Adaptive Exploration of a UAVs Swarm for Distributed Targets

Detection and Tracking

Mario G. C. A. Cimino

, Massimiliano Lega

, Manilo Monaco

and Gigliola Vaglini

Department of Information Engineering, University of Pisa, 56122 Pisa, Italy

Department of Engineering University of Naples “Parthenope”, 80143 Naples, Italy

Keywords: UAV, Swarm Intelligence, Stigmergy, Flocking, Differential Evolution, Target Detection, Target Tracking.

Abstract: This paper focuses on the problem of coordinating multiple UAVs for distributed targets detection and

tracking, in different technological and environmental settings. The proposed approach is founded on the

concept of swarm behavior in multi-agent systems, i.e., a self-formed and self-coordinated team of UAVs

which adapts itself to mission-specific environmental layouts. The swarm formation and coordination are

inspired by biological mechanisms of flocking and stigmergy, respectively. These mechanisms, suitably

combined, make it possible to strike the right balance between global search (exploration) and local search

(exploitation) in the environment. The swarm adaptation is based on an evolutionary algorithm with the

objective of maximizing the number of tracked targets during a mission or minimizing the time for target

discovery. A simulation testbed has been developed and publicly released, on the basis of commercially

available UAVs technology and real-world scenarios. Experimental results show that the proposed approach

extends and sensibly outperforms a similar approach in the literature.

1 INTRODUCTION

In this paper we consider the problem of discovering

and tracking static or dynamic targets in unstructured

environments, with no prior knowledge about their

location and about the obstacles layout (Senanayake

et al. 2016). Examples of scenarios in this context are:

(i) illegal dumps, in peri-urban and rural areas without

waste/sanitation facilities (Persechino et al. 2010);

(ii) anti-personnel landmines, left after a conflict in

areas such as natural parks, agricultural and grazing

lands (Rodriguez et al. 2014); (iii) early wildfire,

intentionally or naturally lighted in large open areas,

e.g., ecological habitats (Howden 2013); (iv) early

toxic or dangerous gas dispersion, in industrial/urban

areas characterized by large plants layouts (Qingchun

et al. 2011). The quality of the process can be

improved either by minimizing the time needed for

discovering the given targets, or by maximizing the

number of discovered targets in the given time.

In such type of missions, a promising approach is

to employ small Unmanned Aerial Vehicles (UAVs)

(Whitehead et al., 2014). The current UAVs

equipment and the available flight control logic offer

good solutions to the problem in a variety of fields.

However, the current solutions for coordinating the

exploration of UAVs swarms are not sufficiently

mature: limited flexibility, complex management and

application-dependent design are the main issues to

solve (Senanayake et al. 2016).

Essentially, in environmental monitoring and

surveillance one of the main requirements is to deal

with circumstances where the target and the space of

exploration are poorly specified. For this purpose, the

UAVs coordination strategy should be autonomous,

robust, resilient, and adaptive. Centralized logic

solutions are not effective for this purpose, due to the

high level of complexity, design and management

effort (McCune et al., 2013). In contrast,

decentralized logic approaches can provide a UAVs

swarm with a certain degree of autonomy (Meng et

al., 2014).

More specifically, a basic swarm of UAVs is

characterized by a large number of homogeneous

individuals, called agents, with local communication,

sensing and actuation capabilities (Maza et al., 2015).

A multi-agent system presents a number of

advantages: (i) it allows parallel/collective scan,

according to the principles of self-organization; (ii) it

is scalable, since by increasing the number of agents

in the swarm its effectiveness is poorly compromised;

(iii) it is flexible, because the agent logic is simple and

Cimino, M., Lega, M., Monaco, M. and Vaglini, G.

Adaptive Exploration of a UAVs Swarm for Distributed Targets Detection and Tracking.

DOI: 10.5220/0007581708370844

In Proceedings of the 8th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2019), pages 837-844

ISBN: 978-989-758-351-3

837

can be easily adapted to the scenario; (iv) it is robust,

and then the task accomplishment is not affected by

the fault of some agents (Aznar et al., 2014).

