BUSINESS INTELLIGENCE THROUGH REAL-TIME TRACKING
Using a Location System Towards Behaviour Pattern Extraction
Pedro Abreu, Vasco Vinhas
FEUP - Faculty of Engineering of the University of Porto, R. Campo Alegre 1021, Porto, Portugal
DEI - Department of Informatics Engineering, R. Campo Alegre 1021, Porto, Portugal
LIACC - Artificial Intelligence and Computer Science Laboratory, R. Campo Alegre 1021, Porto, Portugal
Pedro Mendes
FEUP - Faculty of Engineering of the University of Porto, R. Campo Alegre 1021, Porto, Portugal
DEI - Department of Informatics Engineering, R. Campo Alegre 1021, Porto, Portugal
INESC Porto - Institute for Systems and Computer Engineering of Porto, Rua Dr. Roberto Frias, 378, Porto, Portugal
Keywords:
Location System, Real-Time, Knowledge Extraction, Business Intelligence.
Abstract:
Nowadays, tracking systems constitute an important knowledge support in order to compute important mea-
surements in companys processes efficiency. As consequence of that, this project proposes a methodology and
an application, based on a tracking system to obtain, by automatic means, dynamic location data on items.
This solution assumes that the client carries or drives an item of some kind. In each item there is an identify-
ing tag attached and hidden in order to make the item at hand detectable by all the sensors that are scattered
around the area. Because of the fact that the tag is light and hidden and also has no information regarding
the specific person/agent this process is completely transparent to the client or robot that is being implicitly
tracked. This system produces real-time shop floor visualization maps with intelligible data on online item
localization; individual item complete path routes; online and historical population density rates and path
routes concentration; and also item vision enabled concentration maps as emulation for item omnidirectional
vision considering occlusions. This proposed system might be useful in many different areas, for instance in
a traditional retail environment tracing clients through a commercial area or enabling item tracking and route
analysis in a hospital.
1 INTRODUCTION
Detecting behavioral patterns is a challenging task
that marketing and distribution companies face. The
issue has been addressed through the past years on
several perspectives like deterministic psychology
(Luce, 1999; Choustova, 2007). However due to their
active consciousness, human beings are extremely un-
predictable and so these methods failed to provide any
accurate data that could be used for industrial pur-
poses. Having these approaches been unsuccessful,
statistical inference with large data sets is still one of
the most powerful tools available.
Nowadays, tracking systems can represent a pow-
erful tool to support monitoring activity. With these
systems, some performance measurements on com-
pany process efficiency can be obtained regardless
of the specific tracked target. This research work
presents a methodology and a tool, based on a track-
ing system to obtain by automatic means, movements
data on these elements.
The presented solution assumes that the client car-
ries or drives an item of some kind, inside the space.
It is also assumed that these items are outside the en-
try of the traceable space, although already inside of
its admissible space. In each of the items there is an
identifying tag attached and hidden in order to make
the item at hand detectable by all the radars that are
scattered around the area. By detecting it, the client
or robot is being implicitly tracked in a completely
transparent way for him/it since the tag is light and
hidden and also has no information regarding the spe-
cific person/agent.
Several benefits can be withdrawn from using a
system such as this. Instantaneously one could use it
to monitor the traceable area in a more effective way
51
Abreu P., Vinhas V. and Mendes P. (2008).
BUSINESS INTELLIGENCE THROUGH REAL-TIME TRACKING - Using a Location System Towards Behaviour Pattern Extraction.
In Proceedings of the International Conference on e-Business, pages 51-57
DOI: 10.5220/0001909900510057
Copyright
c
SciTePress
than looking at dozens of screens with images from
security cameras. Live monitoring of the elements’
positions on a specific area allows managers to iden-
tify congested sectors. It also allows him to identify
hot and cold zones which may be related to the inter-
ests points at hand or, for instance, to a local security
issue or other type of event such a medical emergence.
