BUSINESS INTELLIGENCE BASED ON A WI-FI REAL TIME

POSITIONING ENGINE

A Practical Application in a Major Retail Company

Vasco Vinhas, Pedro Mendes and Pedro Abreu

FEUP - Faculdade de Engenharia da Universidade do Porto, Rua Dr. Roberto Frias s/n, Porto, Portugal

DEI - Departamento de Engenharia Informática, Rua Dr. Roberto Frias s/n, Porto, Portugal

LIACC - Laboratório de Inteligência Artificial e Ciência de Computadores, Rua do Campo Alegre 823, Porto, Portugal

Keywords: Wi-Fi, Real-Time Tracking, Business Intelligence, Retail.

Abstract: Collecting relevant data to perform business intelligence on a real time basis has always been a crucial

objective for managers responsible for economic activities on large spaces. Following this emergent need,

the authors propose a platform to perform data gathering and analysis on the location of people and assets

by automatic means. The developed system is retail business oriented and has a fairly distributed

architecture. It couples the core elements of a real-time Wi-Fi based location system with a set of developed

functional views so to better explicit the information that one can observe for each tracked entity, the

undertaken path on the space, demographic concentration patterns. Tests were conducted on a real

production environment as a partnership outcome with a major player in the retail sector and the obtained

results were completely satisfactory having the managers confirmed the provided knowledge relevance.

1 INTRODUCTION

People and asset’s tracking and positioning data

collection – even if with the sole purpose of locating

those entities – on a given space, has always been a

need for those with management responsibilities,

regardless their economic sector. Up until thirty

years ago, the mechanisms to pursue such objective

were manifestly out of reach of the masses mainly

due to financial reasons. At that time, most systems

were part of military projects and made use of radar

technology to passively detect air or marine traffic

(LaFollette et al., 1991), although in this last case

the positioning was achieved through an active

procedure – the traceable water vehicle had to

intentionally transmit a radio signal with its latitude

and longitude encoded (Yu, 2005).

However the recent advances in several domains

such as computer technology, video capturing and

positioning sensors changed the previously

mentioned premises. Nowadays video surveillance

systems, with mid-end camera resolution are cost-

effective even in the context of small and medium

enterprises – SMEs – with the non minor advantage

of complete digital storage and processing. In spite

of a video feed does not provide data on the

positioning of entities, in the corresponding

camera’s scope, in way that such information can be

retrieved automatically. With the latest wireless

networks proliferation, other approaches emerged,

with the obvious advantage of automatically

providing clean data at a semantic level, displayable

and storable in a straightforward way. Among the

technologies that fit in this situation one shall refer

to: Bluetooth; Wi-Fi; RFID; GPS; amongst others.

Naturally, each of them encloses its strengths and

flaws. Some require the traceable items to have an

active behavior while others process them as passive

items. The coverable area and the error involved are

also important factors to be considered as well as the

required resources, such as power and equipment

density. Some of them remain unwavering with

layout changes while others require a full system

recalibration.

The research work proposed in this document

materializes itself on a system that takes advantage

of typical redundant Wi-Fi networks and is based on

a positioning engine built on top of these. The

system provides a visualization platform of such

data on a real time basis. The information can be

displayed through several perspectives, including

fully scalable concentration grids, clean positioning

Vinhas V., Abreu P. and Mendes P. (2009).

BUSINESS INTELLIGENCE BASED ON A WI-FI REAL TIME POSITIONING ENGINE - A Practical Application in a Major Retail Company.

In Proceedings of the 11th International Conference on Enterprise Information Systems - Databases and Information Systems Integration, pages 11-16

DOI: 10.5220/0001858000110016

 SciTePress

of the elements at hand or even a vision inference

assuming that the items to be tracked are associated

with people. This system was originally developed

for a corporate entity operating in the global retail

market, whose institutional designation is obfuscated

by commercial reasons. Following the same line of

reasoning all data provided; application’s

screenshots and shop floor layouts were based on

real production data and were obfuscated, without

compromising functionality and intelligibility.

2 STATE OF THE ART

Despite the advances in this field, leading to

noteworthy breakthroughs, some issues still remain

to be addressed. From these, one ought to point out

those concerning occlusion, which are classifiable in

three distinct categories: self-occlusion, where part

of the object to trace, typically articulated ones,

occludes another part; inter-object occlusion where

two traceable objects occlude each other; and

occlusion by the background scene where the

physical space’s properties propitiate a camouflage

of the object to track. For the inter-object occlusion

research works like (MacCormick and Blake, 2000)

(Elgammal et al., 2000) exploit the a priori

knowledge of the object’s position and attempt to

predict possible occlusions and solve them

smoothly.

