No More Hiding! WALDO: Easily Locating with a Wi-Fi Opportunistic

Approach

Bol

ıvar Silva

, Jo

ao Carlos Lima

, Celio Trois

, William Pereira

and Cristiano da Rocha

Universidade Federal de Santa Maria, Av. Roraima n

1000, Santa Maria, Rio Grande do Sul, Brazil

Penguin Formula, Lisbon, Portugal

Keywords:

Indoor Positioning, Opportunistic Wi-Fi Sensing, Object Detection, Pervasive Computing, Location based

Service.

Abstract:

The popularization of mobile devices and increasing the number of sensors and embedded resources boosted

a large amount of research in the area of context-aware. Among the most relevant contextual information

is the location. In outdoor environments, the GPS technology is already widespread and used. However, the

people tend to spend most of their time indoors, such as universities, hospitals, malls, and supermarkets, where

the GPS location is compromised. Several approaches using mainly radio frequency technologies have been

proposed to solve the problem of indoor location. So far, no widely accepted solution solves the problem of

location indoors. In this way, this work to use an opportunistic approach, making use of the Wi-Fi infras-

tructure available in the environment, to provide the location of mobile stations. Based on this objective, the

WALDO architecture was developed that unites the characteristics of different approaches of the works that

have been produced in recent years, taking into account the techniques that present better results at each stage,

in conjunction with a zone-based approach and ranking, on phase online of the ﬁngerprint technique, which

makes ignoring RSS readings that present noises.

1 INTRODUCTION

Nowadays, it is quite common to ﬁnd mobile devices

with several sensors, which collect information from

the environment in which the device is inserted, aim-

ing to offer greater convenience and improve the user

experience, presenting information and useful ser-

vices, such as showing the user the nearby restaurants

or suggest the best route to reach the desired destina-

tion. In this sense, location information is a funda-

mental parameter. A large part of these technologies

are targeted for indoor use, such as visitor orientation

at malls (Oosterlinck et al., 2017), airports (Ahmed

et al., 2016), real-time monitoring of patient location

in hospitals (Kanan and Elhassan, 2016), identiﬁca-

tion of crowd concentration (Fukuzaki et al., 2015),

among others.

Although GPS (Global Positioning System) tech-

nology is quite widespread for outdoor environments

(Yang et al., 2013), due to signal interference, it often

does not provide satisfactory accuracy indoors (Paul

and Wan, 2009) (Rai et al., 2012). To mitigate this

demand is necessary to use a source of information

more suitable for this type of environment. In this

sense, several technologies have been used as a source

of information to perform the mapping of the environ-

ment. Mostly technologies that emit radio frequency

signals, such as Wi-Fi (Wireless Fidelity), Bluetooth

(Oosterlinck et al., 2017), RFID (Radio Frequency

Identiﬁcation) (Ahmed et al., 2016) and NFC (Near

Field Communication). Due to the popularization and

enabling opportunistic approaches, the Wi-Fi technol-

ogy in IPS (Indoor Positioning System) implementa-

tions are among the most widely used technologies

(Rai et al., 2012). Among the approaches used are the

ﬁngerprint technique and estimates using geometric

calculations (Yang and Shao, 2015). The ﬁngerprint-

ing technique depends on a prior mapping of the en-

vironment, whereas geometric estimates use the dis-

tances between the object to be located and three or

more landmarks with a known location.

The popularization of Wi-Fi networks has made

more and more people stay connected through their

mobile devices (Ge and Qu, 2016), opening up oppor-

tunities for other ways to exploit this technology. In

recent years, several works have used Wi-Fi technolo-

gies to obtain the location of devices in environment

indoors. In addition to allowing the use of existing

Silva, B., Lima, J., Trois, C., Pereira, W. and da Rocha, C.

No More Hiding! WALDO: Easily Locating with a Wi-Fi Opportunistic Approach.

