Available online at www.sciencedirect.

com
Available
Available online
online at
at www.sciencedirect.com
www.sciencedirect.com
ScienceDirect
ScienceDirect
ScienceDirect
Available online at www.sciencedirect.com
Transportation Research Procedia 00 (2020) 000–000
Transportation Research Procedia 00 (2020) 000–000
Transportation Research Procedia 00 (2020) 000–000 www.elsevier.com/locate/procedia
ScienceDirect www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia

Transportation Research Procedia 52 (2021) 645–652

23rd
23rd EURO Working Group on Transportation Meeting, EWGT 2020, 16-18 September 2020, Paphos,
23rd EURO
EURO Working
Working Group
Group on
on Transportation Meeting,
Meeting, EWGT
Transportation Cyprus EWGT 2020,
2020, 16-18
16-18 September 2020, Paphos,
September 2020, Paphos,
Cyprus
Cyprus
Linking
Linking the microscopic traffic flow mechanics with the macroscopic
Linking the
the microscopic
microscopic traffic
traffic flow
flow mechanics
mechanics with
with the
the macroscopic
macroscopic
phenomena
phenomena by exploiting class-type traffic information retrieved from
phenomena by exploiting class-type traffic information retrieved from
by exploiting class-type traffic information retrieved from
online traffic
online traffic maps
traffic maps
maps
online
Vana
Vana Gkaniaaaa and Loukas Dimitriou a,*
a,
Vana Gkania
Gkania andand Loukas
Loukas Dimitriou
Dimitrioua,**

a
LaBfor Transport Engineering, Department of Civil and Environmental Engineering University of Cyprus,75 Kallipoleos Str., P.O. Box 20537, 1678 Nicosia,
aLaB for Transport Engineering, Department of Civil and Environmental Engineering University of Cyprus,75 Kallipoleos Str., P.O. Box 20537, 1678 Nicosia,
a
LaB for Transport Engineering, Department of Civil and Environmental Engineering
Cyprus University of Cyprus,75 Kallipoleos Str., P.O. Box 20537, 1678 Nicosia,
Cyprus
Cyprus

