-
Terminal Airspace Modelling for Unmanned AircraftSystems
Integration
Aaron McFadyen1 and Terry Martin2
Abstract— This paper considers the problem of
integratingunmanned aircraft into low altitude airspace above
urbanenvironments, including major terminal areas and
helicopterlanding sites. A simple set of data-driven modelling
techniquesare used to explore, visualise and assess existing air
traffic in amanner more informative to the unmanned aircraft
community.First, low altitude air traffic data sets (position
reports) areanalysed with respect to existing exclusion/no-fly
zones. Second,an alternative geometric approach to defining and
comparingvarious exclusion zones is derived based on set theory.
Theanalysis is applied to a region of south-east
Queensland,Australia including Brisbane International Airport and
threehelicopter landing areas. The results challenge some of
thecurrent unmanned aircraft regulations, and should help
tomotivate a more rigorous scientific approach to safely
integrateunmanned aircraft.
I. INTRODUCTION
Unmanned Aircraft Systems (UAS) constitute one ofthe fastest
growing sectors of aviation in many countries.Unmanned aircraft
offer an alternate or new solution to alarge number of tasks within
a diverse set of commercialapplications from remote sensing in
disaster response andagriculture, to infrastructure inspection in
mining and realestate [1]. As such, there is an increased demand to
allowunmanned aircraft regular access to both regional and
urbanairspace, including in and around major cities.
Integrating unmanned aircraft into an urban airspace
envi-ronment is difficult. Firstly, most major cities are
co-locatedwith a major international airport, general
aviation/trainingaerodromes, and helicopter landing sites
(heliports) for me-dia, emergency services and police operations.
This meansthere is a large volume of diverse air traffic (fixed
androtary wing) extending from surface level. Secondly, thereis
generally an increased population density as more peoplemigrate to
major trade centres. This means that there aremore people located
below an already crowded airspace filledwith low-level operations.
Collectivity, this raises consid-erable safety concerns when
considering the integration ofunmanned aircraft, particularly
regarding collision risk. Byfurther increasing the volume and
diversity of air traffic, therisk of collision (mid-air or ground)
increases, along withthe potential for aviation related
fatalities.
As a result, many regulatory bodies have stipulated
restric-tions on unmanned aircraft operations within urban
airspace.One of the hazards considered is public safety, so
unmannedflights are confined to operate at a minimum proximity
to
1,2 are with the Science & Engineering Faculty, Queensland
University ofTechnology, Brisbane, Australia.
[email protected]
Fig. 1. Example airspace classifications in south-east
Queensland, Aus-tralia. The image shows some of the military
airspace (�), training airspace(civil and military) (�), controlled
airspace boundaries (−) and a declareddanger area due to unmanned
aircraft operations (�).
persons not involved in the operation. This means operat-ing
unmanned aircraft for many media events, or aroundsports and music
venues, is either prohibited or requiresspecial exemptions. Another
hazard relates to the expectedair traffic density directly, such
that unmanned operationsmay be restricted around key locations or
geographic regions.For example, many countries have established
predefinedexclusion or no-fly zones around aerodromes and
helicopterlanding sites. Access to these exclusion zones depends
onwhether the aerodrome is towered, and the type of
unmannedoperators certificate held.
In defining exclusion zones, there is often minimal or
notransparency in how such regions are determined. From anaviation
safety regulators perspective, it would be reason-able to establish
very conservative exclusion zones. Froman unmanned aircraft
operators perspective, it would bedesirable to have a minimal and
liberal set of exclusionzones. It is difficult to satisfy both
stakeholders given theirfundamental motivations. Either there is
missed commercialopportunity through overly strict regulations that
do notsignificantly reduce safety, or there is more opportunity
(andtherefore more operators) under relaxed regulations whichcould
compromise system safety. In light of these concerns,the
contributions of this paper are:� A spatial analysis of low-level
urban air traffic near a ma-
jor international airport and emergency service
helicopterlanding sites. The airspace region encompasses parts
ofBrisbane, Australia.
� A data-driven and risk-inspired approach to
determiningunmanned aircraft exclusion (no-fly) zones near
terminalareas and helicopter landing sites.
-
This paper is structured as follows. Section II provides
aliterature review of useful terminal air traffic modelling
pa-rameters and approaches in the context of unmanned
aircraftintegration. Section III details the air traffic data,
includingdata mining (extraction) methods, data limitations and a
dataanalysis with respect to current exclusion zone
regulations.Section IV explores an alternate data-driven approach
todefining unmanned aircraft exclusion zones. Conclusions
andfurther work are offered in Section V.
