Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility

Study

Mehmet Cihan Sakman

1 a

, Panagiotis Gkikopoulos

1 b

, Francesco Martella

2 c

Massimo Villari

2 d

and Josef Spillner

1 e

School of Engineering, Zurich University of Applied Sciences, Technikumstrasse 9, Winterthur, Switzerland

MIFT Department, University of Messina, Messina, Italy

Keywords:

Smart Shopping, Product Search, Digital Companion.

Abstract:

Consumers in brick-and-mortar stores are increasingly expecting a personalised shopping experience and dig-

ital assistance in ﬁnding the right products. The retail industry has already undergone a shift towards more

ubiquitous in-store technology but smart shopping assistance for consumers is notably absent. Whether physi-

cal product search, localisation and navigation could work with contemporary technology is an open question.

In an applied research setting conducted on behalf of a retail technology provider, we have analysed the feasi-

bility of introducing a smart shopping workﬂow based on product search and indoor navigation. In this paper

we provide our ﬁndings with key contributions in workﬂow design, mobile application design, and technology

ﬁtting concerning localisation and notiﬁcation of customers.

1 INTRODUCTION

In the retail industry, physical stores are turning from

merely places to sell products towards spaces for con-

sumption experience. Technology is a key driver be-

hind personal recommendations and other contribu-

tors to that experience. The targeted use of technology

turns shopping places into networked cyber-physical

deployments with the involvement of access points,

mobile devices, smart shelves and dispensers, elec-

tronic shelf labels (ESL) and point of sales (POS)

stations, cameras and other sensors, and industry-

speciﬁc software such as Enterprise Resource Plan-

ning (ERP) for dynamic stock management, label

designers, rule engines and campaign dashboards

(Kellermayr-Scheucher et al., 2022). The more tech-

nology is installed, the more options there are for ex-

ploiting it in terms of providing a smart shopping ex-

perience; yet at the same time, store owners are cost-

conscious and prefer low-cost, low-maintenance solu-

tions. Moreover, the added value of the technologisa-

tion of stores is not always clear to owners, although

https://orcid.org/0000-0001-9541-123X

https://orcid.org/0000-0001-6436-8929

https://orcid.org/0000-0002-5022-3259

https://orcid.org/0000-0001-9457-0677

https://orcid.org/0000-0002-5312-5996

it becomes more clear when it translates into higher

customer engagement and satisfaction. Letting the

customer ﬁnd the right product with technology assis-

tance, and not missing out an opportunity of purchase,

is contributing to the added value as evidenced by re-

cent empirical studies (Gong et al., 2022; Linzbach

et al., 2019).

The focus in this paper is speciﬁcally to investi-

gate the feasibility of cost-effective augmented prod-

uct search within the stores. Consumers interested

in fully deﬁned products, brands or less deﬁned cat-

egories need assistance in expressing their interests,

getting an overview situation about the availability,

and receiving guidance through signalling and navi-

gation to ensure that the chosen products end up in the

basket. From a technology perspective, this requires

a complex workﬂow encompassing an interactive de-

vice (usually the consumer’s mobile phone), beacons

for indoor positioning, and shelf labels associated to

products. Building an unconstrained lab-level tech-

nology could address this problem but would not have

chances of being adopted on the market. Instead, we

opt for a real-tech approach, intending to design and

validate a solution that works in the constrained en-

vironments found in real stores and matches real cost

requirements.

Our key contributions in this space are: (i)

An abstract workﬂow for personalised and privacy-

Sakman, M., Gkikopoulos, P., Martella, F., Villari, M. and Spillner, J.

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study.

DOI: 10.5220/0012085100003552

In Proceedings of the 20th International Conference on Smart Business Technologies (ICSBT 2023), pages 43-53

ISBN: 978-989-758-667-5; ISSN: 2184-772X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

preserving physical product search in a shop, (ii) a

concrete mobile application to realise this search with

hybrid notiﬁcations based on plausible technologies,

and (iii) a validation in a lab environment with real-

tech equipment and processes.

The paper is structured as follows: First, we

review existing approaches towards product ﬁnding

and indoor navigation, and introduce additional pre-

liminary knowledge. Then, as ﬁrst contribution we

present a custom abstract workﬂow for personalised

shopping that takes the technological constraints into

account. Next, we explain the customer interaction

with the mobile device and its backend service, and

in this context introduce the application as second

contribution and explain the indoor navigation algo-

rithm which reproduces state-of-the-art results. Fi-

nally, we discuss the technology ﬁtting, validating

the personalised shopping with real-tech infrastruc-

ture as third contribution, outlining potential improve-

ments depending on technological evolution, before

concluding with remarks on possible adoption of the

results in industry.