In general, a target search mission managed by a

multi-agent system is characterized by (i) the number

of targets and agents; (ii) the mobility of targets; (iii)

the complexity of the environment; (iv) the prior

knowledge about the target; (v) the type of swarm

coordination. We propose an environment model to

realistically represent the key characteristics of a

mission, and to test the effectiveness of a given

coordination logic. With regard to the swarm

coordination logic, the approach proposed in this

paper is based on three major aspects: (a) spatial self-

formation in order to better explore the environment,

i.e., the UAV movement is made according to a set of

mutual spatial constraints of arrangement; (b)

collaboration in order to exploit the knowledge

already acquired, i.e., each UAV both contributes and

is subject to some potential field influencing the

steering; (c) adaptation in order to optimize the

global behavior, i.e., the UAVs swarm adapts its

behavioral parameters, considering also (a) and (b), to

the specific type of mission.

In particular, the coordination logic is inspired by

behavioral patterns of biological systems. The

integration of biological patterns in a computational

coordination logic has to consider the enhancements

of the current UAVs information technology, such as

instant communication, simultaneous localization

and mapping, long-range sensing, etc. The final

purpose is to verify on realistic settings whether the

designed logic and the considered technological

enhancements allow a reduction of complexity and a

more effective optimization, keeping the essential

benefits of the original biological models.

For this purpose, we (i) propose a swarm

coordination algorithm that is adaptive to

heterogeneous scenarios with either static or dynamic

targets, (ii) develop and publicly release a simulation

testbed on which the commercially available UAVs

technology and real-world scenarios can be

considered.

The paper is structured as follows. The swarm

coordination logic and the operating environment is

presented in Section 2. In Section 3, the real-world

scenarios and the related UAV technologies are

detailed. Experimental results are presented and

discussed in Section 4. Finally, Section 5 summarizes

conclusions and future work.

2 ENVIRONMENT FOR SWARM

COORDINATION LOGIC

In the design of the testbed, an important distinction

is between flight simulator and exploration simulator.

A flight simulator focuses on control logic: it

recreates the equations that govern UAV fly, how it

reacts to external factors such as air density,

turbulence, wind shear, cloud, precipitation, etc. In

contrast, an exploration simulator focuses on

coordination logic, assuming that external factors are

already managed. It represents the exploration at a

different scale, which depends on the spatial and

temporal resolution needed to detect the target, and

recreates at that scale obstacles and target

distribution. Consequently, in the environment the

basic UAV movements and collision avoidance are

simulated for the specific purpose of exploration.

Figure 1 shows a simplified representation of the

environment with the available elements. For a better

granularity and without loss of generality, the search

problem is formulated by discretizing the

environment into a lattice of cells. In the

environment, a single UAV, or drone, is represented

by a disc with an inner arrowhead. An obstacle or a

target usually covers many cells. In figure, each

obstacle-cell is black, whereas each targeted cell is

colored. A targeted cell can either be discovered, i.e.,

yellow cell, or undiscovered, i.e. red cell. The color

intensity of a targeted cell represents the

quantity/presence of target, when applicable (e.g.,

fire, gas, etc.). Finally, a pheromone mark is

represented as a cluster of grey cells. The grey level

represents the pheromone intensity.

Figure 1: Environment and its elements: (from left to right)

drone, target, pheromone, and obstacle.

The temporal unit (tick) of the simulation

environment is set to a given number of seconds,

depending on the type of mission. On every tick, the

environment changes its current state to the next state,

according to the following rules. An obstacle-cell is

static. A targeted-cell can either be static (e.g.

landmine or dump) or dynamic (e.g., fire or gas). The

target dynamics is supplied as a sequence of frames

whose transition is ruled by a preset time frequency.

ICPRAM 2019 - 8th International Conference on Pattern Recognition Applications and Methods

838

This both avoid the effort of coding the equations

underlying the dynamics of targets and allows to use

real available frames to recreate a new scenario. The

grey level of a pheromone cell is dynamic, and it is

updated following an evaporation rule. A pheromone

mark is released by a drone when a target is found

(release rule). The drone position and direction is

dynamic and set according to exploration and

coordination rules. For a given type of mission, all

rules can be parametrically adapted by an

evolutionary algorithm which improves the overall

quality of the search process.