II. BACKGROUND
Although aviation related fatalities have been decliningover the
years, operations in and around terminal areasremains hazardous
[2]. Recent data shows that over halfthe recorded aviation
fatalities occur during the take-off,initial climb, approach and
landing phases of flight [3].Given the increase in conventional air
traffic and escalatedcommercial pressure to operate unmanned
aircraft in urbanareas, maintaining a high level of aviation system
safety inthe terminal area will remain a priority.
To this end, most regulators have taken a conservativeapproach
and established a set of exclusion or no-fly zones1
for unmanned aircraft around aerodromes and heliports. Akey
component in defining any such exclusion zoning is theprovision of
appropriate air traffic modelling and analysistechniques. Such
techniques should also include robust risk-centric metrics, so that
the merits of various exclusion zoneproposals can be examined. The
overall goal is to then shrinkthe exclusion zone boundaries without
violating any TargetedLevel of Safety (e.g. 1.5 x 10−8 to 5 x 10−9
fatal accidentsper system flight hour.).
Modelling terminal air traffic has been tackled for manyyears.
The diversity in modelling approaches often stemsfrom the intended
application of the derived models. Simplis-tic models do not
consider the air traffic directly, and insteadrely on the airspace
classification and approach/departureprocedures to give an
indication of the expected traffic typeand behaviour in each
region. An example of this type ofmodel is shown in Fig. 1, and is
often used to help flightplanning for unmanned aircraft
operations.
More rigorous approaches attempt to model the risk ofcollision
in a predefined volume as a function of expectedaircraft behaviour.
The variation in pairwise aircraft trajec-tories (often within air
routes) is used to derive a conflictprobability which acts as a
proxy for risk. Several establishedcollision risk models have been
used [4], [5] with moreadvanced approaches emerging [6], [7].
Although useful,many of these models cannot be applied to terminal
airspace[8], given the underlying assumptions relating to
aircraftbehaviour (constant velocity trajectories etc.).
In an attempt to capture a more global view of theaircraft
traffic behaviour in terminal regions, data-drivenmodelling
techniques have recently emerged. Using realtrack data (radar,
ADS-B etc.), modelling is often tackled
1Unmanned aircraft may only operate within these exclusion zones
withspecial permission.
from a flow monitoring or outlier detection perspective.
Forthese applications, the goal is to define common patternsor
traffic tubes, such that the approach and departure pathvariability
can be captured and used to discriminate betweennormal and abnormal
behaviours in real time. For example,terminal airspace models have
been derived using geometricprimitives [9], Fourier analysis [10],
Principle ComponentAnalysis [11], [12] and sequence alignment
techniques [13]with varying success. In some cases, very accurate
andcompact representations of major terminal air traffic routescan
obtained, but often at the cost of discarding less fre-quent (but
still important) flight tracks. From an exclusionzone modelling
perspective, this can be problematic if thediscarded flight tracks
are not sufficiently rare. Interestingly,despite the availability
and importance of track data aroundurban areas, limited effort has
been directed at developingrobust exclusion zone models from the
data. Ideally, anysuch model should support the comparison between
differentboundary choices in terms of collision risk, exclusion
zonearea and ease of implementation.
What is clear in the literature, is that there is no consensuson
the appropriate modelling technique to define sensibleterminal area
exclusion zones. In this initial paper, the goalis to then explore
different exclusion zone boundaries usingsome simple modelling
techniques and comparison metrics.Based on the above findings, this
paper proposes:� A data-driven geometric approach to analysing air
traffic
movement near terminal areas and helicopter landingsites, that
focuses on determining useful parameters todefine suitable unmanned
aircraft exclusion zones.
� A simple set of metrics to discriminate between candi-date
unmanned aircraft exclusion zones. The metrics areaimed at
quantifying the simplicity of the exclusion zoneshape, as well as
capturing the trade-off between reducingcollision risk and
optimising usable airspace.
III. AIR TRAFFIC DATA
A. Data Source
The air traffic data is taken from real position reportsrecorded
by the Australian Advanced Air Traffic System(TAAATS) used for
current Air Traffic Management inAustralia. The data is first
filtered to isolate all recordedflights in south east Queensland
over the summer of 2014-2015 (December-February). This region of
airspace is aninteresting case study as it includes the capital
city ofBrisbane and its associated international airport, two
majorgeneral aviation and training aerodromes, a military
airport,and a number of emergency service helicopter landing
sites.