Regarding long term data analysis the advantages
of using such a system are several and may be more
than the ones presented next. By analyzing all the
paths taken by the elements, it is possible to obtain the
hot zones on any time frame and thus evaluate the suc-
cess rate of a given promotional campaign and among
other things. Erroneous and random movements may
also be correlated with a security issue, and thus this
system could also represent an interesting addition to
conventional security systems. The results later ob-
tained would clearly point out the success of the lay-
out redesign. Of course these last measures imply cor-
relating client positions with goods bought by him.
The paper is structured as follows: section 2 de-
scribes the current state of the art regarding the sev-
eral areas of knowledge involved in the development
of the system described in this paper. Section 3 de-
scribes the tool that simulates the environment as well
as its architecture and core functionalities. Section 4
discusses the results obtained so far and the next sec-
tion concludes the paper by summarizing the focus
of this research work and pointing out future lines of
discussion.
2 STATE-OF-THE-ART
Nowadays, tracking systems represent an important
research area as their applications are transversal to
several areas of knowledge (e.g computer science,
medicine, simulation, robotics as well as industrial
tracks). In the past few years, technology has evolved
in order to provide more accurate measurements. In
the robotics area, for better modeling the world, it is
extremely relevant to accurately process the signals
received by the multiple sensors involved. Locating
objects of the real world to the modeled one is a criti-
cal task for the appliance of the navigation algorithms
and methodologies. Following these advances the
work published by Hyunwoong Park (Park H., 2006)
presents a new kind of sensor system with multiple
rotating range sensors. Such system allows a robot
to guide itself on a priori unknown world. On the
other hand these tracking systems also find interesting
applications on scenarios where the context environ-
ment is already known. Regarding this last system,
locating elements assumes a crucial role. To achieve
this goal, several technologies have been used. By
doing a brief comparison, it is observable that all of
them have their strengths and flaws concerning char-
acteristics like the cost in terms of initial investment
and maintenance. There are others related to environ-
ment specificities. Among these last, other parame-
ters such as coverable area, tracking detection errors
and occlusion problems should also be considered.
One of the most effective technologies is also one
of the most expensive ones and concerns detection of
thermal signatures. This technique is appropriate to
living organisms which emit particular heat waves.
One particular application of this technology is the
monitoring of the fauna in the ocean (Raizer, 2003).
Another interesting technology is Bluetooth because
most modern mobile equipments are prepared to send
a receive data though this protocol. Although the
initial investment is low the coverable area is not
very wide and battery consumption is high, in relative
terms (Jappinen P., 2007).
The cheapest solution is infrared based. Even
though its price attractiveness, infrared systems tend
to fail on most real environments because the signal
is unable to reach the target if there is an opaque ob-
ject between the receiver and the target (Krotosky J.,
2007).
Two of the most emergent technologies for track-
ing are RFID and Wi-Fi based. The first one still lacks
standardization which is somehow reflected in the
pricing of both receivers and transmitters. It is based
on high frequency radio waves having the detectable
tags a passive or an active response. Passive tags are
only detectable on a 13 meter radius and are used
for instance on the new USA passports. Active tags,
alternatively, are detectable on a much wider range
but are more expensive (around 400%) essentially be-
cause the tags require an independent power supply
(Chao C., 2007). Wi-Fi may also be considered as a
tracking technology. This approach is mainly used for
creating wireless computer networks but in this case
the involved tracking only requires the usage of the
low level protocols. This type of solution is interest-
ing because it makes possible reusing existent com-
puter networks for other proposes and takes advan-
tages from possible simple detection with at least one
access point. With only one access point the system’s
precision may not be very high but there is no need
for triangulation. Occlusion problems and signals
losses, with the use of this technique will be reduced
to a residual level in both open spaces and indoors
considering that indoor spaces do not have signifi-
cant metal structures within the walls) (Mingkhwan,
2006).
Another area that suffered several developments
ICE-B 2008 - International Conference on e-Business
52
in the past few years concerns world modeling. In
this research area there are relevant research topics,
most of them related to computer graphics. Most
of the current advances focus on three dimensional
(3D) worlds. In this scope, the evolution on com-
puter graphics is the most notorious. Nowadays, sim-
ple systems are able to represent complex 3D world
including high resolution textures, detail animation
(Vazquez, 2007) and weather condition (Grudzinski,
2007). It is even possible to recreate a 3D world from
textual specifications (Moura J., 2004). On top of
3D world, many algorithms are applicable in other to
optimize rendering performance and obtaining world
data such as visible objects of a certain point.