Specifically, Figure 2 shows a drone model with

the related parameters. The simulator takes into

account the drone cruise speed, acceleration, angular

speed, battery duration, drone size, sensing angle and

sensing radius.

Figure 2: Drone model with parameters.

Figure 3 shows the formation rules, based on

Reynolds’ flocking (1987): rules of alignment,

separation, and cohesion. Alignment aligns the

drones heading to the average heading of nearby

agents (flock mates). Separation keeps a large

formation by maintaining a minimum distance among

flock mates. Cohesion directs each agent towards the

center of the flock mates. A global angle of vision and

different ranges of radius characterize the three areas

of influence.

Figure 3: Formation rules (flocking).

Figure 4 shows a 3D shape of a single pheromone

mark, where (x,y) is the environment and z is the

pheromone intensity. A mark is modelled as a

truncated cone determined by the parameters

radiusTop, radiusDown, initial pheromone intensity.

When overlapping, pheromone marks can aggregate

in pheromone tracks. The track evaporates over the

time: every tick, its intensity is linearly reduced by a

given amount (deltaEvaporate).

Figure 4: Pheromone mark (stigmergy).

In the literature, the indirect communication

mechanism based on pheromone is called stigmergy

(Cimino et al. 2015a): an agent’s action produces a

mark which, in turn, incites another action, which

produces another mark, and so on. In the proposed

stigmergy, when a drone detects a new targeted cell,

the drone releases a pheromone mark on the location

of the sensed target. Pheromone acts as an attractive

potential on neighboring drones. While unknown

targets are sensed, additional pheromone marks are

released by flock members, thus enabling an

incremental positive feedback up to completion of all

targets in the proximity of the initial target. After a

certain time, the pheromone intensity cannot be

reinforced, and in practice disappears.

Figure 5 shows the collision avoidance model.

Figure 5: Collision avoidance.

Adaptive Exploration of a UAVs Swarm for Distributed Targets Detection and Tracking

839

The drone obstacle vision is set via two

parameters, i.e., collision.vision and collision.angle,

creating a circular sector area, whose vertex is

centered on the drone. When an obstacle or another

drone is detected in the collision area, the drone

changes its heading and speed to avoid the obstacle.

The area that will be occupied by the drone in the next

tick can be easily calculated via its velocity and its

possible headings. Thus, the multiple drones can be

accordingly organized in the current instant so as to

avoid overlapping with drones and obstacles at the

next temporal tick.

Overall, the swarm logic at each tick can be

summarized by the following pseudocode:

function SwarmSearch(environment)

tick = 0; targetsFound = 0;

evaporate(pheromone);

foreach drone d in swarm

if targets in d.sensing then

markTargetsFound(targets);

releasePheromone(targets.position);

targetsFound = targetsFound + 1;

endif

if obstacles in d.vision then

turnAway(d.heading, obstacles);

elseif pheromone in d.sensing then

turnTowards(d.heading,pheromone);

elseif flockmates in d.flock then

turnForFlocking(d.heading, d.flock);

else

turnForRandomWalk(d.heading, wiggle);

endif

moveForward(drone);

endfor

tick = tick + 1;

while (targetsFound < targetsThreshold)

or (tick = maxSearchTime);

return {tick, targetsFound};

function QualityMeasure(environment)

if targets.dynamic then

totTargets = 0;

foreach frame in environment

targets = SwarmSearch(environment);

totTargets += targets/targetsFrame;

endfor

return totTargets/numFrame;

else

tick = SwarmSearch(environment);

return tick;

endif

To adapt the swarm behavior to the environment,

the quality of the process is measured and optimized

by using the Differential Evolution (DE) algorithm

(Cimino et al. 2015b). Specifically, let K be the

number of adaptive parameters of the mission. In DE

a solution is represented by a real K-dimensional

vector. The overall quality of the process is optmized

either by minimizing the time needed for discovering

the given targets (static targets), or by maximizing the

average number of discovered targets in the overall

search time (dynamic targets). The adaptation process

is an intrinsic part of the swarm: in some sense only

after the optimization the initial set of UAVs becomes

a swarm, i.e., an effective organism specialized for

the type of mission.