Using aircraft identification tags and report timing,
theposition reports are then condensed into an ordered set
offlights or tracks. All flights with less than 3 position
reportsare then removed, before each flight trajectory is
re-sampledto provide 50 equally spaced data points. As a result
thedata consists of 64,307 flights with 16,292,100 unique
datapoints.
-
Longitude (deg)153.02 153.04 153.06 153.08 153.1 153.12 153.14
153.16 153.18 153.2
Latitude(deg)
-27.48
-27.46
-27.44
-27.42
-27.4
-27.38
-27.36
-27.34
-27.32
-27.3
-27.28Example Runway Boundary Zones and Traffic
Fig. 2. Example summer air traffic in Brisbane terminal area at
or below500 feet above ground. The region shows all position
reports (•) within5nm of the aerodrome reference point (ARP), both
runways, a runway-based exclusion zone (−) and the
take-off/departure splays (−/−−).
B. Terminal Areas and Helicopter Landing Sites
To reduce the data coverage and retain focus on theanalysis of
exclusion zones, the data set is further reducedby partitioning it
into two distinct regions. The first regionincludes the terminal
region under 500 feet and extending5nm from the aerodrome reference
point associated withBrisbane international airport. This region
contains 95,498position related data points. The second region
includes theairspace around three helicopter landing sites under
500 feet,and extending 3nm from each reference heliport
(helipad).This region contains 3,618 position related data
points.
The terminal region traffic is depicted in Fig. 2 alongwith some
current examples of exclusion zones. The leastrestrictive zone is
defined by a radial disc extending 3nmfrom the aerodrome reference
point. A more conservative,and perhaps realistic zone, is defined
by a closed contour(polygon) that has no boundary point less than
3nm fromeach runway threshold or aerodrome boundary (fence).
Whenusing the runway thresholds, this zone can be created
bygenerating 3nm radial discs from each threshold and tracinga
boundary around the overlapping polygons. The averagespeed of the
traffic in the region is approximately 120kts,with peak traffic
during the daylight hours and early eveningsof each day (see
Appendix).
The heliport region traffic is depicted in Fig. 3 alongwith an
example exclusion zone. The zone is determinedusing the same
approach as the runway-based exclusion zonefor terminal regions by
replacing the runway threshold witheach heliport. Of note,
individual 3nm radial exclusion zoneswould have to be considered if
Brisbane’s hospitals weremore sparsely distributed. The average
speed of the traffic inthe region is less than 50kts, with a
consistent flow of trafficthroughout each day. Reduced traffic is
also observed earlyin the morning.
For both airspace regions, these simple visualisationsillustrate
that there are a number of regions with very little
Longitude (deg)152.94 152.96 152.98 153 153.02 153.04 153.06
153.08 153.1 153.12
Latitude(deg)
-27.55
-27.5
-27.45
-27.4
Example Heliport Exclusion Zones and Traffic
Fig. 3. Example summer air traffic around three Brisbane
helicopter landingsites at or below 500 feet above ground. The
region shows all positionreports (•) and a 3nm radial exclusion
zone around each helipad (−−).The combined exclusion zones made up
from the closed contour formed bythe boundary of each radial
exclusion is also shown (−).
traffic. Given the duration of the data set, this can mean
asingle flight over 3 months in some cases. As such, someof the
existing exclusion zones may be overly conservative,reducing the
opportunities for unmanned aircraft to utiliselow-density airspace
around urban areas. With this in mind,the following section
proposes an alternative data-drivenapproach to explore candidate
exclusion zones in urban areas.
Remarks: The size of each aircraft or helicopter is
notconsidered in this analysis, and all point visualisations ofthe
actual aircraft are therefore not to scale. Additionally,some
position reports in each region do not have an altitudedata point.
These reports have been retained to consider theworst case
scenario.
IV. DATA-DRIVEN EXCLUSION ZONES
The data-driven exclusion zones for each landing area arederived
using some basic techniques common to set theoryanalysis. The basic
approach is to derive candidate exclusionzones by considering
compact convex and bounding (non-convex) polygons enclosing
different collections of positionreports contained around each
landing site. More relaxed(tighter) exclusion zones are then
derived by successivelyremoving position reports and recalculating
the polygons.
A. Notation
Each position report is denoted by a single point x(x, y),where
x and y are the latitude and longitude coordinatesof the point. The
convex hull conv(S) of a finite point setS ∈ R2 is the set of all
convex combinations of its points.The bounding polygon ∂S of a
finite point set S ∈ R2 isthe set of points in the closure of S ,
not belonging to theinterior of S .