DOI: 10.5220/0007677505710578

In Proceedings of the 21st International Conference on Enterprise Information Systems (ICEIS 2019), pages 571-578

ISBN: 978-989-758-372-8

571

and widespread infrastructure, another motivator lies

in the energy savings of mobile devices. The use of

other sensors or interfaces, such as the accelerometer,

gyroscope, Bluetooth and magnetometer, for the sole

purpose of providing device location, result in extra

energy expenses (Lane et al., 2013).

For systems that provide localization in indoor en-

vironments, there is a relationship between accuracy

in location, complexity, cost, and scalability estimates

for implementation in real environments (Mainetti

et al., 2014). Many papers in the literature(Yang and

Shao, 2015)(Lymberopoulos et al., 2015)(Chen et al.,

2014)(Rai et al., 2012)(Kannan et al., 2013) discuss

these points separately, or even address just a few of

these. This paper presents a low-cost, scalable, indoor

location approach that uses the Wi-Fi infrastructure

available to obtain the location of commercial smart-

phones.

Among the main contributions presented in the

present work is the use of an opportunistic approach.

Another contribution is the use of a zone scheme

and ranking in the location phase of the ﬁngerprint

technique, which ignores some RSS (Received Signal

Strength) readings that present noise. For this to be

possible, during the location phase, the mapped area

is zoned, which has a set of sampling points with their

respective ﬁngerprint signatures. Areas with the most

extended distances between online measurement and

sampling points are calculated and excluded. Since

only the two zones with the highest probability of the

device are left, a ranking scheme is performed, to pri-

oritize the readings with less noise.

The rest of the paper is organized as follows: Sec-

tion 2 presents the motivation and related works in the

literature. Section 3 explains WALDO architecture

and the methodology used, explaining the implemen-

tation of the modules that make up the architecture.

Section 4 describes the tests and results. Finally, con-

clusion and future work are given in Section 5.

2 MOTIVATION AND RELATED

WORK

The use of the ﬁngerprint technique, coupled with

the use of mobile devices (such as smartphones and

tablets) with Wi-Fi technology, has presented promis-

ing results in the development of indoors positioning

systems. In this approach, the smartphone to be found

must have an active Wi-Fi interface. Among the chal-

lenges encountered in this approach are the inherent

limitations of mobile devices, such as low process-

ing power, limited memory, battery usage, and the di-

versity of distinct hardware found in mobile devices

(Kannan et al., 2013). Also, radio frequency sig-

nals suffer much interference from obstacles present

in closed environments, making the use of this pa-

rameter for the unambiguous characterization of the

environment, become a great challenge. In order to

better characterize each section of the mapped envi-

ronment, Chen et al. (Chen et al., 2014) proposes an

algorithm that relates not only the power of the re-

ceived signal (RSS) but also the proximity order of

the APs according to their power. The authors hope

that in this way, the ﬁngerprint of each room in the en-

vironment will be better deﬁned, making it easier to

locate even when the devices have different hardware

characteristics. Using a different approach, Rai et al.

(Rai et al., 2012) make use of some of the sensors

present in smartphones (accelerometer, compass, gy-

roscope, etc) together with data obtained from Wi-Fi

and a user-informed map containing the characteris-

tics of the environment, and barriers (walls and other

obstacles).

Another challenge for the use of ﬁngerprint-based

approaches lies in the mapping of the environment,

which demands time and work. Since the ﬁnger-

print based on a particular feature of the environment,

such as radio frequency, magnetic ﬁeld or background

sound is used, this information must be captured and

stored (training stage). Therefore, the time demanded

the ofﬂine step is related to the size of the area to be

mapped, as well as the precision that one wishes to

obtain. In order to reduce this time, without com-

promising accuracy, a series of tests with short and

long duration readings were carried out in the work

of Yang et al., and it was veriﬁed that the value that

most frequently appears in the measurement (a few

minutes) is the same as long-term measurements (a

few hours). Thus, there is no need for large amounts

of capture time per site. Also, the mean values of

the readings of each position were compared with the

most frequent value, and a signiﬁcant difference be-

tween the two measurements was found.

The WALDO architecture presents an approach

that detects RSS noise-free readings during the on-

line stage, eliminating them from the localization es-

timation process and thus presenting better estimates.

