On the Support of a Similarity-enabled Relational Database

Management System in Civilian Crisis Situations

Paulo H. Oliveira, Antonio C. Fraideinberze, Natan A. Laverde, Hugo Gualdron,

Andre S. Gonzaga, Lucas D. Ferreira, Willian D. Oliveira, Jose F. Rodrigues-Jr.,

Robson L. F. Cordeiro, Caetano Traina Jr., Agma J. M. Traina and Elaine P. M. Sousa

Institute of Mathematics and Computer Sciences, University of Sao Paulo,

Av. Trabalhador Sancarlense, 400, Sao Carlos, SP, Brazil

Keywords:

Crisis Situation, Crisis Management, Relational Database Management System, Similarity Query.

Abstract:

Crowdsourcing solutions can be helpful to extract information from disaster-related data during crisis man-

agement. However, certain information can only be obtained through similarity operations. Some of them

also depend on additional data stored in a Relational Database Management System (RDBMS). In this con-

text, several works focus on crisis management supported by data. Nevertheless, none of them provide a

methodology for employing a similarity-enabled RDBMS in disaster-relief tasks. To ﬁll this gap, we intro-

duce a methodology together with the Data-Centric Crisis Management (DCCM) architecture, which employs

our methods over a similarity-enabled RDBMS. We evaluate our proposal through three tasks: classiﬁcation

of incoming data regarding current events, identifying relevant information to guide rescue teams; ﬁltering

of incoming data, enhancing the decision support by removing near-duplicate data; and similarity retrieval of

historical data, supporting analytical comprehension of the crisis context. To make it possible, similarity-based

operations were implemented within one popular, open-source RDBMS. Results using real data from Flickr

show that our proposal is feasible for real-time applications. In addition to high performance, accurate results

were obtained with a proper combination of techniques for each task. Hence, we expect our work to provide a

framework for further developments on crisis management solutions.

1 INTRODUCTION

Crisis situations, such as conﬂagrations, disasters in

crowded events, and workplace accidents in industrial

plants, may endanger human life and lead to ﬁnan-

cial losses. A fast response to this kind of situation

is essential to reduce or prevent damage. In this con-

text, software systems aimed at supporting experts in

decision-making can be used to better understand and

manage crises. A promising line of research is the use

of social networks or crowdsourcing (Kudyba, 2014)

to gather information from the crisis site.

Several desirable tasks can be performed by soft-

ware systems designed for aiding in decision-making

during crises. One of such tasks is to detect the evi-

dences that best depict the crisis situation, so that res-

cue teams can be aware of it and prepare themselves

properly. For instance, identifying ﬁre or smoke on

multimedia data, such as images, videos or textual re-

ports, usually points to conﬂagration. Some relevant

proposals in this direction comprehend ﬁre and smoke

detection based on image processing approaches (Ce-

lik et al., 2007) and techniques for ﬁre detection de-

signed over image descriptors that focus on detecting

ﬁre from social media images (Bedo et al., 2015).

Another important task is to ﬁlter the information

received from crowdsourcing solutions dedicated to

collecting data from crises. When reporting incidents,

users might end up sending too much similar informa-

tion, such as pictures from the same angle of the same

object. Such excess of similar data demands a longer

time to be processed. Moreover, it turns the decision-

making process more time-consuming. Therefore, re-

moving duplicates is an essential task in this context.

The task of searching for similar data in historical

databases can support decision-making as well. Take

for instance a database that contains images and tex-

tual descriptions regarding past crisis situations. If the

crowdsourcing system gets, for instance, images de-

picting ﬁre, a query might be posed on the database to

Oliveira, P., Fraideinberze, A., Laverde, N., Gualdron, H., Gonzaga, A., Ferreira, L., Oliveira, W., Rodrigues-Jr., J., Cordeiro, R., Jr., C., Traina, A. and Sousa, E.

On the Support of a Similarity-enabled Relational Database Management System in Civilian Crisis Situations.

In Proceedings of the 18th International Conference on Enterprise Information Systems (ICEIS 2016) - Volume 1, pages 119-126

ISBN: 978-989-758-187-8

119

retrieve similar images and the corresponding textual

descriptions. Then, based on these results, specialists

would potentially infer the kind of material burning in

the crisis, by analyzing the color tone of the smoke in

the retrieved images and their textual descriptions.