Considering other domain issues, the hurdles that

arise when tracking entities with non-linear

movements must be addressed as this point

constitutes one of the major problems in tracking

persons in non controllable environments. A

pragmatic approach to this situation could point out

to time resolution diminishment in which tracking is

achieved or the loosening of the real time

requirements. When the solution is based on video

feed processing, there are some additional problems

directly related to the inherent technology. The

scene’s illumination should be adequate so to

facilitate the image binarization processes and the

network shall be enhanced and optimized in order

not to become a system’s bottleneck.

2.1 Non-Image based Systems

GPS is commonly used to perform real time

detection of different types of vehicles and as a base

tool to analyze their motion (Yu, 2005) (Nejikovsky

et al., 2005). Yet in this scenario, the technology is

applicable in three distinct ways: Cellular Based

Tracking is a solution based in a conventional

mobile phone with a GPS receiver that emits the

vehicles position every five minutes. Wireless

Passive Tracking has core advantage in using GPS,

because once it is set up, there is no monthly fee

associated, and with it is possible to collect

information like for instance, how many pit stops are

made by a vehicle in a given route and how fast is it

moving. Its worldwide coverable area constitutes an

ideal solution for transporting companies.

The radio frequency identification (RFID) is a

non-standardized wireless tag location method. This

technology requires a RFID receiver and a set of

tags which can be divided in two different groups:

Passive - only detectable on a 13-meter radius from

the receiver; Active Group - have their own internal

power source, offer both reliable detection on a

larger scale, and more resilience to occlusion

problems caused for possible obstacles in the

environment. The two major issues about this

technology are the receiver’s cost and the active

tag’s average unit price, mainly due to the need of an

independent power supply (C. et al., 2007).

Wi-Fi 802.11 technology allows establishing

connectivity between a set of devices allowing an

easy setup of wireless data networks in academic

campus, industrial facilities, public buildings, etc.

The technology behind these networks can also be

used for designing a tracking system. By reusing

commonly existing data networks and a low level

protocol it is possible to create a tracking system on

top of this infrastructure. Another advantage of this

technology is the possibility of tracking an object by

using a single access point, though the precision will

weaken due to the lack of signal triangulation.

Because of its technical details, the impact of issues

such as occlusion and signals loss is reduced to a

residual level especially in environments, which do

not have high concentration of metallic materials

(Mingkhwan, 2006).

Bluetooth is a wireless protocol available on any

modern mobile equipment, allowing data exchange

between multiple devices. It is exclusively used for

short-range communications, which is the cause for

its poor applicability on tracking systems. The

battery consumption is also remarkably high

(Jappinen and J., 2007).

2.2 Image based Systems

Thermal signature systems are one of the most

expensive technologies for locating an object on a

scene. The main purpose of these solutions is the

reconnaissance and processing of thermal images.

These systems attempt to recognize specific thermal

ICEIS 2009 - International Conference on Enterprise Information Systems

signatures of the entities being tracked although

some items might not have them.

Multi-camera video surveillance is a technique

that uses a set of cameras to track entities in a given

environment. By accurately crossing the information

coming from cameras which have intersecting

frustums one can enhance the precision of the

system, despite the possible processing overhead

(Mittal and Davis, 2003). Camera calibration and its

positioning on a tridimensional space may be

performed/inputted manually (Collins et al., 2001)

(Cai and Aggarwal, 1999) or even automatically

despite the obvious errors that may occur if the last

method is undertaken on an unsupervised way (Lee

et al., 2000) (Khan and Shah, 2003). Although these

systems are actually used in some scenarios, some

issues still persist. The need to have a dedicated

network for the system, the expenditure required for

high-resolution equipment and the computational

demands, are still major concerns. In the literature,

some research work tries to optimize the

performance of these systems by minimizing the

need of brute force computation (Mittal and Davis,

2003) and by using overlapping camera views

(Huang and Russel, 1997) (Javed et al., 2003)

(Kettnaker and Zabih, 1999).

3 PROJECT DESCRITPION

In this section, the undertaken project is described in

detail in what regards its several components and

analysis perspectives.