Abstract
Abstract
Abstract
The conversion from single-entity level characteristics of traffic flow to comparable system-level characteristics shaped a new era for traffic
The
The conversion
conversion from
from single-entity
single-entity level characteristics of
of traffic flow
flow to
to comparable system-level characteristics shaped aa newnew era for traffic
monitoring and control. Since then level characteristics
landmark studies explored traffic
network-levelcomparable system-level
traffic flow relationshipscharacteristics
across entireshaped
urban networks era or
forregions
traffic
monitoring
monitoring and
and control.
control. Since then
then landmark studies explored network-level traffic
traffic flow relationships across entire urban networks or or regions
and cities, mainly based Since
on simulationlandmark
data butstudies
also explored network-level
with empirical data. Although, flow relationships
the ability across
to observe andentire urban
monitor thenetworks
traffic state regions
of the
and
and cities,
cities, mainly based
based on simulation data but also with
with empirical data. Although, the
the ability to
to observe and
and monitor the traffic
trafficasstate of
of the
system on mainly
a network-wide onlevel
simulation
depends dataonbut
thealso
availabilityempirical data.traffic
of existing Although, ability
surveillance systems, observe
adequately monitor
deployed the such state
to cover thea
system
system on aa network-wide
onnetwork.
network-wide level depends
level this
depends on the
on we availability
the propose
availability of
of existing
existing traffic
traffic asurveillance
surveillance systems,
systems, adequately
adequately deployed
deployed such as
as to
such (MFD) to cover
cover a
complete To overcome deficit, a method to estimate network's Macroscopic Fundamental Diagrams usinga
complete
complete network.
network. To overcome
To overcome this deficit,
this deficit, we
weandpropose
propose a method
a method to estimate
to estimate a network's Macroscopic Fundamental Diagrams (MFD) using
traffic flow mechanics at the microscopic level exploiting class-type traffic ainformation
network's Macroscopic Fundamental
that can be obtained Diagrams
from online (MFD)
traffic maps.using
This
traffic
traffic flow
flow mechanics at
at the
the microscopic
mechanicsdepicted microscopic level
level and exploiting
and based
exploiting class-type
class-type traffic information
traffic techniques,
informationablethat can
that to
can be
be obtained
obtained from
from online
online traffic
traffic maps.
maps.ofThis
This
valuable information on maps is extracted on image processing simultaneously perform discretization the
valuable
valuable information
information depicted
depicted on
on maps
maps is
is extracted
extracted based
based on
on image
image processing
processing techniques,
techniques, able
able to simultaneously
totraffic
simultaneously perform
perform discretization
discretization of
of the
the
urban space-and the road network therein- in seamless pixels and further capture the color-coded information in a suitable data structure
urban
urban space-and
space-and the
the road
road network
network therein-
therein- in
in seamless
seamless pixels and
and further capture the color-coded traffic information in
in aa suitable data structure
valuable for meta-analysis. Then, the fundamental trafficpixels furtherare
flow mechanics capture the connecting
used for color-codedthe traffic information
captured suitable
pixels properties withdata structure
macroscopic
valuable
valuable for meta-analysis.
for meta-analysis. Then,
Then, the fundamental traffic flow mechanics are used for connecting the captured pixels properties with
with macroscopic
traffic phenomena, especially withthe
thefundamental
well-defined traffic flow The
(MFDs). mechanics
validityareofused for connecting
the method is testedthebycaptured
comparing pixels
the properties
estimated MFDs macroscopic
to ground-
traffic
traffic phenomena,
phenomena, especially with
with the well-defined (MFDs). The validity of the method
method is tested
tested by comparing the estimated
estimated MFDs to
to ground-
truth MFD obtainedespecially
using empirical the well-defined
data from loop (MFDs).
detectors.The Thevalidity
results of
arethe
providing is
valuable by comparing
evidence theoperational
on the MFDs ground-
characteristics of
truth
truth MFD obtained using empirical data from loop detectors. The results are providing valuable evidence on the operational characteristics of
large MFD
urban obtained using
areas, while at aempirical data stage
meta-analysis from loop
it wasdetectors. The results
able to capture are providing
spatio-temporal valuableofevidence
phenomena on the operational
urban mobility, characteristics
like concentration, hysteresisof
large
large urban
urban areas,
areas, while
while at aa meta-analysis stage it
it was
was able to
to capture spatio-temporal phenomena of
of urban mobility,
mobility, like concentration, hysteresis
and homogeneity. Sinceatonline
meta-analysis
traffic mapsstageprovide able
almost capture
global spatio-temporal phenomena
coverage the proposed method urban
is practically like concentration,
feasible and offershysteresis
a novel
and
and homogeneity.
homogeneity. Since online
Since large-scaletraffic
online traffic maps
maps provide almost global coverage the proposed method is practically feasible
provide almost global coverage the proposed method is practically feasible and offers a novel and offers a novel
approach for monitoring traffic systems.
approach
approach forfor monitoring
monitoring large-scale
large-scale traffic
traffic systems.
systems.
© 2020 The Authors. Published by ELSEVIER B.V.
This
© 2020is The
an open access
Authors. articleby
Published under the CC
Elsevier B.V.BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
©
© 2020
2020 The
Peer-review
The Authors.
under
Authors. Published
Published by
responsibility
by Elsevier
of the
Elsevier B.V.
scientific committee
B.V. committee of23rd
the 23rd Euro Working Group on Transportation Meeting
Peer-review under responsibility of the scientific of the EURO Working Group on Transportation Meeting.
Peer-review
Peer-review
Keywords:
under
under
Type
responsibility
yourresponsibility
of
keywords here,of
the scientific
the scientific
separated
committee
committee
by semicolons ;
of the 23rd
of the 23rd EURO
EURO Working
Working Group
Group on
on Transportation
Transportation Meeting.
Meeting.
Keywords: Type your keywords here, separated by semicolons ;
Keywords: Type your keywords here, separated by semicolons ;

1. Introduction and Background

1.
1. Introduction
Introduction and and Background
Background
The starting point for observing and modeling road traffic at networks started at the case of specific locations/cross-sections and
The starting
starting point for
for observing and
and modeling
modeling road
road traffic
traffic at
at networks
networks started
started at
at the
the case
case of
of specific
specific locations/cross-sections
locations/cross-sections and
dateTheback to thepoint
seminalobserving
work of Greenshields (Greenshields et al., 1934, Greenshields 1935), and
who carried out tests to measure
date
date back to the seminal work of Greenshields (Greenshields et al., 1934, Greenshields 1935), who carried out
out tests
tests to
to measure
trafficback to traffic
flow, the seminal work
density andofspeed
Greenshields (Greenshields
using photographic et al., 1934,
measurement Greenshields
methods for the 1935), whoThe
first time. carried
conversion frommeasure
single-
traffic flow,
traffic flow, traffic
traffic density
density and
and speed
speed using
using photographic
photographic measurement
measurement methods
methods forfor the
the first
first time.
time. The
The conversion
conversion from
from single-
single-

* Corresponding author. Tel.: +35722892286; fax: +0-000-000-0000 .

* Corresponding author. Tel.: +35722892286; fax: +0-000-000-0000 .
* Corresponding lucdimit@ucy.ac.cy
E-mail address:author. Tel.: +35722892286; fax: +0-000-000-0000 .
E-mail address: lucdimit@ucy.ac.cy
E-mail address: lucdimit@ucy.ac.cy
2352-1465 © 2020 The Authors. Published by Elsevier B.V.
2352-1465©©2020
2352-1465 2020TheThe Authors.
Authors. Published
Published byB.V.
by Elsevier ELSEVIER B.V.
2352-1465
Peer-review©under
2020 The Authors. ofPublished by Elsevier B.V.of the 23rd EURO Working Group on Transportation Meeting.
Peer-review open responsibility
This is an under access article
responsibility
the scientific
under
of the the CC
scientific
committee
BY-NC-ND
committee license
of the 23rd EURO(https://creativecommons.org/licenses/by-nc-nd/4.0)
Working Group on Transportation Meeting.
Peer-review under responsibility of the scientific committee of the 23rd EURO Working Group on Transportation Meeting.
Peer-review under responsibility of the scientific committee of the 23rd Euro Working Group on Transportation Meeting
10.1016/j.trpro.2021.01.077
646 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652
2 Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000