-
Longitude (deg)153.02 153.04 153.06 153.08 153.1 153.12 153.14
153.16 153.18 153.2
Latitude(deg)
-27.48
-27.46
-27.44
-27.42
-27.4
-27.38
-27.36
-27.34
-27.32
-27.3Data-Driven Exclusion Zones and Traffic
Fig. 4. Example data-driven terminal area exclusion zones for
100%, 99%and 98% (−/•,−/•,−/•) of the original air traffic data
contained withina 5nm radius of the aerodrome reference point
(ARP). The dashed andsolid lines define the convex and bounding
polygons around the associatedposition reports. The minimum
Hausdorff distance between the convex andnon-convex polygons is
also depicted −/•.
B. Terminal Areas
For the terminal area, position reports x under 500 feetand
within 5nm of the aerodrome reference point a areconsidered as the
reference data set A0 such thatA0 = {xn ∈ R2 : ||xn−a||2 ≤ 5κ}, n =
{1, ..., N} (1)
where N = |A0| is the number of position reports andκ defines
the conversion constant from degrees to nauticalmiles. The first
set of candidate exclusion zones is found bycalculating the convex
hull CA0 and bounding polygon C̄A0of the reference data such
that
CA0 = conv(A0), C̄A0 = ∂A0 (2)Tighter or less restrictive
exclusion zones can then be calcu-lated by successively removing a
percentage of the outermost(most distant) position reports from the
reference data set.To find the outermost points, a distance metric
d̂ is definedsuch that for each point n in the dataset
d̂n = minv∈V
(||xn − vi||2) n = {1, ..., N} (3)
where v defines a vertex (edge) from the set of vertices Vof the
convex hull enclosing the terminal runway thresholds.
The points corresponding to the largest distances can thenbe
removed to create smaller data sets A1, ...,Ak, wherethe index of
the data set corresponds to the percentageof points removed from
the reference data set A0. Bythen re-applying (2), the convex CA1 ,
..., CAk and boundingpolygons C̄A1 , ..., C̄Ak can then be found.
The result is a setof candidate exclusion zones that eventually
contract to theconvex hull defined by the runway thresholds as k
increases(see Fig. 4).
Remarks: Informally, the convex hull can be defined byplacing a
taut rubber band around the entire set S . Convexhulls are found
using the Quickhull algorithm [14]. The
Longitude (deg)152.98 152.99 153 153.01 153.02 153.03 153.04
153.05 153.06 153.07 153.08
Latitude(deg)
-27.55
-27.5
-27.45
-27.4Data-Driven Exclusion Zones and Traffic
Fig. 5. Example data-driven helicopter landing site exclusion
zonesfor 100%, 95% and 90% (−/•,−/•,−/•) of the original air
traffic datacontained within a 3nm radius of the helicopter landing
sites. The dashedand solid lines define the convex and bounding
polygons around theassociated position reports. The minimum
Hausdorff distance between theconvex and non-convex polygons is
also depicted −/•.
bounding polygon is the set of outermost points in the setthat
when joined using straight line segments (edges) encloseall points
in the set. The set is typically non-convex, but canbe convex for
certain data sets.
C. Helicopter Landing Sites
For the helicopter landing sites, position reports x under500
feet and within 3nm of any of the helicopter landingpads h ∈
{h1,h2,h3} are considered as the reference dataset H0 such that
H0 = {xm ∈ R2 : ||xm − h||2 ≤ 3κ}, m = {1, ...,M},(4)
where M = |H0| is the number of position reports. Similarto the
case for the terminal area, the first set of candidateexclusion
zones is found by calculating the convex hull CH0and bounding
polygon C̄H0 of the reference data such that
CH0 = conv(H0), C̄H0 = ∂H0, (5)
Again, less restrictive exclusion zones can then be calculatedby
successively removing a percentage of the outermostposition reports
from the reference data set. To find theoutermost points in this
case, the same distance metric d̂
d̂m = minz∈Z
(||xm − zi||2) m = {1, ...,M} (6)
is used where z now defines a line from the set of linesZ
joining each helicopter landing zone reference point(helipad).
Using (2) again, a set of candidate exclusion zonesdefined by the
convex CH1 , ..., CHk and bounding polygonsC̄A1 , ..., C̄Hk can
then be found. For helicopter landing areashowever, the exclusion
zones eventually contract to theconvex hull defined by the
helicopter landing site referencepoints as k increases (see Fig.