3.1 Project Description

As revealed through Figure 1, the elaborated

technical design contemplates several independent

modules that communicate in order to achieve a

systematic unit. The first action, in offline mode,

consists in conducting a complete map

creation\edition. The user shall specify, amongst

other details, depicted in subsection 3.2, the image

file representing the shop floor layout, the used scale

and identify, by using a draw-like tool, what items

are to be visible by visitors as well as spawn areas.

This information is compiled in a XML file for both

the position engine and real-time monitoring tool

and submitted to the mentioned database for the

historical BI application.

The Wi-Fi tag consists in a miniaturized active

802.11 a/b/g board with a couple of power batteries

attached. These are configured to connect to a

specific Wi-Fi network – security, DHCP and other

network configurations are also possible – and to

directly communicate with the position engine. By

using this kind of wireless technology, it is possible

to reuse partially or totally the client’s network

infrastructure, having only, for special requirement,

a high density of access points as the accuracy

naturally increases with this factor.

The used position engine periodically collects data

from the tags and updates their position against a

pre-loaded localization model. This model is very

similar to the produced from the map editor differing

only in the available information regarding visible

objects. This model also requires a previous offline

site survey for measuring Wi-Fi signal strength and

for network items – routers and access points –

precise localization. The engine is also web-enabled

and supports a HTTP/XML API so that third-party

applications can interact with it, therefore accessing

localization and status information regarding each

individual registered tag.

Figure 1: System’s Global Architecture.

Using this communication protocol, the developed

real-time monitoring server is responsible for

gathering, at a specific periodicity – typically equals

to the position engine frequency – every tag’s valid

location data. With this information, this module is

directly responsible for updating the database and

caching the session’s data for the real-time

monitoring application.

Having the continuous up-to-date database as a solid

information reference, it was possible to enable both

real-time and historical business intelligence

applications. For real-time knowledge extraction, it

was only used data referring to active sessions, for

historical analysis, and delegating all the process

BUSINESS INTELLIGENCE BASED ON A WI-FI REAL TIME POSITIONING ENGINE - A Practical Application in a

Major Retail Company

effort to the database engine, specific and dynamic

time windows were used to filter data.

The versatility of such application must be referred

as it congregates both web-enabled features and zero

data process – it is all delegated to the database

engine and allocated in a dedicated server – enabling

its usage in a wide range of devices, including PDAs

and mobile phones.

3.2 Real-Time Monitoring Application

This unit is responsible for accessing location data

from the online position engine and, simultaneously,

using a multithread sliding-window approach,

commit new data to the database and compute data

into visual information following distinct

approaches. Each of these tool’s facets is mapped

into a distinct GUI tab enabling independent

analysis.

Before BI extraction, there are two mandatory

configuration requirements that must be met: the

first consists in loading the shop floor layout; the

second consists in establishing a HTTP connection

with the position engine. The third, optional,

requirement resides in opening a database

connection for online data insertion. If this is not

met, there are a virtual infinite number of

application’s instances that can be run at the same

time, enabling simultaneous BI extraction for

numerous organization’s members.

3.2.1 Real-Time Tracking

This feature enables complete item tracking by

overlapping current item’s position directly over the

loaded map information. This capability is

independent of the GUI’s windows size and/or shape

as the coordinate systems are always synchronized.

As depicted is Figure 2, it is also possible to enable

session’s path history directly over shop floor image,

therefore enhancing visual perception of both

current positions and session routes.

In the presented screenshot, it is possible to see

that at the time it was taken, there is only a single

client in the shop, whose location is near the layout

center and that his visit concentrated mainly in direct

routes in the north corridors.

If there would be more clients present, it would

be possible to perceive their current location by their

blue dots representation as well as, optionally

visualize their session routes.

This feature tries to emulate a bird’s eye view of

the all shop floor, with the possibility of recalling all

the client’s routes as if they left a visible trail while

they are touring the facility.

Figure 2: Real-Time Tracking with Visible Path.

3.2.2 Demographic Concentration

The demographic concentration feature enables

space division using a completely dynamic, in terms

of cell size, matrix grid that is colored according to

demographic concentration at given time. Once

again, this computation can be performed using only

strictly real-time positions or by recalling all current

session’s route positions.

Figure 3: Real-Time Tracking with Visible Path.

Each cell is then colored following a three-color

gradient where blue stands for lower levels of

concentration, green for intermediate levels and red

for high levels. A special remark is due to the fact

that the entire colour spectrum is used for the

mentioned purposes.