entity level characteristics of traffic flow to comparable system-level characteristics was first introduced by Lighthill and Whitham
(Lighthill and Whitham, 1955) and Richards (Richards, 1956) who independently proposed a simple continuum model to describe
the characteristics of traffic flow, known as the LWR model. Their study attracted attention as essential features of traffic flow,
such as wave formation and propagation, could be qualitatively well reproduced with the LWR model.
Moving to the network-wide traffic modeling, in the 1960s several landmark studies appeared exploring network-level traffic
flow relationships (Smeed 1966&1968, Thomson, 1967, Wardrop, 1968, Godfrey, 1969). Later Zahavi (Zahavi, 1972) worked on
the relationship between network-level parameters of traffic intensity, road density, and the weighted space mean speed using data
from England and the United States. In the late 1970s, Prigogine and Herman (Herman and Prigogine, 1979) proposed the two-
fluid theory in order to characterize the flow of vehicles in the urban network. Empirical validation of the two-fluid theory appeared
in the 1980s utilizing data obtained using chase-car techniques, supplemented by aerial photography (Herman and Ardekani, 1984,
Ardekani et al., 1985, Ardekani and Herman, 1987). Mahmassani et al. (1984a) and Williams et al. (1985) during their study of the
two fluid models related the three fundamental traffic variables speed-flow-concentration at a network level based on data from
simulation and indicated that they were similar to those on individual road facilities. A comprehensive review of these macroscopic
flow models is included in the Traffic Flow Theory monograph revised in 2001 (Gartner et al., 2001).
More than a decade later, Daganzo (Daganzo, 2007) revisited the concept of the macroscopic traffic flow analysis as part of an
urban traffic dynamics model for improving city mobility through gridlock control. A year later Geroliminis and Dagazo
(Geroliminis and Daganzo, 2008) in a field experiment in Yokohama, Japan, proved that average flow and density are indeed
related by a reproducible curve for a complete network, which has come to be known as the Network or Macroscopic Fundamental
Diagram (NFD or MFD).
So far, many researchers have examined several different applications of the MFD for improving traffic control. For instance
studies including gating strategies (Keyvan-Ekbatani et al., 2012), pricing strategies (Geroliminis and Levinson, 2009, Amirgholy
and Gao, 2017) or routing strategies (Yildirimoglu et al., 2015) that are sensitive to the functional form of the MFD. The factors
that influence the shape of an MFD also drew research attention. Geroliminis and Sun (Geroliminis and Sun, 2011) explored the
effect of the spatial variability of vehicle density on the shape and the scatter of a well-defined MFD. The effects of adaptive driving
on network capacity and traffic instability in a simulated network model was examined in (Saberi et al., 2014).
Given the MFD of a network, effective traffic management and monitoring strategies can be readily developed to moderate
congestion. However, the data needed to plot the MFD are not always readily available. Up to now, there are typically two empirical
data sources considered as viable for the estimation of the MFD: loop detector data and floating car data or fusing both data sources
to estimate a more accurate MFD. In (Ambühl et al., 2016) the limitations of each data source are pointed out while a combination
of them is proposed following an approach by Leclercq et al. (Leclercq et al., 2014) that leads to a well- defined MFD, reducing
key drawbacks of each data source. A preliminary study (Ambühl and Menendez, 2016) has shown that applying a data fusion
algorithm increases the accuracy of the MFD estimation.
In the current study, the feasibility of using online traffic maps information for capturing network traffic phenomena (e.g. MFD
estimation) is investigated as a validation tool for further meta-analysis. The acquisition of the vital traffic information regarding
the fundamental traffic variables (traffic flow, speed, density) and their relationships are achieved by performing a back-engineering
approach based on image processing and using fundamental microscopic traffic relationships to estimate macroscopic phenomena.
Recently, online traffic maps have increased reliability, due to the extensive use of crowd-sourced information and the augmentation
of data from many sources (mobile tracking, GPS information, traffic counts, etc.). As so, they can be considered as a collective
and important framework of combined traffic information, typically depicted in standard color-coding of road links, but with
distinctive characterization on traffic conditions. In the light of other data sources limitation, the capturing of macroscopic
phenomena utilizing online traffic maps’ information presents great value (Gkania and Dimitriou, 2019a).
The remainder of the paper is organized as follows: Section 2 introduces the methodological proposition part while Section
3 is devoted to the experimental setup and Section 4 to the calibration process. In Section 5, the proposed methodology is applied
and demonstrated for the road network of Nicosia, Cyprus. Finally, the last Section presents the conclusions and an outlook on
further research and applications.