5).
-
Fig. 6. Candidate data-driven exclusion zones (�,�,�) for both
terminalC̄Ak and heliport C̄Hk regions around Brisbane. Increasing
k is depictedfrom blue to red. Currently enforced exclusion zones
are also depicted (�).
D. Practical Application
Applying the data-driven approach to exploring
candidateexclusion zones reveals some interesting results. For both
theterminal and heliport regions, less restrictive exclusion
zonescould be adopted, without removing any of the original
datapoints. Therefore, the collision risk may not be effected
byexpanding the operational envelope for unmanned aircraft inthe
Brisbane region. However, this may reduce the separationbetween
aircraft, which may not be palatable to the regulator.
After removing some of the data points from each region,the
candidate exclusion zones start to take a more complexshape with a
reduced area. Of note, large regions becomeaccessible to the west
and east of the terminal region, andnorth and south of the heliport
region.
However, care must be taken in interpreting these results.As
data has been removed there are aircraft present outsideeach
exclusion zone, so it is unclear how the collision riskhas been
affected. A possible way to view the results is tothen consider a
relative risk-opportunity metric R such that
Rk = |1− āk/N̄k| (7)where āk is the normalised area of the
exclusion zone with kpercent of the data points removed, and N̄k is
the normalisednumber of enclosed position reports. The area and
numberof position reports are normalised using the original data
setand associated region area (3nm polygon for heliports and5nm
radial disc for terminal region). As the risk-opportunitymetric
value increases, there are more aircraft per unit surfacearea in
the candidate exclusion zone. Relatively speaking,this means that
there may be more opportunity (usable area)without a significant
increase in collision risk. The variationin Rk between the
candidate exclusion zones defined by thebounding polygons is
depicted in Fig. 7. Similar plots couldbe constructed for an
arbitrary zone to visualise the trade-offbetween increasing the
operational envelope and maintainingan acceptable risk of
collision.
An alternate and softer approach to assessing the utilityof the
data-driven exclusion zones is to consider their shape.This is
important from an implementation perspective, asexclusion zones
should be easy to understand and map (seeFig. 6). A metric that can
be used to measure how far two
Percentage Removed (%)0 5 10 15
Risk-O
pportunity
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9Example Risk-Opportunity
Fig. 7. Example risk-opportunity curve for bounding polygon
exclusionzones for both terminal C̄A (◦) and heliport C̄H (�)
regions. The riskopportunity for the original data set and
exclusion zone is also included.
Percentage Removed (%)0 1 2 3 4 5 6 7 8 9 10
Hau
sdorffDistance
(dH)
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
0.055
0.06
0.065Data-Driven Exclusion Zone Similarity
Fig. 8. Example Hausdorff distance variation between each convex
C(·)and bounding C̄(·) exclusion zone pair for both terminal (◦)
and helicopterlanding areas (�).
subsets of a metric space are from each other is the
Hausdorffdistance dH defined as
dH(C, C̄) = max{supx∈C
infx̄∈C̄
d(x, x̄), supx̄∈Ĉ
infx∈C
d(x, x̄)} (8)
where x ∈ C, x̄ ∈ C̄ and d(x, x̄) = ||x− x̄||2. Applied to
thedata-driven exclusion zones, dH = 0 if and only if C and C̄have
the same closure. That is, the union of the points in eachexclusion
zone and its associated boundary are the same.This means that a
small Hausdorff distance suggests theshape of the convex and
bounding exclusion zones are verysimilar. This is desirable from an
implementation standpoint,as the closer the exclusion is to a
convex shape, the morestraight forward the implementation and
enforcement islikely to be. For the terminal and helicopter landing
areas,the convex and bounding exclusion zones converge as thenumber
of data points are removed. This makes sense as thetraffic density
increases whilst the track shape become lessdiverse with reduced
outliers (see Fig. 8).
-
V. CONCLUSIONS
Determining adequate exclusion zones for unmanned air-craft in
an urban environment subject to frequent conven-tional air traffic
is an important task. Existing regulationscurrently limit where
unmanned aircraft may operate, but insome cases these can be overly
strict without significantly re-ducing the risk of collision. This
paper presented an alternatedata-driven geometric approach to
defining exclusion zones,where each zone can be tailored to each
unique environmentbased on the risk appetite of the regulating
body. To thisend, a number of simple metrics were also proposed to
helpvisualise the trade-off between increasing usable airspace
andmaintaining an acceptable level of collision risk.