3 SCENARIOS AND RELATED

TECHNOLOGICAL SETTINGS

In this section, the scenarios used and the various

quality measurements are illustrated. Table 1

summarizes the main features of each scenario. The

first three scenarios are static: Illegal Dump is based

on the abusive trash map in Paternò, Italy (Trashout

2018); Rural Mine and Urban Mine are based on

publicly available data of landmines in areas near

Sarajevo, in Bosnia-Herzegoniva (See-demining,

2018). The remaining scenarios are dynamic: Fire

Tracking comes by a propagation model developed

by the Northwestern University (Wilensky, 2018);

S Leak is based on a sour gas accident occurred in

December 2003, in Chongqing City, northeastern

Sichuan Gas Field, China (Ma Q.C. et al. 2011); LPG

Leak, is based on an accident occurred in June 2009

in Viareggio, Italy, and involving an LPG railcar

rupture in a congested urban area (Pontiggia et. al.

2011).

Table 1: Characteristics of each scenario.

Scenario

Area size

(m × m)

Targets

Animation

No of

frames

Illegal Dump

400 × 400

0 min.

Rural Mine

400 × 400

0 min.

Urban Mine

400 × 400

0 min.

Fire Tracking

1400 × 1400

20 min.

S Leak

4816 × 4400

48 min.

LPG Leak

500 × 300

4 min.

To show the environmental complexity, Figure 6

and Figure 7 show the satellite map used for Illegal

Dump, and the corresponding initial vector image

represented in the simulation environment,

respectively. Here, obstacles (buildings and trees) are

represented in black, whereas targets are represented

as red points. Drones, represented as purple triangles,

are placed at the corners and are oriented towards the

center of the area. Figure 8 shows another scenario,

Urban Mine, during the search process. Here, the

ICPRAM 2019 - 8th International Conference on Pattern Recognition Applications and Methods

840

targets found are represented in yellow. A large

pheromone cloud is clearly visible in the center of the

area, where a higher concentration of targets has

attracted a relevant number of drones.

Figure 6: Illegal Dump, satellite image (Google Maps ©).

Figure 7: Illegal Dump, vector image.

The environmental characteristics have been

considered for the technical specifications of the

commercially available UAV. Table 2 and Table 3

show, respectively, the technical specifications of the

drone “Dji Matrice 200” (www.dji.com/matrice-200-

series), and the sensing equipment for each scenario.

Such technology has been selected on the basis of

gained knowledge and skill over a number of projects

using UAV technology for environmental monitoring

and surveillance.

Figure 8: Urban Mine, simulator image with pheromone.

Table 2: Technical specifications of the commercial drone

“Dji Matrice 200”.

Parameter

Value

Radius

0.3 m

Max speed

17 m/s

Max acceleration

4.4 m/s

Max angular speed

2.6 rad/s

Max angular acceleration

6.9 rad/s

Battery duration

24-38 min

Obstacle vision distance

3-30 m

Obstacle vision angle

60°

Table 3: Technical specifications of the sensing equipment,

for each scenario.

Scenario

Cruise

speed

(m/s)

Sensing

technology

Sensor model

Sensing

radius

Illegal

Dump

Visual +

Thermal

Dji Zenmuse

XT2

5 m

Rural

Mine

Visual +

Thermal

Dji Zenmuse

XT2

5 m

Urban

Mine

Visual +

Thermal

Dji Zenmuse

XT2

5 m

Fire

Tracking

Visual

Dji Zenmuse

XT2

36 m

Leak

Electro-

chemical

Dräger

X-am 5600

0 m

LPG

Leak

Resistive

NiTiO

0 m

Specifically, the sensing technology proposed for:

(i) Illegal Dump, is based on (Persechino et al. 2010)

Adaptive Exploration of a UAVs Swarm for Distributed Targets Detection and Tracking

841

and (Lega et al. 2012); (ii) Rural and Urban Mine, is

based on (Rodriguez et al. 2014); (iii) Fire Tracking,

is based on (Cruz et al. 2016). The Sensor proposed

for H2S Leak and LPG Leak is based on (Neumann

et al. 2013) and (Chaudhari, 2018), respectively.