2. Methodological Framework

In this section, the traffic information acquisition and traffic layers extraction method developed will be presented. Then,
cardinal assumptions and definitions of variables are presented, prior to the MFDs estimation and the exploration of other
macroscopic phenomena and relationships.

2.1. Traffic information acquisition and traffic layers extraction

The first step to extract traffic information from map images here relies on distinguishing and then capturing the traffic layers
from the rest of the map, providing a structured database for further analysis. The approach followed for achieving this is based on
raster image processing of online traffic maps and makes use of some important image features:

a) The discretization of areas into seamless sections/digits (pixels, in raster images),

 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652 647
Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000 3

b) Each pixel has selected dimensions (length and width) representing an area at a specific location, and
c) Each pixel captures-in a seamless manner- the traffic characteristics of the area it represents.

Image processing of urban networks has been used for many years, but this back-engineering approach is valuable for post-
processing traffic maps information and investigating the quantitative and qualitative characteristics in terms of traffic flow
fundamentals. There are many ways for extracting color layers from raster images. Here the most straightforward and reliable way
is preferred (Gkania and Dimitriou, 2019b), which is based on identifying the color codes that are used for traffic depiction, in a
color mode. Starting from the representation of maps by raster images, these are using a 3-multidimensional matrix format of size
[r, c, 3] where each pixel at location [r, c] reflect the Red-Green-Blue/RGB color model. When traffic maps are captured in raster
images, the map space is discretized into homogeneous digits/pixels, each assigned one color. Several levels of discretization can
be assumed and adopted for different purposes, making use of the pixels’ characterization and the traffic state that they are
representing. For each color code one representative value of the traffic variable can be assigned. For example, since each color
represents a range of values for each traffic variable (traffic volume/flow, average speed, average density, etc.), one can reasonably
assume one characteristic value for each color. Here, as will be later described in detail, values were selected and used for each
traffic variable after empirical calibration.
As a result of the above traffic map image processing, a matrix containing the characteristic values of the depicted traffic
variables is produced, where each non-empty entry (pixel) of the matrix can be viewed as a notional ‘detector’ of traffic states
across the map’s space, while the network properties are retained since each pixel contain the traffic information of a specific link’s
lane part/stretch. Subsequently, a typical map which is processed as an image, provides a discretized version of the network with
many thousands of notional detectors created with a complete system coverage (at least in all locations that traffic information is
broadcasted), offering a large dataset of abstracted (but dynamic) traffic information.

2.2. Nomenclature

The variables and the symbols that will be used in the analysis below, are:

Nomenclature

S Scale of the thematic map

pc Pixels colored with specific color c
Uc Representative value of average speed according to color c from the map legend, in Km/h
c Color index (i.e. g for green, o for orange, r for red and dr for dark red)
j Link index
i Pixel index
m Total number of links in one image
n Total number of pixels in one image
lj Length of link j
t time interval
Ut Average network speed across all the links of a given network, in Km/h at time interval t
Qt Average network flow across all the links of the network, in veh/h at time interval t
Kt Average network density across all the links of the network, in veh/km at time interval t
At Total number of vehicles (accumulation) on the network at time interval t
Uf Free-flow speed of a network
Kjam Jam Density of a network

2.3. Assumptions

Before the estimation of the fundamental variables based on data from traffic maps, we introduce the following assumptions of
the proposed method:

a) The scale of the map (𝑆𝑆) is used to relate the number of the pixels depicted in a raster image and the corresponding linear
distance on the road network using Equation 1:

1 Km = S x Number of Pixels (1)

b) Each link of the road network on the map is depicted by using a total number (𝑛𝑛) of colored pixels, based on Equation 2:
648 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652
4 Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000

𝑛𝑛

𝑙𝑙𝑗𝑗 = ∑(𝑝𝑝𝑐𝑐𝑐𝑐 × 𝑆𝑆 ) (2)

𝑖𝑖=1

c) Although, a pixel has both horizontal and vertical dimensions, for the proposed application the pixel’s dimensions are
strictly related to the length of a link and not to its width.

d) Each pixel is also related to the spatial dimension of traffic and can be regarded as a virtual loop detector providing
information about traffic conditions at each location.

2.4. Definitions

To explore the capacity of online traffic maps for estimating network-wide phenomena (e.g. in the detail that MFDs are offering),
fundamental traffic variables were first estimated utilizing measurements from the whole network. For achieving that, a connection
between the average network flow, the average network speed, the average network density, and the number of pixels that belong
to distinctive operating class of online traffic maps was established. The mathematical expressions used for achieving this follow.
Firstly, by considering that typical online maps offer information about links’ average speed per colored segment (complete
links or link’s stretches), we can estimate the network average speed 𝑈𝑈 𝑡𝑡 at time interval 𝑡𝑡 as:

∑𝑚𝑚 𝑡𝑡
𝑗𝑗=1(𝑙𝑙𝑗𝑗 × 𝑈𝑈𝑐𝑐𝑐𝑐 )
𝑡𝑡
∑𝑛𝑛𝑖𝑖=1(𝑝𝑝𝑐𝑐𝑐𝑐 × 𝑆𝑆 × 𝑈𝑈𝑐𝑐𝑐𝑐𝑡𝑡 )
𝑈𝑈 𝑡𝑡 = = (3)
∑𝑚𝑚
𝑗𝑗=1 𝑙𝑙𝑗𝑗 ∑𝑛𝑛𝑖𝑖=1( 𝑝𝑝𝑐𝑐𝑐𝑐
𝑡𝑡
× 𝑆𝑆)