For all those tasks, it is desirable that a commodity

system provide functionalities over existing software

infrastructure. Commodity systems that can play this

role are the Relational Database Management Sys-

tems (RDBMS). They are largely available in the cur-

rent computing technology and are able to bring new

functionalities without the need of redesigning the ex-

isting software. Moreover, RDBMS provide efﬁcient

data storage and retrieval. However, they do not read-

ily support similarity operations, which are needed to

address the aforementioned tasks.

Several works in the literature aim at embedding

similarity support in RDBMS. Nevertheless, the liter-

ature lacks a methodology for employing a similarity-

enabled RDBMS in the context of crisis management.

This work aims at ﬁlling that gap. Our hypothesis is

that providing similarity support on an RDBMS helps

the decision support in crisis situations.

We contribute with a data-centric architecture for

decision-making during crisis situations by means of

a similarity-enabled RDBMS. Our proposal is evalu-

ated using an image dataset of real crises from Flickr

in performing three tasks:

• Task 1. Classiﬁcation of incoming data regarding

current events, detecting the most relevant infor-

mation to guide rescue teams in the crisis site;

• Task 2. Filtering of incoming data, enhancing the

decision support of rescue command centers by

removing near-duplicate data;

• Task 3. Similarity retrieval from past crisis situa-

tions, supporting analytical comprehension of the

crisis context.

This work has been conducted to cater to demands

of the project RESCUER: Reliable and Smart Crowd-

sourcing Solution for Emergency and Crisis Manage-

ment

, supported by the European Union’s Research

and Innovation Funding Program FP7.

The results of our experimentation show that the

proposed architecture is effective over crisis scenarios

which rely on multimedia data. In addition to the high

performance achieved, accurate results are obtained

when using a proper combination of techniques.

The rest of the paper is structured as follows. Sec-

tion 2 presents the related work and Section 3 presents

the main concepts for similarity support on RDBMS.

http://www.rescuer-project.org/

Section 4 describes the new Data-Centric Crisis Man-

agement architecture, on which the proposed method-

ology is based. Section 5 presents our methodology,

describes the experiments and discusses the results.

Finally, the conclusions are presented in Section 6.

2 RELATED WORK

Existing research on crisis management highlights the

importance of computer-assisted systems to support

this task. The approaches may be categorized into dif-

ferent types according to their purpose.

One type refers to localization, whose purpose is

to determine where victims are located during a disas-

ter. There are works that accomplish this task by em-

ploying cell phone localization techniques, such as In-

ternational Mobile Subscriber Identity (IMSI) catch-

ers (Reznik et al., 2015).

Another type regards logistics. Examples compre-

hend an integer programming technique for modeling

multiple-resource emergency responses (Zhang et al.,

2012) and a methodology for routing rescue teams to

multiple communities (Huang et al., 2013).

A different line of work refers to decision-making

based on social media (Gibson et al., 2014). Most of

the work focus on textual data, specially from services

like Twitter (Ghahremanlou et al., 2015).

Although all the aforementioned approaches have

been conceived to cater to different requirements, all

of them share the characteristic of using Information

Communication Technology (ICT) in response to cri-

sis situations. The decision-making systems based on

incoming data have one more characteristic: the par-

ticipation of people somehow involved in the disaster.

Existing work have focused on the importance of

crowdsourcing data for crisis management in the post-

2015 world (Halder, 2014). Therefore, to describe

our methodology, we assume the existence of crowd-

sourcing as a subsystem dedicated to gathering input

data. Additionally, we assume the existence of a com-

mand center, where analysts evaluate the input data in

order to guide the efforts of a rescue team at the crisis

site.

The work of Mehrotra (Mehrotra et al., 2004) is

the closest approach with respect to our methodology.