These algorithms are too complex for most real-
time tracking systems and therefore for this research
work the world is assumed to be a two and a half di-
mension one (2,5D). In this scenario, a map is rep-
resented considering a bird’s eye view and assuming
that the height of the objects has no maximum value.
In these conceptions it is simpler to obtain the set
of potentially visible objects from a certain point us-
ing a portal culling algorithms; that consider walls as
complete occluders and assigns a vision probability to
each region in the map (Pires, 2001). Such is achieved
by dividing the indoor space into separate sectors and
then portals which represent the breaches between the
sectors. By drawing cones that connect the observer
point and both extremities of a given portal, one can
obtain the areas where all the objects are potentially
visible. Some caution is required when performing
such an operation, since the lines that represent the vi-
sion cone cannot intersect the ones representing zone
divisions. It is also relevant to state that one must as-
sume the observer’s vision direction is the center of a
given ”vision cone”.
3 PROJECT DESCRIPTION
The project description is divided of three distinct
subsections. In the first denominated as Project Ar-
chitecture a description of the system’s architecture
is depicted. In the Real Time Tracking Visualization
and Concentration Maps subsection the principles of
the system features is explained and in the last sec-
tion, Client Vision Module the used vision algorithm
is exposed.
3.1 Project Architecture
This research work proposes a decentralized architec-
ture and prototype tool following that same principle
as detailed in Figure 1.
Figure 1: System’s Architecture.
In the first stage, using a location system that might
be instanciated in a RFID or Wi-Fi based solution,
covering an area of for instance a large open space or
building, with a maximum error of one meter, move-
ment data is collected regarding the tagged elements
present in the given specific floor. In order to gather
the location data one agent has been developed to col-
lect all the positions. This action has a given periodic-
ity and is dependent on the location engine. Typically,
this collected data is guaranteed to be obtained every
second at most for every single tag, although this fig-
ure might be decreased depending on the number of
simultaneously trackable objects.
After the collection process the agent sends the in-
formation to the server application. Prior to this ac-
tion, the agent executes a simple, yet efficient, data
validation that is based on the map of the structure
that is sent by the server before the collector agent
boot process.
The server application, before being able to re-
ceive any position data, must load the floor map that
will contain the trackable objects, from a XML file.
The maps are modeled as 2,5 D worlds and include
several structures that can be easily adapted to many
types of spaces. Before using it, the server validates
the map against a XML Schema.
For instance, a XML file representing a traditional
retail shop includes entry areas and an exit one that
in this case is designated as a payment area. There
are also walls that have infinitive height positioned
around the map. In the supermarket example these
last are named as shelves as can be seen on Figure
2. The proposed XML structure allows specifying a
color and a designation for each half of the wall. The
half is determined by the largest dimension, and in
case of a square it is assumed to be vertically alligned
as can be seen on Figure 2.
For each set of received coordinates, the server
stores them in a database for universality sake. The
BUSINESS INTELLIGENCE THROUGH REAL-TIME TRACKING - Using a Location System Towards Behaviour
Pattern Extraction
53
Figure 2: Drawn Map with Corresponding XML File Struc-
ture.
system executes it in a completely completely agnos-
tic way concerning the database provider. At this
point it is also relevant to state that the server is a
multi-threaded application having a thread per tag.
Each thread writes into a reserved memory location
the tag’s current position. The previous position is
not overwritten; instead it is stored in the database.
Each thread includes some recovery proceedings such
as deleting all the records of a specific tag in a given
time frame if it stops transmitting its location for a
long period of time regarding the context at hand.
As each thread executes, the server GUI is able to
display online data. This data represents where the
tags are located on the map. Other types of views in-
volve processing the data in real-time, using computer
memory and/or by consulting the database registers
for further in depth reports (section 3.2).