Where the scale of the map (𝑆𝑆) is known, while the number of pixels pc colored with specific color c can be estimated based on
the proposed image processing techniques. Regarding the representative value of average speed (𝑈𝑈𝑐𝑐𝑐𝑐 ) according to color c these
can be estimated based on the map’s legend. In our case, the legend of the maps does not give precise ranges for the color-coding
scheme that has been adopted and as so these values were estimated through empirical calibration. Then to estimate the average
network density 𝐾𝐾 𝑡𝑡 that is defined as the average number of the vehicles per unit length at time interval 𝑡𝑡 , a parametric approach
was followed based on the fundamental linear speed-density relation proposed by Greenshields (Greenshields, 1935), that follows:

𝑈𝑈𝑓𝑓
𝑈𝑈 𝑡𝑡 = 𝑈𝑈𝑓𝑓 − ( ) 𝐾𝐾 𝑡𝑡 (4)
𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗

where 𝑈𝑈𝑓𝑓 stands for the free-flow speed and 𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗 is the jam density of the network.
The free-flow speed is relatively easy to estimate, as it generally lies between the speed limit and the design speed of the roadway.
On the other hand, a reliable estimation of jam density, being representative enough of the vehicles using the network of each given
time, is much more difficult. In our analysis the representative values both for free-flow speed and density jam of the network were
estimated through a calibration process utilizing empirical data from loop detectors for the selected road network.
As the variable 𝑈𝑈 𝑡𝑡 is also known from online traffic maps (Equation 3) an estimation of average network density (𝐾𝐾 𝑡𝑡 ) can be
achieved for selected free-flow speed and jam density by substituting 𝑈𝑈 𝑡𝑡 into Equation 4 and solving for 𝐾𝐾 𝑡𝑡 :

𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗
𝐾𝐾 𝑡𝑡 = 𝐾𝐾𝑗𝑗 − ( ) 𝑈𝑈 𝑡𝑡 (5)
𝑈𝑈𝑓𝑓

𝑄𝑄
The equation of flow-density relationship can be derived by substituting for 𝑈𝑈 (based on the fundamental traffic relationship),
𝐾𝐾
into Equation 4 and solving for 𝑄𝑄, as:

𝑈𝑈𝑓𝑓
𝑄𝑄𝑡𝑡 = 𝑈𝑈𝑓𝑓 𝐾𝐾 𝑡𝑡 − ( ) (𝐾𝐾 𝑡𝑡 )2 (6)
𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗

Additionally, the total number of vehicles (accumulation) on the network 𝐴𝐴𝑡𝑡 at time interval 𝑡𝑡, can be estimated by multiplying
the average network density with the total length of the links in the network using Equation 7.
𝑚𝑚 𝑛𝑛
𝑡𝑡
𝐴𝐴𝑡𝑡 = ∑ 𝑙𝑙𝑗𝑗 × 𝐾𝐾 𝑡𝑡 = ∑( 𝑝𝑝𝑐𝑐𝑐𝑐 × 𝑆𝑆) × 𝐾𝐾 𝑡𝑡 (7)
𝑗𝑗=1 𝑖𝑖=1

As it can be observed by the above equations, a relationship between microscopic traffic information (captured in each pixel)
with macroscopic (network-wide) metrics has been established. This interplay between microscopic and macroscopic variables
 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652 649
Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000 5

will be extended, enhanced and used in the later sections of this study. It is explicitly noted here that the fundamental formula
Q=K×U used for integrating pixels/locations traffic variables can be regarded as simplified, reflecting the assumption that in the
complete functional form where E(Q)=E(K)×E(U)+COV(K,U) the term COV(K,U)=0 (E standing for expected value). Though,
the implementation made here reflects an average functional among the three fundamental traffic variables, adequate for the
purposes of the current analysis.

3. Experimental Setup

In this section, the proposed method is applied over a collection of traffic images, collected from a free online traffic map
platform, for the city of Nicosia, Cyprus, for one week (11-17 February 2019). As it can be observed, four different colors are used
for distinguishing the traffic state for the whole road network (green, yellow, red and dark red). After a preliminary data
preprocessing and filtering procedure, 1806 online traffic map images were collected using 5-minute intervals. Then, an extraction
procedure of the color layers was applied to the dataset, organized in seamless pixels’ 3D matrices (see traffic layers extraction
sub-section). Next, the number of pixels and the percentage for each color were calculated for the whole dataset. Fig. 1 presents
the percentage of pixels belonging to each color layer for February 11th, 2019. As it can be observed, the number of pixels
belonging to a color reflects the expected pattern of the traffic dynamics in typical urban network. Particularly, the percentage of
green pixels (i.e. light traffic) is decreasing during the rush hours, giving space to orange (i.e. moderate traffic), red (i.e. heavy
traffic) and dark red (i.e. congestion) pixels. A steep incline of green pixels appears during the night hours as expected for the city
of Nicosia.