That work presents an interesting approach, but it fo-

cuses mostly on textual data and spatial-temporal in-

formation, rather than on more kinds of complex data,

such as images. Furthermore, it lacks a methodology

for employing content-based operations. We ﬁll those

gaps by providing a methodology to perform such op-

erations over disaster-related data and provide useful

information to rescue teams.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

120

3 BACKGROUND

3.1 Content-based Retrieval

Complex data is a common term associated with ob-

jects such as images, audio, time series, geographical

data and large texts. Such data do not present order

relation and, therefore, are unable to be compared by

relational operators (<, ≤, ≥, >). Equality operators

(=, 6=) could be used, but they have little or no mean-

ing when employed on such data. Nevertheless, com-

plex data can be compared according to their content

by using similarity concepts (Barioni et al., 2011).

The interaction with a content-based retrieval sys-

tem starts as the user enters a query, providing a com-

plex object as the query example. This complex ob-

ject is submitted to a feature extractor, which extracts

representative characteristics from it and generates a

feature vector. The feature vector is sent to an evalua-

tion function, which compares another feature vector

stored in the database and returns a value representing

the dissimilarity degree (also known as the distance)

between both feature vectors. This comparison is re-

peated over the database, generating the results at the

end of the process and sending them to the user.

Two of the most common queries used in content-

based retrieval are the Range Query and the k Nearest

Neighbor (kNN) Query (Barioni et al., 2011). Range

Query is deﬁned by the function Rq(s

, ξ), where s

represents an object from data domain S and ξ is the

radius used as distance constraint. The query returns

all objects within a distance ξ from s

. kNN Query

is deﬁned by the function kNNq(s

, k), where s

rep-

resents an object from the data domain S and k is the

number of elements to be returned. The query returns

the k most similar objects to s

. The kNN Queries are

employed in the context of Instance-Based Learning

(IBL) algorithms, such as the kNN Classiﬁer, which is

used in our proposal and thus discussed in Section 3.2.

The feature extraction is usually required because

the original representation of a given complex object

is not prone to useful and efﬁcient computation. Eval-

uation functions are able to compute the dissimilarity

degree of a pair of feature vectors. These subjects are

discussed in Sections 3.3 and 3.4.

The similarity retrieval process can be performed

outside an RDBMS. However, enabling an RDBMS

with similarity is a promising approach and there are

several ways for doing so, as discussed in Section 3.5.

3.2 kNN Classiﬁer

The concept of Instance-Based Learning (IBL) (Aha

et al., 1991) comprehends supervised learning algo-

rithms that make predictions based solely on the in-

stances previously stored in the database. In these al-

gorithms, no model is built. The knowledge is repre-

sented by the data instances already stored and classi-

ﬁed. Then, new instances are classiﬁed in relation to

the existing stored instances, according to their sim-

ilarity. One of the main IBL algorithms is the well-

known kNN Classiﬁer (Fix and Hodges Jr., 1951).

For a given unlabeled instance, the kNN Classi-

ﬁer retrieves from the database the k most similar in-

stances. Following, it predicts the label based on the

retrieved instances, according to some predeﬁned cri-

terion. A simple one is to assign the label of the pre-

vailing class in the k nearest neighbors. Another one

is to weigh the retrieved instances by distance, so the

closest ones have a higher inﬂuence.

3.3 Feature Extractors

One of the main tasks for retrieving complex data by

content is the feature extraction process, which maps

a high-dimensional input data into a low-dimensional

feature space, extracting useful information from raw

data while reducing their content. Using proper fea-

ture extractors for a complex data domain leads to re-

sults closer to what the users expect (Sikora, 2001).

Several feature extractors have been developed for

different application domains. The main characteris-

tics investigated in the context of images are color,

texture and shape. There are several feature extractors

for such characteristics, some of which are part of the

MPEG-7 standard (MultiMedia, 2002). In this work,

we employ a color-based and a hash-based extractors.

Color-based Extractors. Color-based extractors

are commonly used as a basis for other extractors. For

this reason, they are the most used visual descriptors

in content-based image retrieval. The color-based fea-

ture extractors in the MPEG-7 standard are commonly

employed in the literature. One of them is the Color

Structure Descriptor, which builds a color histogram

based on the local features of the image.

Perceptual Hash. An extractor suitable for near-

duplicate detection is the Perceptual Hash

. It gener-

ates a “ﬁngerprint” of a multimedia ﬁle derived from

various features from its content. These “ﬁngerprints”

present the characteristic of being close to one another

if the extracted features are similar.