Also, are used techniques that have already been ad-

dressed and tested in the literature separately.

3 WALDO ARCHITECTURE

WALDO uses the ﬁngerprint technique to perform

mapping and location devices, which in turn uses

the Received Signal Strength (RSS) data of different

WLANs whose signal covers the area to be mapped.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

572

The ﬁngerprinting technique is divided mainly into

two distinct stages: training phase (ofﬂine) and loca-

tion phase (online). In general, this technique consists

of comparing the set of RSS readings obtained during

the training phase and the location phase RSS read-

ings set.

Figure 1: WALDO Architecture.

Figure 1 gives an overview of the WALDO archi-

tecture. As can be seen, the training phase and the

location phase have subdivisions which will be ex-

plained in the following sections.

3.1 Methodology

The expected result of WALDO consists of determin-

ing the location of a smartphone, within a previously

mapped area, using information from a set of nearby

Wi-Fi networks. Figure 2 presents an information

ﬂow that describes the methods and procedures per-

formed in the process. As can be observed, analyzing

from the left, during the training phase the area and

it’s RPs (1a) are deﬁned. In the next step (1b) the ra-

dio map is obtained, which is composed of the read-

ings performed in each of the RPs. The radio map in-

formation is used to generate FDB

and FDB

, used

for comparison purposes in the tracing process. The

location phase starts (2a) with the reading performed

by the smartphone, in the place where it should be lo-

cated. This step consists of calculating the sum of the

RP distances of each zone and the values read at the

time of the estimate of the location. The two zones

that have the sums of smaller distances will be used

in the next step. In the next step (2b), two classi-

ﬁcations are generated containing the RP list of the

two remaining zones. This RP list considers only RPs

that establish neighborhoods with the leading RP (the

nearest neighbors) of the ranking. Finally, the zone

with the highest probability of being the correct one

is identiﬁed and the Naive Bayes algorithm (NB) is

applied at the remaining points.

3.2 Collection Module

The process of the data capture consists of storing in

the database the RSS readings, associated with the

respective identiﬁers of the APs, performed by the

smartphone in each RP of the mapped area. To start

the capture is necessary to have the information about

the dimensions of the area to be mapped, as well as

the location of the RPs used.

During the data collection, were considered four

main characteristics. They are the number of readings

per RP, reading time, device orientation at the time of

reading and the number of RPs. The amount of RPs

and distance of spacing used in the scenario, take into

account information observed in the literature, which

indicates average distances of errors over one meter,

in an approach using Wi-Fi ﬁngerprint. Also, the

variation of RSS readings in close RPs, presents sub-

tle differences, making the distinction between very

close RPs difﬁcult. Once decided the number of RPs,

it is necessary to deﬁne the number of readings per-

formed by RP as well as the time of measurements

demanded. In the present work, the approach used

where the readings are performed pointing the de-

vice to four different directions, using the walls of the

room as references for each direction. In this case,

at least four different readings are taken, pointing in

each direction. Based on works cited in the literature,

such as the (Yang et al., 2013), it is known that, al-

though the oscillations demand a certain amount of

measurements per point, the values that appear with

more frequencies in short duration measurements will

be the same values that will appear more frequently in

long-duration measurements. Therefore, making the

most frequently measured value the best choice for

each RP ﬁngerprint.

3.3 Filter Module

The variability in RSS measurement performed by a

Wi-Fi mobile device is a consequence of radio fre-

quency communication characteristics and interfer-

ence caused by indoor obstacles. Since RSS and the

identiﬁcation of APs (MAC addresses and locations

in the mapped environment) are the main parame-

ters used in the estimation of positioning, the selec-

tion of the data to be labeled and used in the calcula-

tions will directly impact the quality of the position-

ing estimates. In this context, the ﬁlter module selects

the data that will be used as parameters in the loca-

tion phase. Firstly, we tried to disregard data that had

No More Hiding! WALDO: Easily Locating with a Wi-Fi Opportunistic Approach

573

Figure 2: Information Flux.