Fig. 1.Timeseries of pixels and boxplot per traffic layer for Nicosia for a typical weekday.

4. Calibration with empirical data

Prior to the application of the proposed method that enables to estimate MFDs based on online traffic maps information, an
empirical calibration of the proposed models was followed from available loop detectors for the same network. Through this
process the values of jam density and free flow of the network as well as the representative values of speed per colored traffic layer
were selected and applied on the Equations (3),(5),(6), and (7). As a first step, traffic data was collected for the same week (11-17
February 2019), from four available loop detectors, named 1004, 1005, 1006, 1010, located mainly in bidirectional cross-sections
of major arteries in Nicosia. as Fig. 2b shows. These loop detectors provide observations of average speed and traffic flow for 5-
minute intervals. The observations of the loop detectors were used to calibrate the proposed models through optimization process.
Particularly, the objective was to minimize the Sum Absolute Error (SAE) between the estimated values from Equation (3) and
(4), and the empirical observations, at the location of the detectors. The constraints that were set for the six variables, 𝑈𝑈𝑔𝑔 , 𝑈𝑈𝑜𝑜 ,
𝑈𝑈𝑟𝑟 ,𝑈𝑈𝑑𝑑𝑑𝑑 , 𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗 ,𝑈𝑈𝑓𝑓 follow:

• 0 ≤ 𝑈𝑈𝑔𝑔 ≤ 60 𝐾𝐾𝐾𝐾/ℎ, 0 ≤ 𝑈𝑈𝑜𝑜 ≤ 50 𝐾𝐾𝐾𝐾/ℎ, 0 ≤ 𝑈𝑈𝑟𝑟 ≤ 40 𝐾𝐾𝐾𝐾/ℎ, 0 ≤ 𝑈𝑈𝑑𝑑𝑑𝑑 ≤ 20 𝐾𝐾𝐾𝐾/ℎ,

𝑣𝑣𝑣𝑣ℎ
• 100 ≤ 𝐾𝐾𝑗𝑗𝑗𝑗𝑗𝑗 ≤ 180 ,
𝑘𝑘𝑘𝑘
• 40 ≤ 𝑈𝑈𝑓𝑓 ≤ 90 𝑘𝑘𝑘𝑘/ℎ.

A summary of the calibration results is provided in Table 1. For all the four locations of the loop detectors (1010,
1006,1005, 1004 bidirectional) five different models were tested. In the first model the objective function was to simultaneously
650 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652
6 Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000

minimize the SAE both for average flow and speed at all the four location by searching the best fit for all the six variables. In the
second model the same objective function was used while values for jam density and free flow remained constant, as in the third
model. In the fourth model the objective function was to minimize SAE only for average speed while in the fifth model only for
flow, respectively. Fig. 2a, illustrates estimated values (triangles) from all models (1,2,3,4,5) and the loop detectors’ data for the
same week, for comparative purposes. According to Table 1 the first model (1st row) presented a slight lower SAE compared to
the second one (2nd row), but the shape of the model two (blue colour) at the same figure is much closer to the theoretical one. It
is of interest, how well the second model fits to the empirical data (almost in the middle). Thus, values from the 2nd row (Table 1)
were chosen for the estimation of the MFDs for the whole network that follows in the next section.

Table 1. Calibration Results.

Model
Ug(km/h) Uo(km/h) Ur(km/h) Udr(km/h) Kjam(veh/km) Uf(km/h) SAE
Location: 1010, 1006, 1005, 1004
1 44 42 40 20 100 51 272,267(1)
2 46 45 40 4 150 50 283,342(2)
3 55 50 40 20 150 60 410,927(3)
4 40 38 40 20 100 47 282,949(4)
5 46 43 40 20 100 80 933,457(5)
Note
(1) Minimize SAE (flow and speed) by changing values in all variables
(2) Minimize SAE (flow and speed) by changing values in speed variables, Kjam=150veh/km and Uf=50km/h.
(3) Minimize SAE (flow and speed) by changing values in speed varιables, Kjam=150veh/km and Uf=60km/h.
(4) Minimize SAE (flow) by changing values in all variables
(5) Minimize SAE (speed) by changing values in all variables

(a) (b)

Fig. 2.(a) Comparison between the proposed models and the empirical data from the loop detectors and for the whole network(e) for one week (11-17 February
2019); (b) Location of the loop detectors in Nicosia, Cyprus.