3.4 Evaluation Functions

The dissimilarity between two objects is usually de-

termined by a numerical value obtained from an eval-

http://www.phash.org/

On the Support of a Similarity-enabled Relational Database Management System in Civilian Crisis Situations

121

uation function. Objects with smaller values are con-

sidered to be more alike.

The Minkowski Family comprehends evaluation

functions known as L

metrics that are widely used in

content-based retrieval (Wilson and Martinez, 1997).

The L

metric corresponds to the Manhattan Distance,

also noted as City-Block Distance. The L

metric is

the well-known Euclidean Distance. Finally, there is

the L

∞

metric, also noted as the Chebyshev Distance.

The Hamming Distance (Hamming, 1950), which

is another well-known evaluation function, counts the

substitutions needed to transform one of the input data

into the other. It can be employed in near-duplicate

detection tasks, since combining it with the Percep-

tual Hash leads to accurate results.

3.5 Similarity Support on RDBMS

SimDB (Silva et al., 2010) is a similarity-enabled

RDBMS, based on PostgreSQL. The similarity oper-

ations and keywords were included in its core. Equiv-

alence rules were also included, which allows alterna-

tive query plans. However, similarity queries are only

available for numerical data, and traditional queries

over other data types are not supported.

SIREN (Barioni et al., 2011) is a middleware be-

tween a client application and the RDBMS. The client

sends SQL commands extended with similarity key-

words, which are checked by SIREN to identify sim-

ilarity predicates. First, SIREN evaluates the similar-

ity predicates, accessing an index of feature vectors,

then uses the RDBMS for traditional predicates.

FMI-SiR (Kaster et al., 2011) is a framework that

operates over the RDBMS Oracle, employing user-

deﬁned functions to extract features and to index data.

MedFMI-SiR is an extension of FMI-SiR for medical

images in the Digital Imaging and Communications

in Medicine (DICOM) format.

SimbA (Bedo et al., 2014) is a framework that ex-

tends the middleware SIREN. SimbA supports the in-

clusion and combination of feature extractors, evalu-

ation functions and indexes on demand. The queries

are processed just like on SIREN.

4 PROPOSED ARCHITECTURE

This section presents our architecture for crisis man-

agement, named as Data-Centric Crisis Management

(DCCM). Section 4.1 describes the scenario of a typ-

ical crisis situation managed by DCCM. Then, Sec-

tion 4.2 describes our architecture.

4.1 Crisis Management Scenario

Figure 1 shows the scenario of a crisis situation sup-

ported by DCCM.

Figure 1: Scenario of a typical crisis situation considering

our architecture for crisis management.

In a Crisis Situation, eyewitnesses can collect data

regarding the event. For instance, they can take pic-

tures, record videos and make textual reports, which

are sent to the Crowdsourcing System. In Figure 1, the

pictures taken by the eyewitnesses are redirected as an

Image Stream to DCCM. Then, the command center

can query DCCM for the Decision-Making process.

Additionally, the crowdsourcing system could re-

ceive other data, such as metadata (e.g. time and GPS

localization) or other data types (e.g. video and text).

4.2 Data-centric Crisis Management

The Data-Centric Crisis Management (DCCM) archi-

tecture is represented in Figure 2. The whole mech-

anism has three processes, each of them depicted in

the ﬁgure by arrows marked with the letters A, B and

C, which represent the tasks introduced in Section 1.

In a crisis situation, we consider the existence of

a crowdsourcing system that receives disaster-related

complex objects (A1) and submits them to DCCM.

Each object of the data stream is placed in a Buffer

and analyzed by the Filtering Engine. First, the en-

gine checks whether the object is a near duplicate of

some other object currently within the Buffer. For the

near-duplicate checking, the Filtering Engine uses the

Similarity Engine (A2) to extract a feature vector from

the object and compare it to the feature vectors of the

other objects within the Buffer. The object is marked

as a near duplicate when its distance from at least an-

other object is at most ξ, which is a threshold deﬁned

by specialists according to the application domain.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

122

Figure 2: The DCCM architecture, consisting of the tasks:

classiﬁcation (A), ﬁltering (B) and historical retrieval (C).

