt B, Shafer S (2001) Topological world modeling using semantic spaces In: Proceedings of the workshop on location modeling for ubiquitous computing, Atlanta, Georgia, Sep- tember 2001

26 Leonhardt U (1998) Supporting location-awareness in open distributed systems PhD thesis, Department of Computing, Imperial College London

27 Bauer M, Becker C, Rothermel K (2001) Location models from the perspective of context-aware applications and mobile ad hoc networks In: Proceedings of the workshop on location modeling for ubiquitous computing, Atlanta, Georgia, Sep- tember 2001 pp 35–40

28 Hu H, Lee DL (2004) Semantic location modeling for location navigation in mobile environments In: Proceeding of the 5th IEEE international conference on mobile data management (MDM 2004), Berkeley, California, January 2004

O R I G I N A L A R T I C L E

Mark Howell Æ Steve Love Æ Mark Turner

Spatial metaphors for a speech-based mobile city guide service

Received: 1 October 2003 / Accepted: 15 March 2004 / Published online: 8 May 2004

Springer-Verlag London Limited 2004

Abstract Speech-based automated mobile phone services

allow people to access information whilst on the move,

but are difficult to use due to the arbitrary assignment of

numbers to menu options For this study, it was

hy-pothesised that the use of spatial interface metaphors

could lead to higher levels of usability for a mobile city

guide service by capitalising on humans’ well developed

spatial ability One non-metaphor, numbered menu

service, and three different spatial metaphor-based

ser-vices were implemented The metaphors used were: a

travel system, an office filing system and a shopping

metaphor Measures of participant performance with

each service and their corresponding subjective

evalua-tions were recorded for each trial The results indicated

that, for first-time users, the non-metaphor service was

the most usable, but after three trials, the office filing

system metaphor service was the most usable

Naviga-tional cues provided by spatial interface metaphors may

improve user attitudes and interactions with automated

phone services

Keywords Mobile phone Æ Interface metaphor Æ

Automated phone service Æ Speech input

loca-Automated mobile phone services are usually tured hierarchically, and consist of lists of spoken menuoptions arbitrarily assigned to numbers The user selects

struc-M Howell (&) Æ S Love

Department of Information Systems and Computing,

Brunel University, Uxbridge, Middlesex, UB8 3PH, UK

Department of Psychology, University of Portsmouth,

King Henry Building, King Henry I Street,

Portsmouth, PO1 2DY, UK