2 RELATED WORK

Research on digital shopping assistance especially

around navigation to products in the physical space

has been a niche topic for a long time but has seen

progress in the recent years. For visually impaired

people, assistance is obligatory and can be addressed

with autonomous navigation based on computer vi-

sion, text recognition and text synthesis (Miralles

et al., 2022) as well as combinations of computer vi-

sion and barcode detection (Deshmukh et al., 2022)

and the combined use of accelerometer, gyroscope

and magnetometer (Perera et al., 2020). The effec-

tiveness of search increases with the data and suitable

data structure modelling and visualisation, and there-

fore research has also been conducted on taxonomies

and ontologies such as OntoNavShop (Ruijgrok et al.,

2018).

Researchers have also investigated the use of con-

nected devices in shops for other purposes beyond im-

pairment such as smart shopping carts that follows

the consumer autonomously (Heyns et al., 2021),

technology-enabled personalisation (TEP) (Riegger

et al., 2021), and the effects of using mobile de-

vices with augmented reality on consumer behaviour

(Chen et al., 2022). A previous work studied product-

awareness shopping through RFID (Chen et al., 2014)

but required the consumer to be already close to the

product to retrieve its information. Many of the stud-

ies are conducted with an economics background and

do not dive deep into technical matters of feasibil-

ity and realisation. In contrast, our work combines

physical product ﬁnding and hybrid notiﬁcation about

products of interest, an aspect lacking from many of

the proposed approaches, and establishes a techno-

logical grounding. The hybrid notiﬁcations exploit

the growing deployment of electronic shelf labels, a

technology already investigated from a psychological

perspective in terms of revenue effects (Boden et al.,

2020) and customer acceptance (Garaus et al., 2016)

but not yet in the context of navigation.

From an innovation perspective beyond the re-

search, indoor navigation and physical product search

is increasingly commercialised by startups such as

MobiDev and Hyper, and attracting the interest of

large mobile platform operators and advertisement

brokers such as Apple and Google.

3 PRELIMINARIES

Localisation of moving entities, such as customers in

a store, is possible with multiple techniques. Recent

research reports about a precision of around 2 cm

that can be achieved with a high number of Ultra-

Wide Band (UWB) nodes, for instance (Vey et al.,

2022). High deployment cost, low mobile device

adoption and less stringent application requirement

however lead to more balanced decisions on localisa-

tion technologies. Moreover, privacy concerns have

been raised in the camera-based ﬁrst smart shopping

discussions (Bermejo et al., 2020), even leading to

broad media coverage

, leading to further trade-offs.

QR codes alleviate these concerns but require active

scanning, similar to NFC tags. Bluetooth Low Energy

(BLE) beacons are another contender in this space but

require dense deployments to achieve tolerable preci-

sion and expose a highly device-speciﬁc performance

urst et al., 2018). BLE technology is affected by the

inﬂuence of obstacles (Li

zbetin and Pe

cman, 2023).

Limitations of precision or acceptance have little ef-

fect on the use case analysed in this paper. In our

high-level workﬂow, we do not make any speciﬁc as-

sumption and instead merely assume the presence of a

suitable localisation subsystem. Table 1 gives a high-

level indication of advantages and disadvantages of

the main method families.

Similar to the localisation, there is an open design

space concerning the notiﬁcation channels for search-

ing as well as guiding and navigating users. Technolo-

gies should be inclusive, not requiring any particular

device (assuming the search could be initiated with a

e.g. Swiss railways shops https://awiebe.org/en/

sbb-uses-cameras-for-facial-recognition/

ICSBT 2023 - 20th International Conference on Smart Business Technologies

Table 1: Localisation methods and technologies.

Method Indoor Cost Precision

GSM tracking yes high low

Camera tracking yes high med-good

GPS/GNSS no low medium

BLE beacons yes medium med-good

BLE AP yes high medium

UWB yes high good

QR codes, NFC yes low –

kiosk at the entrance or via a service robot), and be of

low cost from the store owner perspective. The cor-

responding overview is given in Table 2. It is evident

that using the personal mobile phone has the advan-

tage of supporting both visual and audible notiﬁca-

tions. Electronic shelf labels (ESLs) are less intrusive

and, despite having a certain installation and main-

tenance cost as well as potential security challenges

(Mandyam et al., 2023), can be a suitable choice if

already installed especially due to their proximity to

the products. Again, our workﬂow abstracts from

the possible notiﬁcation options and only assumes the

presence of at least one.