If the object is not a near duplicate, then it is sub-

mitted to the Classiﬁcation Engine (A3). The classi-

ﬁcation process uses Historical Records in a database

to train classiﬁer methods (A4). Based on such train-

ing, the Classiﬁcation Engine labels the object regard-

ing the event it represents. For instance, it can be la-

beled as “ﬁre” or “smoke”. Finally, the Classiﬁcation

Engine notiﬁes the specialists with the now-classiﬁed

object for the Decision-Making process (A5).

If the object is a near duplicate, then it is not sub-

mitted to the Classiﬁcation Engine. Instead, it stays

in the Buffer to be compared to others that come later.

Moreover, it is associated with the event of the object

of which it is a near duplicate. In Figure 2, the objects

from the same event have the same color. The white

objects marked with “?” have not been analyzed yet.

The Buffer may be determined either by a physical

size, such as the number of elements it holds, or by a

time window. In Figure 2, it is delimited by a time

window of length k, beginning at time t and ending

at time t −k. The Buffer is ﬂushed at every k-th time

instant. Before ﬂushing, the Representative Selector

selects the object of each group that best represents its

event (B1) according to a predeﬁned criterion.

If a near-duplicate object is selected as the repre-

sentative, then it gets the label of the classiﬁed object

of its group. The already-classiﬁed object, in turn, is

marked as near duplicate. On the other hand, if the se-

lected representative is already the classiﬁed object of

its group, then no changes are made. Lastly, the clas-

siﬁed objects are stored in the database and the near

duplicates are discarded, ﬂushing the Buffer (B2).

There is another use case for DCCM, which refers

to the historical analyses. The Decision-Making team

may want to provide complex data samples to retrieve

similar events from the past. For this purpose, DCCM

provides the Historical Retrieval Engine (C1). First,

the engine extracts the features from each provided

sample (C2). Then, it compares the extracted features

against the Historical Records and provides its ﬁnd-

ings to the Decision-Making team (C3).

5 CASE STUDY

In this section, we present the case study for evaluat-

ing the DCCM architecture over the three tasks dis-

cussed earlier. The experiments were carried out over

a real crisis dataset known as Flickr-Fire (Bedo et al.,

2015) containing 2,000 images extracted from Flickr,

1,000 labeled as “ﬁre” and 1,000 as “not ﬁre”.

5.1 Implementation of DCCM

To implement DCCM, we extended the open-source

RDBMS PostgreSQL. Our implementation, named as

Kiara, supports an SQL extension for building simi-

larity queries over complex data (Barioni et al., 2011).

Also through the SQL extension, Kiara allows manag-

ing feature extractors and evaluation functions, which

are dynamically (no recompilation) inserted and up-

dated by user-deﬁned functions written in C++. Kiara

makes use of metatables to keep track of feature ex-

tractors and evaluation functions associated with the

attributes of complex data that a user instantiates.

To support the SQL extension, we built a parser

that works like a proxy in the core of Kiara. It receives

a query and rewrites only the similarity predicates, ex-

pressing them through standard SQL operators. Then,

it sends the rewritten queries to the core of Kiara.

After inserting a new extractor, the features are au-

tomatically extracted from the complex data (e.g. im-

age, video or text) and then stored in user-deﬁned at-

tributes dedicated to representing such data. Similar-

ity queries can be included through PL/pgSQL func-

tions and new indexes can be included through the in-

terface known as Generalized Search Tree (GiST), al-

ready present in PostgreSQL. Moreover, Kiara allows

exploring alternative query plans involving traditional

and similarity predicates.

5.2 Classiﬁcation of Incoming Data

5.2.1 Methodology

Classifying disaster-related incoming data is helpful

because of two reasons. One of them is to identify the

characteristic that best depicts the crisis situation. The

other is to store data properly labeled, which improves

On the Support of a Similarity-enabled Relational Database Management System in Civilian Crisis Situations

123

further queries on a historical database. To do so, the

DCCM architecture employs the kNN Classiﬁer.

The parameter k can be selected arbitrarily. How-

ever, too small values can be noise-sensitive, whereas

too large values allow including more instances from

other classes, leading to misclassiﬁed instances.

