Visualizing Large Scale Vehicle Trafﬁc Network Data

A Survey of the State-of-the-art

H. W. A. S. Gondim

, H. A. D. do Nascimento

and D. Reilly

Institute of Informatics, Federal University of Goias, Campus II, Caixa Postal 131. CEP 74001-970, Goiania - GO, Brazil

Faculty of Computer Science - Dalhousie University, 6050 University Ave., Halifax, Nova Scotia B3H 4R2, Canada

Keywords:

Urban Trafﬁc, Trafﬁc Congestion, Information Visualization.

Abstract:

Analyzing and improving large urban trafﬁc networks is a difﬁcult process due to complex interrelationships

between the many variables that impact vehicle trafﬁc behavior. Information visualization techniques can

facilitate the tasks of analyzing large amounts of data and of exploring potential solutions to practical trafﬁc

problems. Surprisingly, there is a relative lack of investigation focused on how information visualization

techniques should be applied and adapted to the ﬁeld of Trafﬁc Engineering. This paper presents an overview

of what has been done on this topic by reviewing the use of information visualization in trafﬁc systems over

the years, and highlighting the current state-of-the-art by focusing on several innovative pieces of research.

We provide a classiﬁcation of the reviewed work and identify areas that have been understudied.

1 INTRODUCTION

Large urban areas in many countries suffer from in-

creasing trafﬁc congestion. Such congestion costs

money, and impedes the growth of a country. It also

affects people’s health; for example, vehicles moving

at low speed tend to consume more fuel and produce

more toxic gases than when traveling faster (Seong

et al., 2011). A study conducted in 45 countries on

the growth of world income and vehicle ownership

forecasted an increase of two billion vehicles by 2030

(Dargay et al., 2007). Therefore, mitigating the urban

trafﬁc congestion problem is important for econom-

ical and social development, as well as for environ-

mental reasons.

Trafﬁc problems are not simple to solve as they

depend on several dynamic and interrelated aspects

(urbanism, population distribution, location of work-

places, social, economical and leisure activities, etc.)

which are hard to describe and to model in full de-

tail (Button, 2001). As a result, the study of this sub-

ject usually involves dividing a trafﬁc problem into

steps or stages, in order to facilitate understanding.

The famous Four-Step Model (McNally, 2000) is an

example. It is comprised of the following steps: trip

generation, trip distribution, modal split (trips by car,

buses, subway, etc.) and trafﬁc assignment. There

are many other ways of analyzing these problems. In

general, they can be divided into three major phases:

(1) collecting information and describing the network

and the travel demands (at different periods of the day

and days of the week), (2) deﬁning a trafﬁc assign-

ment model that considers the characteristics of the

network and the transportation modes, and (3) experi-

menting with the model and variations of the network

through simulation.

The issues that need to be addressed in each phase

have been well studied and a variety of strategies have

been devised for dealing with them. For instance, ve-

hicle trafﬁc networks are often modeled using graph

theory concepts, while travel demands are usually

represented in the form of two-dimensional arrays

called origin-destination (OD) matrices, for distinct

modes of transportation (car, bus, etc.), periods of the

day (rush hours, night time, etc.), and days of the

week (Ashok and Ben-Akiva, 1993). Macroscopic,

mesoscopic and microscopic methods (either static or

dynamic) exist for trafﬁc assignment at different reso-

lutions (Baskar et al., 2009) and each helps to predict

vehicle trafﬁc behavior in decision support systems.

However, algorithms and calculations are not suf-

ﬁcient to obtain feedback quickly, and to uncover pat-

terns in the complex data sets collected for and gen-

erated by these systems. Thus, tools for visualization

are important for such applications.

Information visualization combines aspects of

human-computer interfaces, computer graphics and

data mining, representing data in a graphical format

337

Gondim H., do Nascimento H. and Reilly D..

Visualizing Large Scale Vehicle Trafﬁc Network Data - A Survey of the State-of-the-art.

DOI: 10.5220/0004686603370346

In Proceedings of the 5th International Conference on Information Visualization Theory and Applications (IVAPP-2014), pages 337-346

ISBN: 978-989-758-005-5

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

that expands users’ visual perception and promotes

better understanding of the data (Agrawal et al., 1993;

Fayyad et al., 1996). A visualization can support three

general activities:

• Exploratory Analysis – ﬁnding new knowledge

contained in the data through an analytical pro-

cess that explores visual representation for high-

lighting trends / patterns that can lead to new hy-

pothesis.

• Conﬁrmatory Analysis – visually demonstrating

the validation or rejection of a particular hypothe-

sis.

• Presentation – graphically representing relation-

ships, structures, behaviors, and other intrinsic

characteristics to the data.

Given the great advances in the area of informa-

tion visualization research, surprisingly few studies

have focused on the role of visualization in the anal-

ysis of trafﬁc problems. There is certainly some form

of visualization present in many computational sys-

tems that assist in the modeling and the simulation of

vehicle trafﬁc. But they have been mostly employed

for illustrating a process or a problem. Reported work

that does employ information visualization generally

does not consider the techniques used per se; rather,

it assumes that the use of visualization (in some form)

yields certain de facto beneﬁts.

The limited application of information visualiza-

tion to trafﬁc problems is perhaps understandable,

since it is not trivial to identify a graphical representa-

tion that is appropriate for the type of data and that ad-

equately considers the principles of human perception

(Gershon et al., 1998), (Chen, 2005). One major issue

is the difﬁculty in representing multi-attribute data in

ways that don’t overwhelm our perceptual abilities.