Table 2: Notiﬁcation methods and technologies.

Method Inclusive Cost

Mobile phone no low

Mobile scanner yes high

Earplugs no low

ESL LED yes medium (battery)

ESL pageﬂip yes low

Kiosk screen yes high

4 PERSONALISED SHOPPING

WORKFLOW

This section describes our ﬁrst contribution, the work-

ﬂow that allows customers to search for a product

in the shop. The workﬂow shall be characterised by

combining personalisation, i.e. considering the con-

sumer’s search preferences, and privacy preservation,

i.e. allowing anonymous use. These characteristics

furthermore relate to the coupling of search and no-

tiﬁcation through temporarily assigned numbers or

colours, depending on the notiﬁcation channel, in or-

der to support multiple concurrent physical product

search activities within a store.

In conjunction with the various options for local-

isation and notiﬁcation, the scoping of the workﬂow

is determined according to Fig. 1. It connects the

three main activities with the necessary data struc-

tures, indicating an initial data curation effort by the

Figure 1: Scope for the personalised shopping workﬂow.

store owner which can however draw on what stores

using shelf labels already have, thus not causing addi-

tional cost related to the input data.

Referring to the detailed workﬂow speciﬁcation

expressed as sequence diagram in Fig. 2, each phase

will be described with greater attention to the most

innovative technological components. The workﬂow

either starts from the consumer who desires to search

for a product in the shop, or by the system upon the

consumer entering the shop with previous preferences

saved. For simplicity, we focus on the ﬁrst variant

(Step 1 - User search product in the shop). The user

types a text reference of the product (possibly us-

ing voice recognitition and speech-to-text conversion)

and chooses the one that interests him/her from the

list of available products, or a set of products match-

ing a desired category. Again, for simplicity, we fo-

cus on the single product search case. At this point

the application sends the data with the searched prod-

uct to the Backend server (Step 2 - Receives user re-

quest through API). The server checks the availability

of the product in the database (Step 3 - Looking for

product availability). The database is updated by the

shop owner or automatically from the shop manage-

ment system. The Backend Server responds to the

User App with a positive or negative ack of the re-

search (Step 4a - Send Response). If successful, it

returns the position of the product and some infor-

mation. In parallel, the indoor navigation algorithm

calculates the initial route to reach the product (Step

4b - Call indoor navigation system). Map and naviga-

tion information is then sent to the User App (Step 5

- Send map information).

The User App communicates via API with the po-

sitioning devices installed in the store; again for sim-

pliﬁcation, we refer to one option, BLE beacons (Step

6 - Exchange BLE info). The data exchange allows

the indoor navigation algorithm to guide the user to-

wards the shelves with the product (Step 7 - Navigate

the shop). The User App is updated indicating the

distance from the product which is recalculated dur-

ing navigation (Step 8 - Calculate product distance).

When the user arrives within ”visibility” distance of

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study

Figure 2: Sequence diagram of indoor consumer interaction.

the labels on the shelves, the User App via API passes

the information to the Cloud Label Controller (Step 9

- Call API for blinking label). This component knows

the position of the labels for each product. Moreover,

depending on the chosen notiﬁcation method, it is

aware of the assigned color of the LED or number on

the ﬂipped label itself that the user expects to see. If

the label exists and is working, the Cloud Label Con-

troller sends a command to the label to make it ﬂash

or pageﬂip (Step 10 - Send command for blinking la-

bel). At this point, the label containing the informa-

tion on the product sought ﬂashes with a speciﬁc color

or number which will be recognised by the User (Step

11 - Blink for user). The workﬂow described allows

a user who is looking for a speciﬁc product in a shop

to check its characteristics and availability and search

for it on the shelves without wasting time physically

searching for it. The ﬂexibility of the workﬂow opens

up the possibility of future developments that can, for

example, suggest a product to the user on the basis of

a proﬁling process and allow him/her to reach it, take

it and paying the product directly in the app.