3.2 Real-Time Tracking Visualization
and Concentration Maps
The Server GUI includes several different views of
both the online and historical data. In all these views
there is a visual representation of the map. In order
to draw a map the server requires to systematically
perform a scale transformation involving real world
coordinates and pixel coordinates. This kind of trans-
formation must be dynamic because, in any instance,
the GUI can be resized. When the view involves po-
sitioning tags on the map this transformation is also
applied to their centre positioning.
The simplest view allows representing online the
tags in their actual positions. Other views are ob-
tained through the server’s knowledgment extraction
features. The Zone Matrix consists in determining
in real-time which are the most and least populated
zones, denominated as hot and cold zones.
It is also possible to consult this data on a wider
time frame considering the same space with or with-
out the same layout. This last feature requires
database access. The zones are automatically ob-
tained by dividing the space into a grid with flexi-
ble dynamic resolution. This dynamic division allows
both a more in depth study of the hot and cold zones
and also a less detailed one in order to study, for ex-
ample, the spaces quadrants occupation.
Several other results are obtainable by accessing
memory-based data structures for limited time frame
analysis and by querying the database in similar
modes. It is possible to obtain historical client paths,
the shop areas walls, shelves or objects that were
more observed by the clients.
In order to reproduce historical client path recog-
nition all the clients coordinate are stored into the
database with a timestamp that is related to a given
map. Figure 3 exposes the paths taken by several
clients in a given time frame.
Figure 3: Paths taken by the Elements.
3.3 Client Vision Module
The client vision module uses a simplified version of
the occlusion portal culling detection algorithm. This
simplified version discards all zones that have a low
probability of being seen and also disregards orienta-
tion and assumes that the observer can see simultane-
ously north, south, east and west directions. For each
ICE-B 2008 - International Conference on e-Business
54
of these directions the observer throws a vision cone
having its center coincident with the direction at hand.
The first intersected walls are considered to be vis-
ible, and the others invisible. The Figure 4 summa-
rizes with descriptive colors which were the shelves
that were most observed by the moving targets.
Figure 4: Visible Walls with Historical Data.
4 RESULTS
In this section, the project’s results will be de-
picted taking into account three generic levels: sys-
tem’s main features; simulation statistics extraction
and global aplication; and architecture stability and
feasability.
For simulation purposes, there had been consid-
ered two standard computer configurations: a high-
end machine with 4GB of RAM, an Intel Core 2 Duo
E820 CPU and a SATA II 320 GB 16 MB cache hard
disc denominated as configuration A, and a low-end
configuration, denominated as B, equiped with a 2 GB
of RAM, an Intel Pentium D 3.00GHz and a SATA II
250 GB 8MB cache hard disc. Both systems were
equiped with Windows Vista Ultimate and the simu-
lations were performed with similar workload condi-
tions. In Figure 5 the experiment’s results are fully
depicted as for both hardware configurations four dif-
ferent scenarios were simulated. For each one it was
recorded the CPU time needed to perform the most
demanding task real-time dynamic grid concentra-
tion levels with memory-based historical data and
the presence of absence of image flicker, with a dif-
ferent number of tracked items ranging from a single
one to one thousand.
As previous note, one shall point that for single
item tracking, the measured CPU time for both con-
figurations is not available as the benchmarking tool
reported zero seconds. The results showed that for ten
items, the differences between low-end and high-end
computers is absolutely negligible. For one hundred
items, configuration B needs twice the time of con-
figuration A but real-time visualization is not jeop-
erdized in anyway. In both cases, for this scale there
was not registered any flicker effect and the process
time was compatible with a real-time system. Only if
the scale is pushed to one thousand, configuration B
takes three point seven seconds to compute and even
configuration A takes two point three seconds. These
figures show that for this kind of scale it is needed
a high-end computer system even if one consider
the traditional consumer market products and hard
real-time requirements are not met but one might still
assume near real-time features that are perfectly ad-
equated for this kind of management/monitoring sys-
tems.
ConfigurationA ConfigurationB
CPUTime(ms) N/A N/A
Flicker N N
CPUTime(ms) 15 23
Flicker N N
CPUTime(ms) 125 215
Flicker N N
CPUTime(ms) 2300 3700
Flicker Y Y
NumberofItems
1
10
100
1000
Figure 5: Simulation Performance Benchmarking.