4 EXPERIMENTAL RESULTS

The environment and the coordination logic are

implemented on NetLogo, a leading simulation

platform for swarm intelligence (ccl.northwestern

.edu/netlogo). The adaptation module is implemented

on MATLAB ©, a numerical optimization framework

(www.mathworks.com). The source code of the

integrated system, called Sciadro 3.1, together with

the scenarios, has been publicly released on the

Github platform (Cimino et al. 2018).

As a pilot example, Figure 9 and Figure 10 show

two frames of the Fire Tracking scenario. Here, the

pheromone clouds clearly show that the swarm is

tracking the fire evolutions.

Figure 9: Fire Tracking, simulation frame, tick = 1013.

Figure 11 shows the number of targeted cells found

(%) against time (sec). The plot indicates a constant

trend of targets found per second, up to about 95%.

Since this is commonly a point of trend variation, to

shorten the simulation duration the targetThreshold

value is set to 95% in the function SwarmSearch, for

static scenarios. For a dynamic scenario, the

maxSearchTime, i.e., the frame period, can be

calculated as the ratio between targets animation and

number of frames. For example, in the Fire Tracking

scenario, it is set to 2060/5 = 240 sec.

Figure 10: Fire Tracking, simulation frame, tick = 1277.

Thus, for static scenarios the quality measure is the

time needed for the target threshold, whereas for

dynamic scenarios it is the average percentage of

targets discovered in each frame. The purpose of the

DE is to find the parameters minimizing the quality

measure, namely the fitness.

More formally, given a simulated scenario , made

of: (i) simulation instants of time   



; (ii) a set of

drones , each drone having a dynamic position















; (iii) a set of targets   , each target

having a fixed position 



. The set of targets

already found 



  , at a given instant of time t,

is the set of targets for which it exists a time   

and a drone  such that the drone’s position allows

the detection of the target’s position (relationship

denoted as “”):





 











 

























(1)

The fitness of the static simulated scenario  is then

defined as the minimum instant of time for which





 has cardinality greater than or equal to 















 





 











 









(2)

In case of dynamic scenarios, the targets can change

every frame transition period P, i.e., 



,  

    , where  is the predefined

final instant of the simulation. The fitness of the

dynamic simulated scenario  is then defined as the

average percentage of targets discovered in all frames:



































(3)

ICPRAM 2019 - 8th International Conference on Pattern Recognition Applications and Methods

842

Figure 12 shows the average best fitness, over 10

trials, against the number of generations, for the Rural

Mine scenario.

Figure 11: Percentage of targets found against time.

The figure clearly shows the structural importance

of DE adaptation, since it improves the performance

by 27%. Table 4 shows the performance of 80 UAVs

swarm, adapted for each scenario, in terms of the 95%

confidence interval over 10 repeated trials. The

number of UAVs has been determined by setting

incremental values and assessing the impact on

performance. For example, table 5 shows the

performance of 20, 40, 60, 80 UAVs for Fire

Tracking, in terms of 95% confidence interval over

10 repeated trials.

Figure 12: DE-based adaptation, average best fitness

against number of generations.

Finally, to better show the effectiveness of the

proposed approach, the same parameters have been

set in the previous version, valid for static scenarios

only, published in (Alfeo et al. 2018), hereafter called

Sciadro 2.0. For this purpose, the sensing radius has

been set to 4. The results, in Table 6, clearly show that

the proposed version sensibly outperforms Sciadro

2.0.

Table 4: 80 UAVs swarm performance adapted for each

scenario.

Scenario

Performance

Illegal Dump

121.70 ± 4.75 sec.

Rural Mine

125.96 ± 8.90 sec.

Urban Mine

152.38 ± 5.25 sec.

Fire Tracking

99.88 ± 0.06 %.