As illustrated in Figure 3, the concentration matrix is

drawn, with a partially transparency, over the shop

floor plant. In this practical example, the illustration

corresponds to a situation where there were two

clients present in the shop, and it was selected, for

concentration calculus concerns, not only current

position but all routes’ positions.

This tool’s feature enables swift, yet effective,

hot and cold zones analysis, current bottlenecks,

unvisited versus most visited areas and online queue

alerts and management. Also, as reported in the

previous subsection, all the graphical information is

independent of the GUI’s window size or shape.

ICEIS 2009 - International Conference on Enterprise Information Systems

3.3 Real-Time & Historical BI

Application

In order to extract significant business intelligence

knowledge based on historical data and not only

real-time information, the authors decided to make

an immediate use of the raw position data stored in

the database – in conjunction with visible structures

pre-processed information – so that significant retail

BI facts could be easily extracted and presented to

end-users – both shop managers and organization’s

top executives. Taking advantage of using Oracle as

equally laboratory and production database, Oracle’s

Application Express was used to generate a web

application responsible for processing data and, most

important, aggregate information in an

understandable way.

Apex’s engine is directly embedded to the

database, thus directly dealing with client’s web

requests. With this architecture, several systems can

easily access BI application as all heavy processing

is database server’s responsibility, leaving the client

with only chart rendering computation.

Despite extensive tool’s analysis being object of

discussion in the next section, by using the described

architecture and technology, several practical

measures were considered for extraction, namely:

hot and cold zones, average visit duration and

distance and number of visits. All these benchmarks

can be targeted to use only real-time data or recall

specific historical time windows.

4 RESULTS

One is able to state that the system performs swiftly

as a whole as its constituting parts are able to

exchange data harmoniously both in laboratory and

production environments. As for the core tool, the

real-time monitoring application, it was observed

that the displayed data is quickly interpretable, thus

making its purpose: enabling real time business

intelligence. In fact, just by looking, for instance, to

the grid tab, managers are able to see what the

current hot and cold zones are and make decisions

based on such information. According to the retail

store managers, by using the application, one is able

to, real time, assess the effectiveness of several

decisions, namely: the store’s layout, number of

open cash registers or the work being carried out by

employees in the zones where the customer service

is made face-to-face.

Figure 4: Web Application – Zone Distribution.

Figure 4 represents the retail store grid

concentration which has the layout as shown on

Figure 2. The data comprehends one week of

activity where each chart entry corresponds to the

number of accounted presences, on a specific zone,

every two seconds. The zones are numbered from

one to the number of rows times the number of

columns. Zone 1 is located on the top left corner.

By analyzing the exposed charts, some

interesting aspects on the customer’s behaviour

when visiting the shop, can be assessed. Although

zones 13 and 14 correspond to the store’s entry, it is

observable that very little time is spent by the

customers. Another interesting point resulting from

this analysis concerns the zones where more

presences were accounted. All of them include

central corridors, which are naturally part of the

paths that clients have to cross to reach the products

they are searching for.

5 CONCLUSIONS & FUTURE

WORK

One ought to affirm that all the most important goals

were fully accomplished. First, a fully functional

item real-time location and tracking system were

pursued. The Wi-Fi based solution, not only

complied but did it reusing most of the client’s

network infrastructure, thus reducing negative

impact in both financial and logistic terms.

Secondly, the designed system’s architecture proved

to be reliable, efficient and flexible enough to

contemplate vast and diverse application scenarios.

These features are more visible in what concerns

dynamic and user-friendly shop floor layout

definition. Also within this scope the distributed

communication architecture performed as predicted

BUSINESS INTELLIGENCE BASED ON A WI-FI REAL TIME POSITIONING ENGINE - A Practical Application in a

Major Retail Company

enabling computation across distinct machines,

therefore improving overall performance and

reliability. This feature also enabled simultaneous

multi-terminal access, both to the real-time analysis

tool and the historical statistical software.

Taking into consideration the project’s tools,

both were classified, by the retail company’s end-

users – mainly shop managers, marketing directors

and board administrators – as extremely useful and

allowed swift knowledge extraction, preventing

them the excruciating, and not often useless process

of getting through massive indirect location data.