characteristics that were inappropriate for the applica-

tion. In this sense, were identiﬁed (using interquartile

range) and excluded the outliers, generated by noise

and other phenomena such as multipath. In addition

to the outliers, data that did not appear in all RPs

was also removed. This is because the AP is distant

physically or has the signal obstructed by an obsta-

cle. In this way, some areas of the mapped environ-

ment are not covered by the signal, while others, get

ﬁckle readings with low power. The second question

is related to the distance between the mapped loca-

tion and the device. If the long distance between the

AP and the mobile station has caused poor signal cov-

erage, and this signal is not available in all regions of

the mapped environment, the chances of obstacles oc-

curring between the transmitter and the receiver will

increase.

In approaches that use probabilistic calculations,

the larger the database, the better the results tend to

be. On the other hand, when the comparison is per-

formed with distance calculations (such as Euclidean

distance), it is more appropriate to use a less extensive

FDB, containing the measures that best characterize

each RP. In the present work, the two approaches were

used, in different stages. In this sense, two FDBs were

generated, where one presents all the readings (after

ﬁltering) and the other presents only the RSS read-

ings that best characterize the RP (most constant RSS

values).

3.4 Fingerprint Generator

The ﬁngerprint generator module is responsible for

obtaining the two (FDB

and FDB

) database corre-

sponding to the training phase. The ﬁrst ﬁngerprint

database (FDB

) is formed by all remaining data af-

ter the deletion of the outliers. The variation of the

signal, caused by the interference of the environment,

causes the intensity of the received signal of the same

AP in the same point is variable. Thus, it is known

that the higher the signal strength, the less interfer-

ence will be. In this sense, in order to prioritize the

data that suffered less interference, the records were

organized in order (decreasing) of signal power. The

reading that presents the highest RSS value for an AP

and a speciﬁc point is put ﬁrst in the sequence of the

records. This organization is performed for each AP

of each point. Also, APs are also organized in de-

scending order of power. In addition to the organi-

zation and arrangement of data in an orderly manner,

the FDB

will be used to perform the comparison ap-

proach by probabilistic means.

FDB

is used to create FDB

. In the FDB

, each

RP will only have one RSS value per AP, that is,

only the RSS that most often appeared in each AP. In

this sense, the number of values present in the FDB

will depend on the amount of RPs, and the amount

of APs used. The two factors are self-explanatory.

The higher the number of RPs and APs, the greater

the data set associated with them. During the creation

stage of the FDB, were captured the data of all the

upcoming APs. Several APs can be observed in the

readings at each point, however, it is worth noting that

in the location phase some APs that were found dur-

ing the creation of FDB will not appear, as well as

some APs observed in the training phase will not ap-

pear in the location phase. Only values observed in

the two steps will be used.

3.5 Zone Module

The use of zones has the objective of improving the

location estimates, prioritizing the values measured in

the location phase with a lower incidence of noise.

For this, the mapped area is divided into a certain

amount of zones, as can be observed in Figure 3.

The division of the area into zones takes into ac-

count the characteristics of the mapped area as the

dimensions of the area and the total amount of RPs

(T RP). The scenario presented in Figure 3 was di-

vided into 5 zones (T z), identiﬁed by upper case let-

ters. The equation 1 give the number of RPs for

each zone. The division begins with the choice of

ﬁve strategic points on the map. From these points

each zone is segmented. In the example, the area of

each zone (Az) corresponds to 24 RPs plus the chosen

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

574

Figure 3: The total area of the computer lab divided into

ﬁve zones.

point, totaling 25 RPs per zone. In this scenario, the

four extreme RPs (1, 9, 73 and 81) besides the central

RP (41) were chosen as starting points of each zone.