S Leak

98.78 ± 0.17 %

LPG Leak

93.88 ± 0.28 %

Table 5: Fire Tracking: swarm performance for a different

number of adapted UAV.

No of UAVs

Performance

60.64 ± 2.06 %

90.36 ± 0.54 %.

98.43 ± 0.25 %

99.88 ± 0.06 %

5 CONCLUSIONS

The paper summarizes the design of a bio-inspired

approach for the coordination of UAVs swarm

involved in distributed targets detection and tracking.

The coordination logic includes a spatial self-

formation and a collaboration based on dynamic

potential field. Moreover, the swarm adapts its

parameters to the specific mission by using an

optimization algorithm. A simulation testbed has been

developed and publicly released, using real-world

UAV technology and scenarios. Experiments are

encouraging, since the proposed approach extends and

sensibly outperforms a similar approach in the

literature. To provide comparative results with other

approaches is considered a key investigation task for

future work.

Table 6: Comparative analysis of performance (sec.).

Scenario

Sciadro 2.0

Sciadro 3.1

Illegal Dump

363.20 ± 102.6

159.03 ± 5.35

Rural Mine

195.80 ± 49.60

193.43 ± 6.79

Urban Mine

303.00 ± 85.70

208.76 ± 5.27

ACKNOWLEDGEMENTS

This work was carried out in the framework of the

SCIADRO project, co-funded by the Tuscany Region

(Italy) under the Regional Implementation

Adaptive Exploration of a UAVs Swarm for Distributed Targets Detection and Tracking

843

Programme for Underutilized Areas Fund (PAR FAS

2007-2013) and the Research Facilitation Fund

(FAR) of the Ministry of Education, University and

Research (MIUR).

REFERENCES

Alfeo, A.L., Cimino, M.G.C.A., De Francesco, N., Lega, M.,

Vaglini, G. "Design and simulation of the emergent

behavior of small drones swarming for distributed target

localization", Journal of Computational Science, Elsevier

Science, Vol. 29, Pages 19-33, 2018, (ISSN 1877-7503).

Aznar, F., Sempere, M., Pujol, M., Rizo, R., and Pujol, M.

(2014). Modelling oil spill detection with swarm drones.

In Abstract and Applied Analysis, volume 2014.

Hindawi.

Chaudhari, P. and Mishra, S., 2018. Ilmenite phase nickel

titanate nanowhiskers as highly sensitive LPG sensor at

room temperature. Journal of Materials Science:

Materials in Electronics, 29(1), pp.117-123.

Cimino, M.G.C.A., Lazzeri, A., Vaglini, G. "Combining

stigmergic and flocking behaviors to coordinate swarms

of drones performing target search", in The Sixth

International Conference on Information, Intelligence,

Systems and Applications (IISA), pp. 1-6, IEEE, 2015.

Cimino, M.G.C.A., Lazzeri, A., Vaglini, G. "Improving the

analysis of context-aware information via marker-based

stigmergy and differential evolution." International

Conference on Artificial Intelligence and Soft

Computing (ICAISC). Springer, Cham, 2015.

Cimino, Mario G.C.A., Monaco M., Piscione P., Alfeo L.

(2018). Github platform, Sciadro 3.1 repository,

(https://github.com/mlpi-unipi/drones-swarm). Accessed

November 10, 2018.

Cruz, H., Eckert, M., Meneses, J., Martínez, J.-F. "Efficient

forest fire detection index for application in unmanned

aerial systems (UASs)." Sensors 16, no. 6 (2016): 893.

Howden, D. J., “Fire tracking with collective intelligence

using dynamic priority map”, in IEEE Congress on

Evolutionary Computation (CEC), 2013. IEEE, 2013, pp.

2610-2617.

Lega, M., J. Kosmatka, C. Ferrara, F. Russo, R. M. A. Napoli,

and G. Persechino. "Using advanced aerial platforms and

infrared thermography to track environmental

contamination." Environmental Forensics 13, no. 4

(2012): 332-338.

Ma Q.C., Zhou L.B. “CFD simulation study on gas

dispersion for risk assessment: a case study of sour gas

well blowout”. Safety Science 2011;49(8):1289–95.