5.2.2 Experimentation and Results

In this task, we classiﬁed the elements of the Flickr-

Fire dataset. For a robust experimentation, we used

the procedure 10-fold cross-validation for a kNN clas-

siﬁer using k = 10. We used the Manhattan Distance

and the extractor Color Structure Descriptor because

existing work showed that they allow accurate results

for ﬁre detection (Bedo et al., 2015).

After 10 rounds of evaluation, we took the average

accuracy and the average F1 score. The result was the

same for both measures, which was 0.86. Considering

a real event, this capability would be able to automati-

cally group data according to their content, indicating

the main characteristics of the crisis and thus saving

the command center crucial time for fast response.

5.3 Filtering of Incoming Data

5.3.1 Methodology

In the task of ﬁltering, we are interested in preventing

duplicate information from being classiﬁed and sub-

sequently sent to the command center.

To determine whether the incoming data is a near

duplicate of existing data, they must be compared by

their content. For this purpose, we must employ sim-

ilarity queries. In this case, though, we are restricted

to Range Queries. If the new object is a near duplicate

of an object in the buffer, then the distance from each

other is at most ξ, which is supposed to be a small

threshold (range), since we want to detect pairs of ob-

jects that, in essence, represent the same information.

Range Queries allow restricting results based on their

similarity, differently from kNN Queries, which do it

by the number of objects retrieved.

Hence, the DCCM architecture prevents near du-

plicates by using Range Queries. Each object that ar-

rives in the buffer is submitted to a default feature ex-

tractor. Then, a Range Query is performed by using

the extracted features as the s

object. The range value

ξ must be predeﬁned as well, according to the appli-

cation domain. If at least one object from the buffer

is retrieved by the Range Query, then the s

object is

marked as a near duplicate.

5.3.2 Experimentation and Results

For this experiment, we employed the Hamming Dis-

tance with the Perceptual Hash extractor and assumed

a buffer size of 80. We ﬁlled the buffer with 80 images

from Flickr, of which 37 depict “small ﬁre” events and

43 depict “big ﬁre” events. Each of the 80 images was

used as the s

to a Range Query with ξ = 10.

The ξ parameter was set to retrieve around half of

the images (40 images approximately), in order to be

able to return an entire class (“big ﬁre” or “small ﬁre”)

of images. This allows evaluating the precision of the

queries with the Precision-Recall method.

Figure 3 shows the Precision-Recall curve for this

set of queries. The curve falls off only after 80% of

retrieval. This late fall-off is characteristic of highly

effective retrieval methods. In this result, one can no-

tice a precision above 90% up to 50% of recall.

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Precision

Recall

Precision-Recall for filtering incoming data

Figure 3: Precision-Recall in the process of ﬁltering incom-

ing data in the buffer.

The results show that DCCM is expected to ﬁlter

out 90% of near-duplicate data. This is a strong indi-

cation that such capability can signiﬁcantly improve

both efﬁciency and efﬁcacy of a command center. Fil-

tering is the most desirable functionality considered

in this work. This is because crowdsourcing is highly

prone to produce redundant data. Right after a crisis is

installed, if ﬁltering is not possible, the ﬂow of infor-

mation streamed by eyewitnesses may be too high for

the command center to make good use of them. How-

ever, a similarity-enabled RDBMS in DCCM is able

to handle such situation with basic similarity queries.

5.4 Retrieval of Historical Data

5.4.1 Methodology

In the context of crisis management, the experts from

the command center might be willing to analyze data

from past events that are similar to the current ones.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

124

Such data may lead to decisions about how to proceed

with the current crisis. In these situations, similarity

queries play an important role.

Considering the DCCM architecture, this task can

be performed whenever the Decision-Making experts

want information similar to the current data. For ev-

ery notiﬁed data at point A5 of Figure 2, they might

provide it as the s

element to the Historical Retrieval

Engine in order to get similar information.

5.4.2 Experimentation and Results

For this experiment, we combined the Color Structure

Descriptor and the Manhattan Distance.

We performed Range and kNN Queries using each

of the 2,000 elements from the Flickr-Fire dataset as

the s

element. We set the k parameter to 1,000 and

the ξ parameter to 7.2, retrieving an entire class.