Az = b(T RP/T z)c + [b(T RP/T z)/2c] (1)

3.6 Location Phase

Once the environment is mapped, the next step is the

location of the smartphone. The location of the de-

vice is performed by comparing the data obtained at

the time of screening (online step) and the present

data from the system database. The process of lo-

cation estimation starts when it is in the mapped en-

vironment, with the active Wi-Fi interface. Given the

characteristics of the Wi-Fi technology, described by

the IEEE 802.11 standard (Potort

ı et al., 2016)(Group

et al., 1999), in order to obtain the device data (MAC

address and RSS), it is not necessary that it be con-

nected to the Wi-Fi network. This information is

broadcast on the mobile device in order to ﬁnd the

available APs within reach. This information can be

obtained at both ends of the communication, that is,

in both the mobile station and the AP. In the present

work, the validation tests and implementation of ar-

chitecture were created with the help of two applica-

tions. These are designed to obtain the data from all

APs close and send them to a server. Thus, in the im-

plementation, the RSS data obtained from the mobile

station were used. Therefore, there are no change re-

quirements in the existing network infrastructure.

3.6.1 k-NN Module

Upon receiving the data, the manager application,

running on a remote server, performs the necessary

procedures to obtain the approximate location of the

mobile station. First, it checks which APs read from

the database FDB

, and discarded the readings from

unknown APs. The ﬁrst comparison uses the Eu-

clidean distance equation, where are comparing the

location parameters with each of the RPs. The closer

to zero is the value resulting from the comparisons,

the more similar the values are.

From the Euclidean distance calculation, it is pos-

sible to verify the locations with less probability of

the smartphone being. In this sense, an approach

based on the kNN (k-Nearest Neighbors) algorithm,

which uses a measure of similarity (or dissimilar-

ity) between neighboring points, was used to classify

datasets (Ge and Qu, 2016). In this approach, the

measure of similarity used is the Euclidean distance,

and the neighbors are the data of the RPs present in

the FDB

, whose Euclidean distance is the smallest.

When obtaining the similarity value for each RP, they

are added between each zone separately, in order to

obtain the sum of the distances of each zone. By iden-

tifying the two zones with the smallest sums of Eu-

clidean distances, we discard the zones whose calcu-

lations have indicated to be the least likely (farthest)

of the smartphone to be located.

With two zones remaining, the next action consists

of identifying and eliminating the zone with the least

probability of the device being, so that only one re-

mains to be analyzed. The next action taken takes into

account RSS variability characteristics. It is known

that within the set of parameters obtained, some of

them reﬂect the characteristics of the environment

where the device is, while others, due to interference

suffered at the time of reading, are with parameters

that do not characterize that point. From this infor-

mation, it is observed the need to identify which data

was read with interference and which suffered less

variation. In this sense, a ranking of results is gen-

erated for each remaining zone, in ascending order,

from the most similar to the least similar. From this

step, the comparison between the zones is no longer

made by the sum of all the results, and starts to be

done by comparing the positions of each ranking, pri-

oritizing the most similar values with those present in

the FDB

, and keeping the least similar (which have

undergone signiﬁcant variations) for tiebreaking ends.

The table 1 presents an example of ranking, with

the most probable points of each zone. To reach these

points are taken into account two Euclidean distances.

The ﬁrst is the distance between RSS values calcu-

lated in the ﬁrst location phase. The second is the

physical distance between the RPs. This distance does

not take into account radio frequency information, but

rather the distance of all points of the zone, concern-

ing the leader point of the ranking. In other words,

it seeks to ﬁnd the neighboring points of the leader

RP. With this information, the ﬁrst 13 points closest

to the leader of each zone are selected, which in the

test scenario correspond to half the total number of

No More Hiding! WALDO: Easily Locating with a Wi-Fi Opportunistic Approach

575

points in the zone. Of the 13 selected points, 12 are

located within the zone. Of the remaining 12, only 6

are points near the ﬁrst point of the ranking. These 6

points are those presented by the table 1.

Table 1: Ranking table for zones B and D.