Maza, I., Ollero, A., Casado, E., and Scarlatti, D. (2015).

Classication of multi-UAV architectures. In Handbook

of unmanned aerial vehicles, pages 953-975. Springer.

McCune, R., Purta, R., Dobski, M., Jaworski, A., Madey, G.,

Madey, A., Wei, Y., and Blake, M. B. (2013).

Investigations of dddas for command and control of uav

swarms with agent-based modeling. In Simulation

Conference (WSC), 2013 Winter, pages 1467-1478.

IEEE.

Meng, W., He, Z., Su, R., Shehabinia, A. R., Lin, L., Teo, R.,

and Xie, L. (2014). Decentralized control of multi-UAVs

for target search, tasking and tracking. IFAC Proceedings

Volumes, 47(3):10048{10053.

Neumann, P.P., Asadi, S., Bennetts, V.H., Lilienthal, A.J. and

Bartholmai, M., 2013. Monitoring of CCS areas using

micro unmanned aerial vehicles (MUAVs).

Persechino, G., P. Schiano, M. Lega, R. M. A. Napoli, C.

Ferrara, and J. Kosmatka. "Aerospace-based support

systems and interoperability: The solution to fight illegal

dumping." WIT Transactions on Ecology and the

Environment 140 (2010): 203-214.

Persechino, G., P. Schiano, M. Lega, R. M. A. Napoli, C.

Ferrara, and J. Kosmatka. "Aerospace-based support

systems and interoperability: The solution to fight illegal

dumping." WIT Transactions on Ecology and the

Environment 140 (2010): 203-214.

Pontiggia, M., Landucci, G., Busini, V., Derudi, M., Alba,

M., Scaioni, M., Bonvicini, S., Cozzani, V., Rota, R.,

2011. “CFD model simulation of LPG dispersion in

urban areas”. Atmospheric Environment 45, 3913e3923.

Qingchun, Ma, Laibin, Zhang, “CFD simulation study on gas

dispersion for risk assessment: a case study of sour gas

well blowout”, Safety Science, 2011; 49(8):1289–95.

Reynolds, C. W. “Flocks, herds and schools: A distributed

behavioral model”. In ACM SIGGRAPH computer

graphics, volume 21, pages 25-34 (1987), ACM.

Rodriguez, J., C. Castiblanco, I. Mondragon, and J. Colorado.

"Low-cost quadrotor applied for visual detection of

landmine-like objects." In Unmanned Aircraft Systems

(ICUAS), 2014 International Conference on, pp. 83-88.

IEEE, 2014.

Rodriguez, J., Castiblanco, C., Mondragon, I., Colorado, J.:

Low-cost quadrotor applied for visual detection of

landmine-like objects. In: International Conference on

Unmanned Aircraft Systems (ICUAS), pp. 83–88. IEEE

(2014).

See-demining, Mine detection Center for the Region of

South-Eastern Europe, accessed February 2018,

http://www.see-demining.org/main.htm,

Senanayake, M., Senthooran, I, Barca, J. C., Chung, H.,

Kamruzzaman, J., Murshed, M., “Search and tracking

algorithms for swarms of robots: A survey”, Robotics

and Autonomous Systems 75 (2016) 422–434.

Trashout, Trash Maps Data, accessed February 2018,

https://www.trashout.ngo/trashmap/?ref=trashout.me.

Whitehead, K., Hugenholtz, C. H., Myshak, S., Brown, O.,

LeClair, A., Tamminga, A., Barchyn, T. E., Moorman,

B., and Eaton, B. (2014). Remote sensing of the

environment with small unmanned aircraft systems

(uass), part 2: Scientic and commercial applications.

Journal of unmanned vehicle systems, 2(3):86-102.

Wilensky, U. Center for Connected Learning and Computer-

Based Modeling, Northwestern University, Evanston, IL.

NetLogo Fire models library, November 2018,

http://ccl.northwestern.edu/netlogo/models/Fire.

ICPRAM 2019 - 8th International Conference on Pattern Recognition Applications and Methods

844