Regarding the first aspect, all the enunciated pre-
dicted functionalities, thecnically described in the
previuos section, were successfully implemented and
fully tested. As illustrated in Figure 6, it is possi-
ble to visualise in real-time the location of up to one
thousand items overlapped with the shop floor layout.
This number of items can be increased but it is de-
pendent of the external location system’s features. In
the same illustration, it is visible the tool’s flexibility
in what concerns to layout management and design
as all shop floor static structures are fully defined and
described through a simple, yet flexible XML config-
uration file. With this approach, it is possible to model
heterogeneous environments and, therefore, apply the
proposed system to several domains.
In spite of the importance of the mentioned func-
tionalities, the greated added value resides in the
knowledge extraction extendend features. Having in
consideration the online item location gathering, the
system is able of real-time item path reconstruction
and visualization operating both in memmory-based
or database access, depending on data dimension.
Conducted experiments showed that real-time mem-
mory access is feasible using a low-end computer
with 2 GB of RAM – for tracking one hundred items
for a period of an hour at a medium pace.
BUSINESS INTELLIGENCE THROUGH REAL-TIME TRACKING - Using a Location System Towards Behaviour
Pattern Extraction
55
R.'"-T
.....
srwFW[z....,M..:r~ICle<-N~~L
_
" -
----
Figure 6: Real-Time Item Location.
As illustrated in Figure 7 the location data is used to
extract more significant information about item dis-
persion/concentration both in the present and also
considering historical data. It was used a gradient
scale where concentration levels vary through the
RGB scale where red means high levels of concen-
tration and blue low levels.
Figure 7: Historical Concentration Matrix.
One significant functionality of this model is the pos-
sibility to perform calculations based on completely
flexible and dynamic projection grid. This option
proved to be efficient on online data processing for
a significant number of tracked items - approximately
one hundred - without database access by using a tem-
poral location matrix. This feature enables a full de-
tailed concentration analysis in real time when recur-
ring to a high definition grid that divides the layout in
small areas; and enables swift high big-picture studies
when using a less tight net. This capability is appro-
priately described in Figure 8.
Finally, considering the features results description,
one shall paint the relevance of the vision module.
Figure 8: Dynamic Concentration Grid Example.
This application requirement performs the emulation
of an omnidirectional vision of each trackable item.
The described algorithm is able of identifying the
visible objects, having into consideration both single
instantaneous data and historical information, previ-
ously collected and stored in direct memmory ac-
cess or in a database. In the conducted experiments,
this system’s module also showed high levels of effi-
ciency and correctness; much similar to the ones al-
ready described in the above paragraph.
Regarding the system’s global architecture def-
inition and implementation, the undertaken simula-
tions demonstrated its adaptative capability through
its flexibility in what concerns to both the database
provider and, perhaps more important, to the loca-
tion system tecnology. These caracteristics greatly
enhance the whole system’s applicability in several
scenarios. Still in this domain, the distribuited sys-
tem’s design enables the usage of low-end computers.
Therefore it constitutes an incentive to client’s IT in-
frastructure reusage while minimizing the solution’s
economical impact. Simultaneously, this approach
enables greater site manager’s empowerment through
real-time information access to all system’s features
visualization. These actions can be triggered for both
partial and global organization providing more and
deeper analysis points of view.
5 CONCLUSIONS
Considering the project’s simulation environment and
the achieved results, one shall state that, although the
location engine had been implemented in order to re-
alistically simulate traceable items, all concept has
been demonstrated. The developed prototype proved
to be efficient and effective in large scale distributed
data gathering and real-time item location visualiza-
tion.
Taking into account the system’s architecture, it
was verified the concern in allowing multi-store man-
agement with both distribuited modules and central
integration concerns that enable consistent and online
knowledge extraction and visualization. Having in
mind the different application modules, one ought to
refer that the integration with the XML-based layout
manager proved to be extremely flexible to accom-
ICE-B 2008 - International Conference on e-Business
56
modate distinct real scenarios and yet realizable and
realistic.