AP Zona B / Zona D PA B/D Dist

ancia B/D

fe:d1:b8 / be:b4:ab 24 / 41 0 / 0

f7:8f:26 / f7:8f:26 27 / 20 0 / 1

8e:d0:a0 / be:b4:ab 33 / 31 0 / 2

be:b4:ab / 87:22:01 35 / 51 1 / 1

87:22:01 / fe:d1:b8 24 / 49 1 / 2

be:b4:06 / - 21 / - 2 / -

The comparison between the remaining zones, to

determine which zone is most likely, follows the fol-

lowing conditions:

• First (most common) case: the values of the ta-

ble 1, the distance from the leading point of zone

B will be compared with the distance of the lead-

ing point of zone D. When verifying that they are

equal, the second point of zone B will be com-

pared with the second point of zone D. At this

point, it is veriﬁed that the distance value of zone

B is smaller. Then the set of points in Zone B is

elected the most probable set;

• Second: when one of the two zones compared

presents all values of distance equal to zero, the

zone will be chosen as most probable;

• Third, because the two rankings have different

sizes, in some cases, during the comparison the

values of one of the rankings will be exhausted,

while the other will still have values to test. In

these situations, where the tie in the compar-

isons persisted, the zone identiﬁed with the high-

est probability during the comparison between the

ﬁve possible, will be the zone chosen.

With the set of RPs reduced, it is necessary to

identify which of the remaining RPs is most likely to

be the location of the device. For this applied Naive

Bayes classiﬁer algorithm.

3.6.2 NB Module

When using Naive Bayes classiﬁer algorithm, the

odds obtained through the measures for each of the

APs are independent of each other since the RSS mea-

sures read by the smartphone about each AP are also

independent. In the current scenario, with a reduced

set of RPs to be analyzed, the use of Naive Bayes be-

comes an interesting approach. In this sense, the data

stored in the training phase corresponds to the training

dataset, while the data received from the client station,

in the location phase, is the data to be classiﬁed. In or-

der to NB classiﬁer present better results, in this step

is to use the FDB

, which has a more extensive data

set.

Continuing the example of the smartphone loca-

tion, six points remain. Starting with the RP number

15, the parameters to be passed to the Naive Bayes

are: the identiﬁer of the RP, one of the APs sent by the

mobile station and its respective RSS measure. The

result returned is the probability that the AP and RSS

belong to the set of values of RP 15. This process

is repeated in RP 15 as there are distinct APs pro-

vided by online reading. If the mobile station sent a

set of ﬁve APs in the reading, the ﬁve APs and RSS

would be compared with the RP 15. The same pro-

cess carried out in RP 15 is done in the other RPs.

To conclude, the sets of probabilities are summed and

veriﬁed which RPs have the most signiﬁcant sum.

4 TESTING AND RESULTS

Among the methods used to evaluate IPSs are the

precision, accuracy, scalability, robustness, cost and

complexity tests (Hossain and Soh, 2015). As noted

in the literature, short distance technologies usually

provide better accuracy than others because they suf-

fer less interference and ensure device position more

accurately. Thus, the tests used for the validation of a

tool, are dependent on the approach that was used. A

Wi-Fi-based IPS does not offer as good an accuracy

as an IPS that uses NFC TAGs, for example. On the

other hand, the need to install TAGs, besides requiring

the use of mobile devices compatible with technology,

make the NFC-based approach more expensive and

less scalable. In that sense, while a technique gains in

accuracy, it loses in scalability and cost.

Considering the characteristics of WALDO, the

tests carried out have the objective of evaluating the

precision and accuracy of the location estimates. The

precision is determined by some method of distance

measurement, such as Euclidean distance, accuracy is

obtained based on a set of precision test results. Due

to the amount of Wi-Fi available for use in most urban

areas, the challenge of scalability, as well as for the

complexity of the implementation, is related to map-

ping the area, which demands time. Because it is an

opportunistic approach that uses the existing infras-

tructure, the cost of implementation is reduced.

To validate the architecture, in addition to the

dataset used in the architecture development, we used

the KIOS Wi-Fi RSS dataset, more speciﬁcally the

database called Desire, made available by (Laoudias

et al., 2013). KIOS consists of a collection of RSS

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

576

samples, a mobile device with Wi-Fi support. Data

were collected in a typical 560 square meter ofﬁce

space consisting of a conference room, laboratories,

and corridors. Within this area, data were selected at

105 different sampling points.