The immediate visual information provided by the

system proved to be effective in direct applications

such as queue management and hot and cold zones

identification, and most significant, in what concerns

to visit’s pattern extraction across different time

dimensions, thus enhancing marketing and logistic

decisions’ impact. One must refer to Oracle’s APEX

technology adoption. It has demonstrated to be able

to allow multiple simultaneous accesses and,

consequently, dramatically enhancing analysis

empowerment, while, at the same time, eliminated

heavy data computation from end-users terminals,

concentrating it in controlled and expansible

clusters.

Regarding future work areas, there has been

identified a set of potential project enhancements

that would be able to suppress some hurdles and,

somehow, wide potential new applications.

Considering business intelligence extraction, it

would be useful to build or reuse an inference engine

capable of determining the odds of a given customer

turn right or left in the next decision point, taking for

that, into account his past actions and comparing

them to other customers’ action that are classified in

the same cluster. This aspect should be also applied

to historical data so that efficient customer clusters

would be defined and maintained.

There have been identified several application

domains that go beyond the retail segment. Amongst

these, one shall mention the possible system’s

adoption by large warehouse management where

traffic jams are not unusual.

As a summary, it is fair to state that the project’s

initial ambitions were fully met and that the close

cooperation with an important stakeholder in the

global retail market was extremely important for

better measuring the system’s positive impact and

potential firstly unseen applications. The technology

transparency, allied with the future work areas, is

believed to greatly improve potential applications in

several domains, thus significantly widening the

project’s initial horizons.

REFERENCES

C, C., J, Y., and W, J. Determining technology trends and

forecasts of RFID by a historical review and

bibliometric analysis from 1991 to 2005. In

Technovation27, Elvisier Ltd, page 268-279, 2007.

Cai, Q. and Aggarwal, J. Tracking human motion in

structured environments using a distributed camera

system. In IEEE Trans. Patt. Analy. Mach.

Intell.,pages 1241–1247, 1999.

Collins, R., Lipton, A., Fujiyoshi, H. and Kanade, T.

Algorithms for cooperative multisensory surveillance.

In Proceedings of IEEE, pages 1456–1477, 2001.

Elgammal, A., Duraiswami, R., Harwood, D. and Davis,

L. Background and foreground modeling using

nonparametric kernel density estimation for visual

surveillance. In Proceedings of IEEE, pages 1151–

1163,2000.

Huang, T. and Russel, S. Object identification in a

Bayesian context. In Proceedings of International

Joint Conference on Artificial Intelligence, pages

1276–1283, 1997.

Jappinen, P. and J, P. Preference-aware ubiquitous

advertisement screen. In e-commerce 2007

Proceedings, IADIS Press, pages 99–105, 2007.

Javed, O., Rasheed, Z., Shafique, K. and Shah, M.

Tracking across multiple cameras with disjoint views.

In IEEE International Conference on Computer Vision

(ICCV), pages 952–957, 2003.

Kettnaker, V. and Zabih, R. Bayesian multi-camera

surveillance. In IEEE Conference on Computer Vision

and Pattern Recognition, pages 117–123, 1999.

Khan, S. and Shah, M. Consistent labeling of tracked

objects in multiple cameras with overlapping fields of

view. In IEEE Trans. Patt. Analy. Mach.Intell., pages

1355–1360, 2003.

Lee, L., Romano, R. and Stein, G. Monitoring activities

from multiple video strams:establishing a common

coordinate frame. In IEEE Trans. Patt.Recogn.

Mach.Intell., pages 758–768, August 2000.

MacCormick, J. and Blake, A. Probabilistic exclusion and

partitioned sampling for multiple object tracking. In

Int. J. Comput. Vision, pages 57–71, 2000.

Mingkhwan, A. Wi-fi tracker: An organization wi- fi

tracking system. In IEEE CCECE/CCGEI,

Ottawa,page 1353-1356, 2006.

Mittal, A. and Davis, L. M2 tracker: A multiview

approach to segmenting and tracking people in a

cluttered scene. In Int.J. Comput. Vision

, pages 189–

203, 2003.

Nejikovsky, B., Kesler, K. and Stevens, J. Real time

monitoring of vehicle/track interaction. In Rail

Transportation, pages 25–31, 2005.

Yu, Z. GPS train location and error analysis which based

on the track fitting of the railway’s geometric locus. In

ICEMI 2005: Conference Proceedings of the Seventh

International Conference on Electronic Measurement.

ICEIS 2009 - International Conference on Enterprise Information Systems