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Precision

Recall

Precision-Recall for retrieving historical data

Range 7.2

kNN 1000

Figure 4: Precision-Recall for retrieving historical data.

We generated the Precision-Recall curve depicted

in Figure 4. From these results, one can observe the

high precision around 0.8 when fetching 10% of rele-

vant data, roughly 100 images, and around 0.9 when

fetching 5%, nearly 50 images — a more realistic sce-

nario. These results point to an effective retrieval of

images based on their class.

From the point of view of a command center user,

there would be an ample knowledge bank from which

initial considerations could be drawn from the current

crisis. This initial knowledge has the potential of sav-

ing time of rescue actions by preventing past mistakes

and fostering successful decisions.

5.5 Overall Performance

Concerning a computer system, it is important to re-

ceive the correct response in a timely manner. There-

fore, we analyzed the overall performance of DCCM.

For this purpose, we carried out one experiment re-

garding scalability and three regarding the tasks.

Overall Scalability. A solution based on DCCM

spends most of its time receiving, storing and index-

ing data for the sake of similarity retrieval. Therefore,

such processing must be efﬁcient. We carried out an

experiment to evaluate the time spent extracting fea-

tures and inserting them into the database. The aver-

age time of ﬁve rounds is presented in Figure 5.

From the results presented in the ﬁgure, one can

calculate that the solution is able to process up to 3

images per second — sufﬁcient for most scenarios.

These numbers refer to a machine with a hard disk of

5,400RPM; such results could be improved by using

SSD disks or RAID subsystems.

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

200

400

600

800

1000

1200

1400

1600

1800

2000

Time (s)

Number of instances in the database

Time spent to store one image

Extraction

Insertion

Figure 5: Time to extract features from one image and insert

them into the database.

From Figure 5, one can also observe that the time

spent inserting images is mostly taken by the feature

extraction, while the time for inserting the features re-

mains constant. The extraction time varies according

to the image resolutions, which range from 300x214

to 3,240x4,290 pixels in the Flickr-Fire dataset.

Table 1: Overall performance of DCCM over Flickr-Fire.

Task Average time (sec)

Classiﬁcation 0.851

Filtering 0.057

Retrieval — Range Query 1.147

Retrieval — kNN Query 0.849

Table 1 presents the performance of DCCM to per-

form the three tasks of our methodology. We ran the

classiﬁcation and ﬁltering tasks 10 times, whereas the

retrieval tasks were performed 2,000 times, once for

each element in the dataset. For the classiﬁcation task,

we used the distanced-weighted kNN classiﬁer, with

k = 10 and performing 10-fold cross-validation. For

the ﬁltering task, we used the 80 aforementioned near-

duplicate images and the range value ξ was set to 10.

Finally, in the retrieval tasks, ξ was set to 2.8 for the

On the Support of a Similarity-enabled Relational Database Management System in Civilian Crisis Situations

125

Range Queries, retrieving around 50 tuples, and k was

set to 50 for the kNN Queries. The results, which rep-

resent the average time to perform each task once, in-

dicate that our proposal is feasible in a real-time crisis

management application.

6 CONCLUSIONS

Fast and precise responses are essential characteris-

tics of computational solutions. In this paper, we pro-

posed the architecture of a solution that can achieve

these characteristics in crisis management tasks. In

the course of our work, we described the use of a

similarity-enabled RDBMS in tasks that could assist a

command center in guiding rescue missions. To make

it possible, we implemented similarity-based opera-

tions within one popular, open-source RDBMS.

The core of our work is related to an innovation

project led by the European Union; accordingly, we

applied similarity retrieval concepts in an innovative

manner, putting together relational and retrieval tech-

nologies. To demonstrate our claims, we carried out

experiments to evaluate both the efﬁcacy and the ef-

ﬁciency of our proposal. More speciﬁcally, we intro-

duced the following functionalities:

• Classiﬁcation of Incoming Data. We proposed

to employ kNN classiﬁcation to classify incoming

data, aiming at identifying and characterizing cri-

sis situations faster;

• Filtering of Incoming Data. We proposed to em-

ploy Range Queries to ﬁlter out redundant infor-

mation, aiming at reducing the data load over the

system and over a command center;

• Retrieval of Historical Data. We proposed to

employ Range and kNN Queries to retrieve data

from past crises that are similar to the current one.