Considering the most noble project’s slice, one
ought refer to the previously enunciated knowledge
extraction features. Having as support the results de-
picted in the last section, it is secure to state that the
system is able to produce real-time shop-floor visual-
ization maps with intelligible data on online item lo-
calization; individual item complete path routes; on-
line and historical population density rates and path
routes concentration; and also item-vision enabled
concentration maps as emulation for item omnidi-
rectional vision, yet considering occlusions. All of
these features are allowed for graphical user interface
through different grid dimensions for distinct analy-
sis granularity. Bearing in mind the project as whole,
one shall state that the developed knowledge extrac-
tion platform with online and diversified visualization
tools constitutes a solid ground for online item track-
ing and heterogeneous space management with dis-
tributed capabilities. One final major advantage of the
proposed system resides in the total independence re-
garding the external position engine both in terms of
suppliers and even more important in terms of base
technology.
In spite of the enunciated project’s accomplish-
ment, an even by being in prototype stage yet re-
liable and fully functional there are several future
work areas that are able to greatly enhance the sys-
tem’s appliance and success. From these, the most
relevant ones are believed to be centered in eccen-
tric shop-floor layouts both in terms of shape and
multi-level buildings; complete path routes analysis
enabling common node fusion for global paths proba-
bilistic construction; flexible and dynamic report def-
inition tool with configurable alarm triggering; and
perhaps the most interesting would be the character-
ization of ’what-if scenarios with simulated traffic
based on real historical data. Considering both the
project current achievements and the depicted future
work areas, one might identify the most desirable im-
plementation domains. Although there are not limited
to, the proposed system might be useful for traditional
retail environment for shopping cart tracking; tracing
clients through a commercial area such as shopping
centers; enabling item tracking and route analysis in
an hospital; producing activity reports and analysis in
controlled areas such as penal complexes, mental in-
stitutions or closed educational organizations.
ACKNOWLEDGEMENTS
The authors would like to thank Professor Eugnio
Oliveira for inspiration and the contribution regard-
ing behavior pattern recognition and artificial intel-
ligence guidelines, Professor Augusto Sousa for en-
lightenment in the portal cells algorithms.
REFERENCES
Chao C., Yang J., J. W. (2007). Determining technology
trends and forecasts of rfid by a historical review and
bibliometric analysis from 1991 to 2005. In Techno-
vation 27, Issue 5, pages 268–279. Elvisier Ltd.
Choustova, O. (2007). Price-dynamics of shares and
bohmian mechanics: Deterministic or stochastic
model? In Foundations of Probability and Physics,
pages 274–282.
Grudzinski, J. (2007). Real-time clouds and weather sim-
ulation in computer games. In Learning with Games
2007.
Jappinen P., P. J. (2007). Preference-aware ubiquitous ad-
vertisement screen. In eCommerce 2007 Proceedings,
pages 99–105. IADIS Press.
Krotosky J., T. M. (2007). A comparison of color and in-
frared stereo approaches to pedestrian detection. In
Intelligent Vehicles Symposium, pages 1226–1231.
Luce, R. (1999). Where is mathematical modeling in psy-
chology headed? In Theory and Psychology, pages
723–737.
Mingkhwan, A. (2006). Wi-fi tracker: An organization wi-
fi tracking system. In IEEE CCECE/CCGEI, Ottawa,,
pages 1353–1356.
Moura J., Coelho A., L. P. e. a. (2004). Readings in Infor-
mation Society. Universidade Fernando Pessoa.
Park H., Lee S., C. W. (2006). Obstacle detection and fea-
ture extraction using 2.5d range sensor system. In
SICE-ICA SE, 2006. International Joint Conference,
pages 2000–2004.
Pires, P. (2001). Dynamic algorithm binding for virtual
walkthroughs. Master’s thesis, Technical University
of Lisbon.
Raizer, V. (2003). Validation of two-dimensional ocean
microwave signatures. In Geoscience and Remote
Sensing Symposium, 2003. IGARSS ’03., pages 2694–
2696.
Vazquez, P. (2007). Real time falling leaves. In Second In-
ternacional Conference on Computer Graphics The-
ory and Applications, pages 244–251.
BUSINESS INTELLIGENCE THROUGH REAL-TIME TRACKING - Using a Location System Towards Behaviour
Pattern Extraction
57