Using the dataset obtained in the computer lab, a

test was performed in each RP, aiming to obtain a per-

ception of the results throughout the room area. In this

sense, 81 location estimates were made. The readings

were made from a smartphone and sent to a server

with the implementation of the WALDO architecture,

which performed the calculations.

Figure 4: Location errors using the dataset of the computer

lab.

In the graph of Figure 4 the results of steps A, B

and C are presented respectively. Of all the points

tested, the location estimate with the greatest discrep-

ancy between the estimated location and the actual lo-

cation presented just under 8 meters away. Despite

this, most of the results shown in C (using WALDO),

present results with errors of little more than 2 me-

ters. In the histogram presented by Figure 5, the data

from step C are shown, evidencing the frequency with

which each estimate was made. Among the tests that

presented estimates with a precision of 0 to 4 meters,

most of these show estimates with a little more than 2

meters of error.

Figure 5: Histogram: location estimation errors using

WALDO architecture with the computer lab datase.

The same tests were performed by the KIOS

dataset and the results are presented in Figure 6. In

total, 792 tests were performed in several RPs of the

mapped environment. Observing the results of the es-

timates made through WALDO (step C), it can be ob-

served that most of the tests present results with errors

between 2 and 5 meters, with a median of approxi-

mately 3 meters. In addition, some outliers can be

observed, with errors over 10 meters.

Figure 6: Location errors using the KIOS dataset.

Figure 7: Histogram: location estimation errors using

WALDO architecture with the KIOS dataset.

5 CONCLUSION AND FUTURE

WORK

In the present work, an opportunistic system was pre-

sented that uses the available WLAN infrastructure to

provide the indoor location. Since there is no need to

install additional sensors, the costs for implementing

the WALDO architecture are greatly reduced. In addi-

tion, Wi-Fi technology is widely used by smartphone

users, so it is easy to access.

Because of the high variability of the strength of

the Wi-Fi signal, due to the interference and noise

common indoors, the estimation of positioning calcu-

lations can often suffer great inﬂuences. With the ap-

proach used in WALDO, by dividing the mapped en-

vironment into zones and using the rankings scheme,

it was possible to prioritize the readings (trace stage)

with less noise, causing outliers to receive less weight

or to be disregarded in the calculations.

No More Hiding! WALDO: Easily Locating with a Wi-Fi Opportunistic Approach

577

With the increasing growth of ubiquitous and per-

vasive technologies directed to internal environments,

it is evident the need of the applications to know the

location of the mobile devices to better adapt to sit-

uations, offering better services. Identifying that a

user is in a certain environment, because it is port-

ing a smartphone, creates several new possibilities

for applications. The WALDO provides parameters

to identify the location of mobile devices, with low

location errors, making it suitable for many types of

applications. The precision tests were performed us-

ing two distinct datasets, aiming to obtain non-biased

results. Although this is an opportunistic approach,

using variable signal technology, the test results were

satisfactory.

As future work, tests will be performed regard-

ing the inﬂuence caused by the use of heterogeneous

devices during the estimation of positioning. In addi-

tion, it is intended to implement a system that obtains

the RSS and MAC information through the APs, in

a passive tracking approach. In this way, it is possi-

ble to estimate the location of both sides of the com-

munication, seeking to improve the precision in the

estimates. In addition, the use of other information

sources along with Wi-Fi will be tested.

ACKNOWLEDGEMENTS

The authors would like to thank Penguim Formula

for partial supporting/funding of this research and

UFSM/FATEC through project number 041250 -

9.07.0025 (100548).

REFERENCES

Ahmed, T., Calders, T., and Pedersen, T. B. T. (2016). An-

alytics on indoor moving objects with applications in

airport baggage tracking.

Chen, L.-H., Wu, E. H.-K., Jin, M.-H., and Chen, G.-H.

(2014). Homogeneous features utilization to address

the device heterogeneity problem in ﬁngerprint local-

ization. IEEE Sensors Journal, 14(4):998–1005.