5 MOBILE INTERACTION AND

NAVIGATION

This part of the paper discusses customers’ interac-

tion with the mobile application and underlying tech-

nologies. For the purpose of a better understanding

of the technologies, the case of the search for a sin-

gle product is reported. The functionality can be ex-

tended to a list of products. The mobile application is

designed for the user’s smartphone. Nowadays many

people use smartphones on a daily basis. These de-

vices are designed to work with different technolo-

gies including BLE-optimising battery consumption.

There are three subsections through this section: Lo-

cate User, the part where the navigation process takes

place; Search Product, where the customers search

for a speciﬁc product they want to buy; and Navigate

to Product, detailed information about the searched

product. Each of those refers to a subprocess from the

previously explained workﬂow, providing a concrete

realisation for the mobile device side while remaining

ICSBT 2023 - 20th International Conference on Smart Business Technologies

ﬂexible for the infrastructure side in terms of beacons

and ESLs. The interaction-centric discussion is based

on a distributed software architecture connecting the

necessary system components for search, localisation

and notiﬁcation as shown in Fig. 3.

5.1 Locate User

On the mobile application home screen, there are two

main paths on with which customers interact: Lo-

cate User and Search Product. On the Locate User

path, there is a straightforward process: indoor nav-

igation using BLE and locating the customer on the

ﬂoor map of the shopping store. Depending on the

physical deployment, the Bluetooth signals may ar-

rive from a ceiling-mounted access point; in this case,

either a single AP provides angle-of-arrival support

to determine the direction (and the user’s mobile de-

vice supports the necessary BLE protocol version), or

multiple APs are used for trilateration. Alternatively,

if no AP is available or does not provide a suitable

API, a mesh of BLE beacons can be deployed, cal-

ibrated and used for the same purpose, with conﬁg-

urable density to balance deployment cost and locali-

sation precision. In our implementation, based on ex-

isting research we provide a Neural Network-based

navigation algorithm which is competitive in accu-

racy. Combination of Bluetooth ﬁngerprinting, a Neu-

ral Network, and a Kalman ﬁlter to predict the posi-

tion of a user is used for the navigation, as expressed

in Fig. 4. The algorithm is separated into two phases,

which we refer to as the preparation and localization

phases. For the preparation phase, training data is col-

lected by moving a Bluetooth receiver device between

as many different points on the shop ﬂoor as possible

and collection signal strength (RSSI) measurements

from the BLE beacons. This data forms our BLE ﬁn-

gerprint database and serves as the training data for

a feed-forward neural network. It should be noted

that by conventional terms this model is over-ﬁtted,

as all the training data is collected from the same lo-

cation and so it would not work in a different location

unless retrained. This is however the state-of-the-art

in neural network-based ﬁngerprint localisation, and

the traditional alternative of multilateration based on

the RSSI measurements (Cant

on Paterna et al., 2017)

also requires manual calibration on location. The

model can then predict the location of the receiver (a

user’s smartphone) and the prediction passes through

a Kalman ﬁlter for smoothing in between measure-

ments to reduce the jittering of the position the user

sees on their screen.

5.2 Search Product

The Search Product path in the application is designed

for sending search queries to the database where all

the products are stored, typically an ERP, but alterna-

tively a Firestore database with generic schema that

works out of the box in our implementation. Cus-

tomers can easily search for any products they want

on this page and then connect to the localisation to

Figure 3: Mobile application architecture.

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study

Figure 4: Beacon-based navigation algorithm.

correlate both the customer position and the product

position. In addition to the search query feature of this

page, the other important function is the personalised

assignment of an anonymised results indicator, in the

form of a colour or number related to the notiﬁcation

channel.

The ESLs available on the market and used in

that research have limited ﬂashing colours for ﬂash-

ing commands, and limited preloaded e-ink pages for

pageﬂipping. To avoid customer confusion, each cus-

tomer should have a unique ﬂashing colour or display

number to track the ELSs applying to the appropri-

ate search results. Nevertheless, it is impossible to as-

sign unique colours or numbers to each customer with

the current hardware technology. Colours are usu-

ally limited to single-digit amounts, and e-ink pages

to low double-digit amounts. Therefore, each cus-

tomer will be assigned different colours or numbers

temporarily during the product search, with the mo-

bile application informing about the assignment. If all

possible colours or numbers are occupied, customers

will be informed and move to the standby list if they

wish. Other possible approaches to increase the phys-

ical notiﬁcation options beyond the phone itself are

possible, but not currently implemented by us, such

as combinations of colours and numbers, or different

blinking LED frequencies or patterns.