This work constitutes a particularly unique contributiontoward
the integration of unmanned aircraft into an urban en-vironment,
and provides a good foundation in which to stemfurther research and
development. Extended research willexplore the inclusion of more
comprehensive collision riskmetrics that explicitly consider
aircraft speed and frequencyto compare and refine different
exclusion zone boundaries.Alternatively, the affect of placing
additional buffer zonesor separation boundaries on the data-driven
exclusion zonescould be studied. Collectively, it is hoped that
advancingthis work will provide a sensible and realistic framework
inwhich to optimise the placement and structure of exclusionzones
for unmanned aircraft.
APPENDIX I
Example aircraft speed and report time (hours) histogramfor 3
months of aircraft position reports taken from a regionof south
east Queensland, Australia.
Fig. 9. Example speed distribution of traffic contained within
the runway-based exclusion zone around Brisbane terminal (�) and
heliport-basedexclusion zone around helicopter landing sites
(�).
ACKNOWLEDGEMENTS
The authors would like to thank the Australian ResearchCentre
for Aerospace Automation (ARCAA) at Queens-land University of
Technology, the Queensland Governmentthrough the Department of
Science, Information Technology,Innovation and the Arts (DSITI),
Thales Australia and the
Fig. 10. Example time distribution of traffic contained within
the runway-based exclusion zone around Brisbane terminal (�) and
heliport-basedexclusion zone around helicopter landing sites
(�).
Civil Aviation Safety Authority (CASA) for the funding tosupport
this work. The authors also acknowledge AirservicesAustralia for
supplying the air traffic data.
REFERENCES[1] K. Valavanis, “Unmanned aircraft systems: the
current state-of-the-
art,” Springer, 2013[2] International Civil Aviation
Organisation, “ICAO Safety Report,”
Technical Report, 2014[3] Boeing Commercial Airplanes,
“Statistical summary of commercial
jet airplane accidents,” Technical Report, Aug. 2015[4] P.
Reich, “Analysis of long range air traffic systems separation
standards I, II, III,” The Journal of Navigation, vol. 19, pp.
88-98,pp. 169-186, pp. 331-347, 1966
[5] D. Anderson and X. Lin, “A collision risk model for a
crossing trackseparation methodology,” The Journal of Navigation,
vol. 49, pp. 337-349, 1996
[6] M. Prandini, J. Hu, J. Lygeros, and S. Sastry, “A
probabilistic approachto aircraft conflict detection,” IEEE Trans.
Intelligent TransportationSystems, vol. 1, no. 4, pp. 199-220, Dec.
2000
[7] Y. Matsuno, and T. Tsuchiya, “Probabilistic conflict
detection in thepresence of uncertainty,” Air Traffic Management
and Systems, pp. 17-33, Feb. 2014
[8] M. Fujita, “Collision risk model for independently operated
homo-geneous air traffic flows in terminal area,” Electronic
NavigationResearch Institute Papers, no. 130, May 2013
[9] J. Cózar, F. Sáez and E. Ricaud, “Radar track segmentation
with cubicsplines for collision risk models in high density
terminal areas,” Proc.Institution of Mechanical Engineers, Part G:
Journal of AerospaceEngineering, vol. 229, no. 8, pp. 1371-1383,
2015
[10] R. Annoni and C.Forster, “Analysis of aircraft trajectories
using fourierdescriptors and kernel density estimation,” IEEE Int.
Conf. IntelligentTransportation Systems, pp. 1441-1445, Sep.
2012
[11] M. Gariel, A. Srivastava and E.Feron, “Trajectory
clustering and anapplication to airspace monitoring,” IEEE Trans.
Intelligent Trans-portation Systems, vol. 12, no, 4, pp. 1511-1523,
2011
[12] E. Salaün, M. Gariel, A. Vela and E.Feron, “Aircraft
proximity mapsbased on data-driven flow modeling,” AIAA Guidance,
Control andDynamics, vol. 35, no, 2, pp. 563-577, Mar.-April
2012
[13] A. Mcfadyen, T. Martin, M. O’Flynn and D. Campbell,
“Air-craft trajectory clustering techniques using circular
statistics,” IEEEAerospace. Conference, (accepted), March 2016
[14] C. Barber, D. Dobkin and H. Huhdanpaa, “The quickhull
algorithmfor convex hulls,” ACM Trans. on Mathematical Software,
vol. 22,no. 4, pp. 469-483, 1996