Fukuzaki, Y., Mochizuki, M., Murao, K., and Nishio, N.

(2015). Statistical analysis of actual number of pedes-

trians for wi-ﬁ packet-based pedestrian ﬂow sensing.

In Adjunct Proceedings of the 2015 ACM Interna-

tional Joint Conference on Pervasive and Ubiquitous

Computing and Proceedings of the 2015 ACM Inter-

national Symposium on Wearable Computers, pages

1519–1526. ACM.

Ge, X. and Qu, Z. (2016). Optimization wiﬁ indoor posi-

tioning knn algorithm location-based ﬁngerprint. In

Software Engineering and Service Science (ICSESS),

2016 7th IEEE International Conference on, pages

135–137. IEEE.

Group, I. . W. et al. (1999). Part11: Wireless lan medium

access control (mac) and physical layer (phy) speciﬁ-

cations. ANSI/IEEE Std. 802.11.

Hossain, A. M. and Soh, W.-S. (2015). A survey of

calibration-free indoor positioning systems. Com-

puter Communications, 66:1–13.

Kanan, R. and Elhassan, O. (2016). A combined batteryless

radio and wiﬁ indoor positioning for hospital nursing.

Kannan, B., Meneguzzi, F., Dias, M. B., and Sycara, K.

(2013). Predictive indoor navigation using commer-

cial smart-phones. In Proceedings of the 28th Annual

ACM Symposium on Applied Computing, pages 519–

525. ACM.

Lane, N. D., Chon, Y., Zhou, L., Zhang, Y., Li, F., Kim,

D., Ding, G., Zhao, F., and Cha, H. (2013). Pig-

gyback crowdsensing (pcs): energy efﬁcient crowd-

sourcing of mobile sensor data by exploiting smart-

phone app opportunities. In Proceedings of the 11th

ACM Conference on Embedded Networked Sensor

Systems, page 7. ACM.

Laoudias, C., Pich

e, R., and Panayiotou, C. (2013). De-

vice self-calibration in location systems using signal

strength histograms. 7:165–181.

Lymberopoulos, D., Liu, J., Yang, X., Choudhury, R. R.,

Handziski, V., and Sen, S. (2015). A realistic evalua-

tion and comparison of indoor location technologies:

Experiences and lessons learned. In Proceedings of

the 14th international conference on information pro-

cessing in sensor networks, pages 178–189. ACM.

Mainetti, L., Patrono, L., and Sergi, I. (2014). A survey on

indoor positioning systems. In Software, Telecommu-

nications and Computer Networks (SoftCOM), 2014

22nd International Conference on, pages 111–120.

IEEE.

Oosterlinck, D., Benoit, D. F., Baecke, P., and Van de

Weghe, N. (2017). Bluetooth tracking of humans in

an indoor environment: An application to shopping

mall visits. Applied Geography, 78:55–65.

Paul, A. S. and Wan, E. A. (2009). Rssi-based indoor

localization and tracking using sigma-point kalman

smoothers. IEEE Journal of Selected Topics in Sig-

nal Processing, 3(5):860–873.

Potort

ı, F., Crivello, A., Girolami, M., Traﬁcante, E., and

Barsocchi, P. (2016). Wi-ﬁ probes as digital crumbs

for crowd localisation. pages 1–8.

Rai, A., Chintalapudi, K. K., Padmanabhan, V. N., and Sen,

R. (2012). Zee: Zero-effort crowdsourcing for indoor

localization. In Proceedings of the 18th annual in-

ternational conference on Mobile computing and net-

working, pages 293–304. ACM.

Yang, C. and Shao, H.-R. (2015). Wiﬁ-based indoor posi-

tioning. IEEE Communications Magazine, 53(3):150–

157.

Yang, S., Dessai, P., Verma, M., and Gerla, M. (2013). Free-

loc: Calibration-free crowdsourced indoor localiza-

tion. In INFOCOM, 2013 Proceedings IEEE, pages

2481–2489. IEEE.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

578