A sample view of the Search Product entry page

for a hypothetic store associated to our physical re-

search lab premises, as outlined in the validation sec-

tion below, is shown in Fig. 5.

5.3 Navigate to Product

Customers will arrive at the Navigate to Product Page

if they search for a product and click that product on

the Search Product page. The Navigate to Product

step is the ﬁnal yet potentially longer-run destination

of the customer. Here, customers can ﬁnd informa-

tion about the product, such as product location on the

ﬂoor map, distance from the product, price, and a de-

scriptive image. In case no assignment was performed

yet, the assigned colour or number will be ﬁrst dis-

played on the Navigate to the Product page. The same

assignment is then shown as a reminder on the Navi-

gate to the Product page. In case of all colors are oc-

cupied, customers will see that in the pop-up screen,

and if they wish, they will move to the standby list un-

til a color becomes available. Even without assigned

colour, the map-based navigation on the mobile de-

vice itself provides a suitable fallback, although it ex-

cludes customers without a phone or without the ap-

plication installed.

Again, a sample view of this subprocess is pro-

vided in Fig. 6. It shows the ﬂoor map on the left

side, with an overlay for navigation consisting of two

to three main items of information: The current loca-

tion of the customer, the location(s) of the product(s)

resulting from the search, and possibly, although not

presently implemented by us, a preferred path to col-

lect all products, for instance based on the shortest

path navigation. The addition of the path would be

more useful in practice in larger stores or malls. On

the right side, the page shows the next product in the

results list along with navigation information and the

assigned personalised indicator.

6 TECHNOLOGY FITTING

To determine the feasibility of our approach, we

have validated it under realistic conditions, in a re-

search laboratory for smart technologies, following a

real-tech approach by using commercial technology

ICSBT 2023 - 20th International Conference on Smart Business Technologies

Figure 5: Search Product sample view.

widely deployed in stores today as integration points.

This concerns especially the ERP to obtain product

information, the ESLs, and the label controller to in-

teract with the ESLs. Due to proprietary commu-

nication protocols, the tight coupling between ESLs

and label controller is unavoidable, whereas the other

technological choices permit a degree of ﬂexibility.

Table 3 contains the details on all chosen integration

points. The table also informs about an approximate

and rounded price point in C in order to facilitate

the discussion on how economic the resulting solu-

tion can be especially for smaller stores. Of particular

research interest in this context is the ability to replace

existing functionality with an open source implemen-

tation that can be used to foster innovation. This anal-

ysis provides our third contribution.

6.1 Experiment Testbed

Our testbed setup resembles a small store with three

longer shelves, a total of 30 ESLs in use to mark prod-

ucts, and 12 BLE beacons. The one-time hardware

cost is around 200

C for the ESLs, 200 C for the IoT

adapter, 700 C for the AP and 120 C for the beacons.

In order to have greater ﬂexibility for investigating

Table 3: Integration points (APIs, portals) for end-to-end

validation and comparison.

Category Solution Cost

ERP ExtendaGo 100 C/y

Label designer Vusion Studio 400 C/y

Label controller Vusion Optipick 350 C/y

Self-localisation Mist API 300 C/y

Self-localisation our approach –

Mobile application our apporach –

mobile device behavior, we have used a Linux-based

notebook instead of a mobile phone to interact with

the system.

By interacting with the ERP and generating label

images dynamically on our backend system, both for

the product and the numbered pageﬂip pages, we are

in a position to discard the label designer. Moreover,

by being able to tap into beacon-based positioning, we

are also able to discard the existing localization API.

Store owners who prefer to use those online platforms

will still be able to do so with our implementation.

We set up the products on the shelves with a 1:1

mapping to ESLs. For labels that emit BLE signals,

these could be used as a high-density grid for the nav-

igation. However, most products on the market do not

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study

Figure 6: Navigate to Product sample view.

emit signals. Therefore, in our testbed, we assume

one beacon per running shelf meter, with the aim to

lure customers nearby the target shelf area. Once

nearby, the local notiﬁcations such as label ﬂashing

and pageﬂipping can occur. An impression of the

testbed is given in Fig. 7.

Figure 7: Schematic view of physical experiment layout,

and impression of a product.