5. MFD Estimation from Online Traffic Maps

In order to appraise the value of the traffic information that is broadcasted in online traffic maps, the MFD estimation (and its
properties) will be used, since MFD is an instrument that encompasses the most (if not all) of the important network-wide
information related to traffic operations. As so, the estimation of the MFD based on the earlier described image-processing
technique and its macroscopic traffic flow interpretation are presented in this section.
For the city of Nicosia, the reference method is implemented, and results are shown in Fig. 3. The plotted data set was calculated
from traffic maps’ images, utilizing solely pixels’ information for the selected week in February 2019, using a 5-minutes time
interval 𝑡𝑡. After the empirical calibration described in the previous section, the following values were selected and applied in
Equations (3), (5), (6) and (7) to estimate 𝑈𝑈 𝑡𝑡 , 𝐾𝐾 𝑡𝑡 , 𝑄𝑄𝑡𝑡 and 𝐴𝐴𝑡𝑡 , respectively:

𝐾𝐾𝑗𝑗 = 150 veh/km, 𝑈𝑈𝑓𝑓 = 50Km/h, 𝑈𝑈𝑔𝑔 =46Km/h, 𝑈𝑈𝑦𝑦 =45Km/h, 𝑈𝑈𝑟𝑟 =40Km/h, 𝑈𝑈𝑑𝑑𝑑𝑑 =4Km/h and 𝑆𝑆=0.038

The results of the Nicosia network shown in Fig 3, follow the theoretical shape of MFDs, although, the congested portion of the
curve is missing. Consequently, it can be concluded that the selected network remains uncongested for the examined time period.
 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652 651
Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000 7

(a) (b) (c)

Fig. 3. Nicosia’s MFDs, for the selected week, (5-minute intervals): (a) average flow-average density diagram; (b) average speed-average density diagram; (c)
average speed-average flow diagram.

Next, for the city of Nicosia, the investigation of the relationship between the total vehicles 𝐴𝐴𝑡𝑡 in the network and the average
network speed 𝑈𝑈𝑡𝑡 and flow 𝑄𝑄𝑡𝑡 has been performed and presented in Fig. 4. A connection among selected data (a-h) on the diagrams
and the network’s traffic state for the same intervals is highlighted on the same figure, illustrating the different network operating
conditions as these are reflected in typical color-coding schemes.
(a)

(b)

Fig. 4. The relationship between average network speed(a); flow(b) versus the total number of vehicles, and network’s traffic state for selected values of average
accumulation during February 11th, 2019 (5-minute intervals).

6. Concluding Remarks

In the current study, the properties of traffic depiction and the feasibility of using online traffic maps information for capturing
macroscopic traffic phenomena has been investigated. The analysis is based on traffic information extraction module for a realistic
network, where traffic layers are captured by straightforward image processing techniques and processed as a categorical type of
data. Fundamental traffic relationships are then used, connecting microscopic information to macroscopic phenomena. To
investigate the quality of traffic information depicted in online dynamic traffic maps, the estimation and the properties of MFD are
used. The application of the proposed methodological framework is tested on commercial freely available online maps. Then, a
thorough investigation about the type and value of spatio-temporal information that it is retained in the abstracted depiction of
652 Vana Gkania et al. / Transportation Research Procedia 52 (2021) 645–652
8 Vana Gkania, Loukas Dimitriou/ Transportation Research Procedia 00 (2020) 000–000

traffic conditions has been conducted. The results revealed that the fundamental macroscopic properties still hold even in cases of
significant information reduction like the representation scheme adopted in online traffic maps. Not only the shapes and the
magnitude of scale among the traffic variables are consistent with field observations, but also valuable additional information can
be extracted from traffic maps (like vehicles concentration, network heterogeneity, etc.). The novelty of this paper lies on the
proposed method to estimate MFD by using online traffic maps as input data in a back-engineering process, an element that could
be valuable to recent advances in Data Fusion and Big Data analysis. The comparison of the estimated MFDs with the empirical
data shows that indeed the fundamental macroscopic properties still hold. Further research is scheduled for extending this type of
analysis to different networks and information platforms, the use of additional analytical and computational means for monitoring
and possible prediction of the mobility patterns of large-scale areas