The results we obtained for each of these tasks al-

lowed us to claim that a similarity-enabled RDBMS is

able to assist in the decision support of command cen-

ters when a crisis situation strikes. We conclude by

stating that our work demonstrated the use of cutting-

edge methods and technologies in a critical scenario,

paving the way for similar systems to ﬂourish based

on the experiences that we reported.

ACKNOWLEDGEMENTS

This research has been supported, in part, by FAPESP,

CAPES, CNPq and the RESCUER project, funded by

the European Commission (Grant: 614154).

REFERENCES

Aha, D., Kibler, D., and Albert, M. (1991). Instance-based

learning algorithms. Machine Learning.

Barioni, M., Kaster, D., Razente, H., Traina, A., and Traina

Jr., C. (2011). Querying Multimedia Data by Sim-

ilarity in Relational DBMS. In Advanced Database

Query Systems.

Bedo, M., Blanco, G., Oliveira, W., Cazzolato, M., Costa,

A., Rodrigues Jr., J., Traina, A., and Traina Jr., C.

(2015). Techniques for effective and efﬁcient ﬁre de-

tection from social media images. ICEIS ‘15.

Bedo, M., Traina, A., and Traina Jr., C. (2014). Seamless

integration of distance functions and feature vectors

for similarity-queries processing. JIDM.

Celik, T., Ozkaramanli, H., and Demirel, H. (2007). Fire

and smoke detection without sensors: Image process-

ing based approach. EUSIPCO ‘07.

Fix, E. and Hodges Jr., J. (1951). Discriminatory analysis

— Nonparametric discrimination: Consistency prop-

erties. Technical report, DTIC Document.

Ghahremanlou, L., Sherchan, W., and Thom, J. (2015).

Geotagging Twitter messages in crisis management.

The Computer Journal.

Gibson, H., Andrews, S., Domdouzis, K., Hirsch, L., and

Akhgar, B. (2014). Combining big social media data

and FCA for crisis response. UCC ‘14.

Halder, B. (2014). Crowdsourcing collection of data for

crisis governance in the post-2015 world: Potential of-

fers and crucial challenges. ICEGOV ‘14.

Hamming, R. W. (1950). Error detecting and error correct-

ing codes. Bell System Technical Journal.

Huang, M., Smilowitz, K., and Balcik, B. (2013). A contin-

uous approximation approach for assessment routing

in disaster relief. Transportation Research Part B.

Kaster, D., Bugatti, P., Ponciano-Silva, M., Traina, A.,

Marques, P., Santos, A., and Traina Jr., C. (2011).

MedFMI-SiR: A powerful DBMS solution for large-

scale medical image retrieval. ITBAM ‘11.

Kudyba, S. (2014). Big Data, Mining, and Analytics: Com-

ponents of Strategic Decision Making.

Mehrotra, S., Butts, C., Kalashnikov, D., Venkatasubra-

manian, N., Rao, R., Chockalingam, G., Eguchi, R.,

Adams, B., and Huyck, C. (2004). Project Rescue:

Challenges in responding to the unexpected. EI ‘04.

MultiMedia, I. (2002). MPEG-7: The generic multimedia

content description standard, p. 1. IEEE MultiMedia.

Reznik, T., Horakova, B., and Szturc, R. (2015). Advanced

methods of cell phone localization for crisis and emer-

gency management applications. IJDE.

Sikora, T. (2001). The MPEG-7 visual standard for content

description — An overview. IEEE Trans. Cir. Sys. Vid.

Silva, Y., Aly, A., Aref, W., and Larson, P. (2010). SimDB:

A similarity-aware database system. SIGMOD ‘10.

Wilson, D. and Martinez, T. (1997). Improved heteroge-

neous distance functions. J. Artif. Int. Res.

Zhang, J., Li, J., and Liu, Z. (2012). Multiple-resource and

multiple-depot emergency response problem consid-

ering secondary disasters. Expert Syst. Appl.

ICEIS 2016 - 18th International Conference on Enterprise Information Systems

126