In the experiment environment, 12 BLE beacons

were positioned in the research lab space at speciﬁc

places to create a 4m x 7.2m grid spaced by 2m rep-

resenting the shopping store along room dividers rep-

resenting the shelves. The schematic grid is shown

in Fig. 7, left side. RSSI measurements were then

collected at different positions to collect a ﬁnger-

print dataset to train the Neural Network model. The

dataset was used as training data to generate a Neu-

ral Network model that predicts customers’ positions

based on RSSI readings. The RSSI readings are col-

lected in the background from the 12 beacons in the

grid every 4 seconds. The frequency of 4 seconds was

selected to ensure all beacons are given the chance to

advertise and be received by the mobile device. The

neural network inference itself is actually much faster.

Since that technology will be used in shopping stores,

the Neural Network model is purposefully overﬁtted

to get more accurate predictions, meaning the model

is only usable in the location it is trained in. Once the

model is trained, it is used to predict a position which

then passes through a Kalman ﬁlter to smoothen the

value and rule out spikes. The output of the ﬁlter is the

ﬁnal output exposed by the localisation service. The

ﬁnal result represents the location of the customer on

the ﬂoor map.

ICSBT 2023 - 20th International Conference on Smart Business Technologies

6.2 Software Implementation

The implementation of the web application used for

validation purposes was built using several underly-

ing technologies. For the front end, React JS was

used to create the user interface. React JS is a widely

used JavaScript library for building user interfaces

and allows for the efﬁcient and scalable development

of complex web applications. The back-end service

was built using Flask API, a microweb framework

written in Python. Flask API allowed for the cre-

ation of RESTful APIs that could be used to inter-

act with physical devices (label controller for ESLs,

beacon/WiFi scanner). Requests sent through the Re-

act front-end were able to interact with these APIs to

retrieve data from and send data to the IoT devices.

Finally, the Google Firebase platform was used as

the database for the web application. Firebase is a

cloud-based database service that provides real-time

updates, secure user authentication, and scalability,

making it a reliable and effective choice for the web

application’s database needs.

The software implementation resulting from our

research is available as open source (Sakman et al.,

2023). We expect that it helps accelerating the setup

of real-labs for personalised shopping and product

search in the future.

6.3 Experiment Findings

Our ﬁndings cover both economic and technological

considerations. Concerning the localisation, we can

conﬁrm that indoor navigation based on beacons is

feasible for a shop environment concerning the nav-

igation precision towards an area close to the target

shelf, and that despite additional investment, the over-

all cost may be lower if such a deployment is planned

from the start.

The mean positioning error of the model is typi-

cally 50–100cm (the resolution of the grid), although

spikes occasionally occur (Step 6 - Exchange BLE

info and Step 7 - Navigate the shop). The accuracy

is generally higher than RSSI-based multilateration as

used in the state-of-the-art and comparable with more

advanced solutions such as angle-of-arrival-based de-

tection (Paulino et al., 2022) or UWB (Vey et al.,

2022) which is more expensive to install and also

incompatible with the majority of customer smart-

phones. With further training rounds, which could be

automated by piggy-backing on cleaning or restock-

ing robots in stores, the precision can be expected to

increase slightly.

Assuming a write-off period of ﬁve years for per-

sonalised shopping equipment, store owners today

will have to invest 6850 C (including all hardware

except for the beacons) to get production-grade sup-

port for introducing ESLs and for being able to access

a raw positioning API. At this cost, they would still

need an on-top solution for personalised navigation

and physical product search. In contrast, our solution

requires the beacons but is able to discard two exist-

ing platforms and the AP as mentioned above. If WiFi

is required in the store, a more reasonably priced AP

could replace it, with a presumed cost of 200 C. This

results in a total cost of 2970 C, equivalent to 43% of

the comparative investment, and with the added ben-

eﬁt of obtaining an integrated solution for search and

navigation.

7 CONCLUSIONS

With our work, we have achieved to demonstrate

technical and economic feasibility of physical prod-

uct search and navigation to these products in stores.

Through conscious technological choices, our result

has proven to work in a real-tech environment, can

be implemented with low-cost hardware and a min-

imum set of online platforms, and works with low

power consumption. For the customers in the store,

a privacy-preserving and inclusive experience is pro-

vided, increasing the likelihood to boost sales and re-

vive physical shopping across target populations such

as digital natives, elderly people or tourists. As a

tangible result of our work, we have published an

open source software implementation (Sakman et al.,

2023).