References

Ambühl, L., Loder, A., Menendez, M., Axhausen, K.W., 2016. Empirical Macroscopic Fundamental Diagrams : New Insights from Loop Detector and Floating
Car Data. 96th Annu. Meet. Transp. Res. Board, Washington, DC.
Ambühl, L., Menendez, M., 2016. Data fusion algorithm for macroscopic fundamental diagram estimation. Transp. Res. Part C Emerg. Technol. 71, 184–197.
https://doi.org/10.1016/j.trc.2016.07.013
Amirgholy, M., Gao, H.O., 2017. Modeling the dynamics of congestion in large urban networks using the macroscopic fundamental diagram: User equilibrium,
system optimum, and pricing strategies. Transp. Res. Part B Methodol. 104, 215–237. https://doi.org/10.1016/j.trb.2017.07.006
Ardekani, S., Herman, R., 1987. Urban Network-Wide Traffic Variables and Their Relations. Transp. Sci. 21, 1–16. https://doi.org/10.1287/trsc.21.1.1
Ardekani, S., Torres-Verdin, V., Herman, R., 1985. The Two-Fluid Model and Traffic Quality in Mexico City (El Modelo Bifluido y La Calidad Del Transito En
La Ciudad de Mexico). Rev. Ing. Civ.
Daganzo, C.F., 2007. Urban gridlock: Macroscopic modeling and mitigation approaches. Transp. Res. Part B Methodol. 41, 49–62.
https://doi.org/10.1016/j.trb.2006.03.001
Gartner N., Messer C., Rathi A., (Editors), 2001. Traffic Flow Theory A State-of-the-Art Report Revised, TRB’s Committee on Traffic Flow Theory and
Characteristics.
Geroliminis, N., Daganzo, C.F., 2008. Existence of urban-scale macroscopic fundamental diagrams: Some experimental findings. Transp. Res. Part B Methodol.
42, 759–770. https://doi.org/10.1016/j.trb.2008.02.002
Geroliminis, N., Levinson, D.M., 2009. Cordon Pricing Consistent with the Physics of Overcrowding. Transp. Traffic Theory 2009 Golden Jubil. 219-240\r740.
https://doi.org/10.1007/978-1-4419-0820-9_11
Geroliminis, N., Sun, J., 2011. Properties of a well-defined macroscopic fundamental diagram for urban traffic. Transp. Res. Part B Methodol. 45, 605–617.
https://doi.org/10.1016/j.trb.2010.11.004
Gkania, V., Dimitriou, L., 2019a. Identification of Macroscopic Network Performance and Phase Transition Phenomena in Urban Areas Based on Information
Retrieved By Online Traffic Maps, in: 98th Transportation Research Board Annual Meeting.
Gkania, V., Dimitriou, L., 2019b. Uncovering human mobility patterns in cities through the lens of traffic maps, in: 98th Transportation Research Board Annual
Meeting. Washington D.C., U.S.A.
Godfrey, J.W., 1969. The mechanism of a road network. Traffic Eng. Control 11, 323–327.
Greenshields, B.D., 1935. A study of traffic capacity, in: 14 Annual Meeting of the Highway Research Board Proceedings. pp. 448–477.
Greenshields, B.D., Thompson, J.T., Dickinson, H.C., Swinton, R.S., 1934. The Photographic Method Of Studying Traffic Behavior. Highw. Res. Board Proc.
Herman, R., Ardekani, S., 1984. Characterizing Traffic Conditions in Urban Areas. Transp. Sci. 18, 101–140. https://doi.org/10.1287/trsc.18.2.101
Herman, R., Prigogine, I., 1979. A two-fluid approach to town traffic. Science 204, 148–151. https://doi.org/10.1126/science.204.4389.148
Keyvan-Ekbatani, M., Kouvelas, A., Papamichail, I., Papageorgiou, M., 2012. Exploiting the fundamental diagram of urban networks for feedback-based gating.
Transp. Res. Part B Methodol. 46, 1393–1403. https://doi.org/10.1016/j.trb.2012.06.008
Leclercq, L., Chiabaut, N., Trinquier, B., 2014. Macroscopic Fundamental Diagrams: A cross-comparison of estimation methods. Transp. Res. Part B Methodol.
62, 1–12. https://doi.org/10.1016/j.trb.2014.01.007
Lighthill, M.J., Whitham, G.B., 1955. On Kinematic Waves. II. A Theory of Traffic Flow on Long Crowded Roads. Proc. R. Soc. A Math. Phys. Eng. Sci. 229,
317–345. https://doi.org/10.1098/rspa.1955.0089
Mahmassani, H.S.., Williams, J.C., Herman, R., 1984. Investigation of Network-Level Traffic Relationships: Some Simulation Results. Transp. Res. Rec. 971,
121–140.
Richards, P., 1956. Shock Waves on the Highway. Oper. Res. 4, 42–51. https://doi.org/10.1287/opre.4.1.42
Saberi, M., Mahmassani, H.S., Zockaie, A., 2014. Network capacity, traffic instability, and adaptive driving: findings from simulated urban network experiments.
EURO J. Transp. Logist. 3, 289–308. https://doi.org/10.1007/s13676-013-0040-2
Smeed, R.J., 1968. Traffic Studies and Urban Congestion. J. Transp. Econ. Policy 2, 33–70. https://doi.org/10.1177/03063127067078012
Smeed, R.J., 1966. Road Capacity of City Centres. Traffic Eng. Control 8, 455–458.
Thomson J.M, 1967. Speeds and flows of traffic in central London: Speed-flow relations. Traffic Eng. Control 8, 721–725.
Wardrop J.G, 1968. Journey speed and flow in central urban areas. Traffic Eng. Control 9, 528–532.
Williams, J.C., Mahmassani, H.S., Herman, R., 1985. Analysis of Traffic Network Flow Relations and Two-Fluid Model Parameter Sensitivity. Transp. Res.
Rec. 95–106.
Yildirimoglu, M., Ramezani, M., Geroliminis, N., 2015. Equilibrium analysis and route guidance in large-scale networks with MFD dynamics, in: Transportation
Research Procedia. pp. 185–204. https://doi.org/10.1016/j.trpro.2015.07.011
Zahavi, Y., 1972. Traffic Performance Evaluation of Road Networks by the α- Relationship. PartsI & II. Traffic Eng. Control Vol. 14, 228–293.