Our research has focused on support for hybrid

notiﬁcations including navigation on the mobile de-

vice. From a human-computer interaction perspec-

tive, additional modalities to receive nagitation advice

and notiﬁcations could be built on top of our work.

This includes augmented reality (AR) navigation to

maintain the overview in larger shops with multiple

separate shelves hosting the desired products. Our im-

plementation is prepared for this modality and we are

in the progress of building the ﬁrst AR-based naviga-

tion as sketched in Fig. 8.

Our research moreover leads to a unique economic

value proposition. According to our interation with

store owners, especially for high-value stores there is

a high need for such a solution. The economic po-

tential is therefore in commercialising the research re-

sults. If a price tag of around 8000 C is aimed at, only

slightly above the current mark but with better func-

tionality, this would imply a possible range of more

than 5000 C, divided into both the development cost

and sales proﬁts.

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study

Figure 8: Ongoing augmented reality integration prototype.

In our follow-up work, we will therefore focus on

the following directions. First, we will study the per-

formance and price point for AR-based navigation.

Second, we will investigate the scalability of the solu-

tion in larger stores, and the operational/maintenance

perspective including the adoption of more sustain-

able technology such as solar-powered ESLs and bea-

cons that are technically suitable for indoor shopping

lighting conditions. As a third research direction, in

order to support both customer and shop owner we

will investigate additional features. For that matter,

we expect an emerging mobile application to incor-

porate cutting-edge technology that enables users to

search for a product and receive recommendations for

related items. Once the customer has added all de-

sired products to their basket, the application will gen-

erate the shortest walking path from their current lo-

cation to the cash service, including all products in the

route. The path precision will be increased by lever-

aging multi-sensor fusion and multi-perspective con-

sensus voting (Gkikopoulos et al., 2022). Along this

path, the application will suggest additional products

nearby to the customer as they navigate through the

market. If the customer accepts any of the recom-

mended products, the application will automatically

generate a new walking path with the same logic. This

feature will provide a seamless shopping experience

for users, helping them discover new products while

efﬁciently navigating through the store.

ACKNOWLEDGEMENTS

Research partially supported by Innosuisse - Swiss In-

novation Agency in project Indoor Navigation for Per-

sonalised Shopping/62895.1 INNO-ICT. This work

has been partially supported by the European Union -

FSE, PON Research and Innovation 2014-2020 Axis

I - Action I.1 ”Dottorati innovativi con caratteriz-

zazione industriale” CUP: J75F20000100007.

REFERENCES

Bermejo, C., Chatzopoulos, D., and Hui, P. (2020). Eye-

shopper: Estimating shoppers’ gaze using cctv cam-

eras. In Proceedings of the 28th ACM International

Conference on Multimedia, pages 2765–2774.

Boden, J., Maier, E., and Dost, F. (2020). The effect of

electronic shelf labels on store revenue. International

Journal of Electronic Commerce, 24(4):527–550.

Cant

on Paterna, V., Calveras Aug

e, A., Paradells Aspas, J.,

and P

erez Bullones, M. A. (2017). A bluetooth low

energy indoor positioning system with channel diver-

sity, weighted trilateration and kalman ﬁltering. Sen-

sors, 17(12).

Chen, C., Huang, T., Park, J. J., Tseng, H., and Yen, N. Y.

(2014). A smart assistant toward product-awareness

shopping. Pers. Ubiquitous Comput., 18(2):339–349.

Chen, J. V., Ruangsri, S., Ha, Q., and Widjaja, A. E. (2022).

An experimental study of consumers’ impulse buying

behaviour in augmented reality mobile shopping apps.

Behav. Inf. Technol., 41(15):3360–3381.

Deshmukh, S., Fernandes, F., Chavan, A., Ahire, M., Borse,

D., and Madake, J. (2022). SANIP: shopping assis-

tant and navigation for the visually impaired. CoRR,

abs/2209.03570.

urst, J., Chen, K., Kim, H.-S., and Bonnet, P. (2018).

Evaluating bluetooth low energy for iot. In 2018

IEEE Workshop on Benchmarking Cyber-Physical

Networks and Systems (CPSBench), pages 1–6.

Garaus, M., Wolfsteiner, E., and Wagner, U. (2016). Shop-

pers’ acceptance and perceptions of electronic shelf

labels. Journal of Business Research, 69(9):3687–

3692.

Gkikopoulos, P., Kropf, P. G., Schiavoni, V., and Spillner, J.

(2022). AVOC: history-aware data fusion for reliable

iot analytics. In Zhang, K., Gherbi, A., and Bellavista,

P., editors, Proceedings of the 23rd International Mid-

dleware Conference: Industrial Track, Middleware

2022, Quebec, Quebec City, Canada, November 7-11,

2022, pages 1–7. ACM.

Gong, T., Wang, C.-Y., and Lee, K. (2022). Effects of char-

acteristics of in-store retail technology on customer

citizenship behavior. Journal of Retailing and Con-

sumer Services, 65:102488.

Heyns, R., Ndiaye, M., and Abu-Mahfouz, A. M. (2021).

Enabling user-oriented features at the edge: A case

of an iot-based smart shopping cart. In 30th IEEE In-

ternational Symposium on Industrial Electronics, ISIE

ICSBT 2023 - 20th International Conference on Smart Business Technologies

2021, Kyoto, Japan, June 20-23, 2021, pages 1–6.

IEEE.

Kellermayr-Scheucher, M., H

orandner, L., and Brandtner,

P. (2022). Digitalization at the point-of-sale in gro-

cery retail - state of the art of smart shelf technology

and application scenarios. Procedia Computer Sci-

ence, 196:77–84. International Conference on EN-

TERprise Information Systems / ProjMAN - Interna-

tional Conference on Project MANagement / HCist -

International Conference on Health and Social Care

Information Systems and Technologies 2021.

Linzbach, P., Inman, J. J., and Nikolova, H. (2019). E-

commerce in a physical store: Which retailing tech-

nologies add real value? NIM Marketing Intelligence

Review, 11(1):42–47.

zbetin, J. and Pe

cman, J. (2023). Possibilities of using

bluetooth low energy beacon technology to locate ob-

jects internally: A case study. Technologies, 11(2).

Mandyam, G. D., Scagnol, M., and Graube, N. (2023). Se-

cure onboarding and management of electronic shelf

labels in retail. In 2023 15th International Conference

on COMmunication Systems & NETworkS (COM-

SNETS), pages 96–101.

Miralles, P. S., Gonzalez, L. F., Yu, X., and Saniie, J. (2022).

Assisting visually impaired people using autonomous

navigation system and computer vision for grocery

shopping. In 2022 IEEE International Conference on

Electro Information Technology, EIT 2022, Mankato,

MN, USA, May 19-21, 2022, pages 203–208. IEEE.

Paulino, N., Pessoa, L. M., Branquinho, A., and Gonc¸alves,

E. (2022). Design and experimental evaluation of a

bluetooth 5.1 antenna array for angle-of-arrival esti-

mation. In 2022 13th International Symposium on

Communication Systems, Networks and Digital Signal

Processing (CSNDSP), pages 625–630.

Perera, G. S. T., Madhubhashini, K. W. R., Lunugalage,

D., Piyathilaka, D. V. S., Lakshani, W. H. U., and

Kasthurirathna, D. (2020). Computer vision based in-

door navigation for shopping complexes. In ICVISP

2020: 4th International Conference on Vision, Image

and Signal Processing, Bangkok, Thailand, Decem-

ber, 2020, pages 15:1–15:6. ACM.

Riegger, A.-S., Klein, J. F., Merfeld, K., and Henkel, S.

(2021). Technology-enabled personalization in retail

stores: Understanding drivers and barriers. Journal of

Business Research, 123:140–155.

Ruijgrok, P., Frasincar, F., Vandic, D., and Hogenboom, F.

(2018). Ontonavshop: an ontology-based approach

for web-shop navigation. Journal of Web Engineering,

pages 241–269.

Sakman, M. C., Gkikopoulos, P., Martella,

F., Villari, M., and Spillner, J. (2023).

Smart personalised shopping web prototype.

https://doi.org/10.5281/zenodo.7859411.

Vey, Q., Dalc

e, R., van den Bossche, A., and Val, T.

(2022). Indoor UWB localisation: Locura4iot testbed

and dataset presentation. In Oteafy, S., Bulut, E.,

and Tschorsch, F., editors, 47th IEEE Conference

on Local Computer Networks, LCN 2022, Edmonton,

AB, Canada, September 26-29, 2022, pages 258–260.

IEEE.

Indoor Navigation for Personalised Shopping: A Real-Tech Feasibility Study