This is especially relevant for trafﬁc, where we need

to analyze complex data sets over both temporal and

spatial dimensions (Boyandin et al., 2011).

This paper surveys the state of the art in the use of

information visualization techniques to support mod-

eling, simulation and analysis of urban trafﬁc. Vi-

sualizations supporting navigation are not considered

here, as our focus is to investigate techniques that

help trafﬁc engineers and city government planners

improve trafﬁc conditions. In addition, we focus on

urban trafﬁc networks due to their complexity.

The paper is organized as follows: Section 2

presents the state-of-the-art, describing research that

uses information visualization for vehicle trafﬁc mod-

eling, simulation and analysis. Section 3 provides a

characterization of the research in this area and identi-

ﬁes some understudied aspects. In Section 4 we draw

our conclusions and suggest areas for future work.

2 INFORMATION

VISUALIZATION FOR

VEHICLE TRAFFIC

Many visualization techniques proposed to support

the study of trafﬁc problems have their foundation

in well-established ﬁelds, most notably Cartogra-

phy (Taylor et al., 2000; Meyer and Miller, 2001).

The majority of visualizations that appear in the traf-

ﬁc engineering literature are common 2D geographic

maps, or geographic maps employing a 3D perspec-

tive. Standard map features from cartography are

widely used, including the division of political re-

gions (suburbs, estates, etc.) by polygonal area fea-

tures, the use of line features for representing streets,

avenues, water channels and rivers, and the use of vi-

sual attributes like color, shade and texture to classify

or highlight map elements.

Geographic Information System (GIS) and Deci-

sion Support System technologies have been applied

to trafﬁc engineering to yield modeling and simula-

tion systems that combine georeferenced data with

an interactive multiple layer visual representation, for

the analysis and planning of trafﬁc networks (Huang

and Pan, 2007). Some popular commercial GISs

in this category are the software Saturn (Van Vliet,

2013) and Dracula (Liu et al., 2008), TransCAD

(Caliper, 2012) and VISSIM (PTV, 2013). Many

other trafﬁc modeling systems are also gaining pop-

ularity, such as the free software SUMO (Krajzewicz

et al., 2013), MITSIM (MitSim, 2013) and PetGyn

(Jradi et al., 2009).

Recently, several research projects have explored

innovative ways of presenting trafﬁc data. They differ

from previous research and from commercial systems

in their application of new presentation metaphors

and more complex visual approaches, or by being

highly interactive. We survey such pieces of work

here. They are grouped in this section according to

the information visualization classiﬁcation proposed

by (Keim and Kriegel, 1996), to provide a prelim-

inary categorization of the work. This classiﬁca-

tion includes Geometric Projections (consisting of all

techniques that show projections of multidimensional

data sets), Pixel-oriented views (which map each data

value to a colored pixel), and Hierarchical presen-

tations. Keim et al. also described a Graph-based

group, but we have realized that all papers that fall in

this category also draw a graph as a geographic pro-

jection of a map. Therefore, those papers were classi-

ﬁed as belonging to the Geometric Projection group.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

338

2.1 Geometric Projection

As expected, geometric projection remains a key ap-

proach for generating interactive visualizations of

trafﬁc network data. Increasingly, researchers are

applying advanced graphical techniques (animations,

perspective transformations, morphing), expanding

the expressive power of geometric projections in ways

that suit the presentation of trafﬁc ﬂow, multidimen-

sional trafﬁc data, and the complex relationships be-

tween network elements (e.g., origins, intersections,

destinations).

Cascade on Wheels (Thirion et al., 2008) is a vi-

sualization project started during the Visualizar Work-

shop at the Medialab Prado in Madrid in 2007. The

project’s main goal was to express the amount of vehi-

cles in big cities using metaphor. Two visualizations

were developed and applied to data regarding taxis,

buses and light and heavy vehicle ﬂows in the central

region of Madrid in 2006.

The ﬁrst visualization, referred to as Walls Map

Piece (Figure 1-(a)), shows trafﬁc data using a wall

metaphor. It presents a 2D satellite map of the city,

in which it is possible to select streets and visualize

their names and trafﬁc data (the amount and type of

vehicles) in a given day. By selecting a tool, the visu-

alization then turns into a 3D map with streets raised

according to the amount of daily trafﬁc ﬂow in them.

Each street segment creates a wall with height propor-

tional to its amount of trafﬁc. The user can rotate the

3D map to view the sides of walls, which show the

amount of vehicles of each type using bar charts. A

similar approach applied to vehicle crash analysis can

be found in (Li et al., 2007).

The second visualization (called Trafﬁc Noise

Mixer) makes use of an audio layer combined with a

network satellite map (see Figure 1-(b)). The user can

interact with the visualization by drawing, dragging

and growing circular regions over the map. Sounds

with speciﬁc frequencies and amplitude are then gen-

erated and played, representing the amount and the

types of vehicles traveling in the selected regions.

The LIVE Singapore project (Kloeckl et al.,

2011), created by the SENSEable City Laboratory at

MIT, explores the development of an open platform

for the collection, elaboration and distribution of a

large and growing number of different kinds of real-

time data that originate in a city (e.g., from cellular

networks, taxi ﬂeets, public transport, seaports and

airports). The platform provides support for gener-

ating interactive visualizations employing geometric

projections. Some applications built with the LIVE

Singapore framework are discussed next.

Members of the LIVE Singapore Project (Kloeckl

Figure 1: (a) Walls Map Piece - a 2D/3D map with walls

representing streets with dense trafﬁc. The greater the ﬂow

of vehicles in a street segment, the higher the corresponding

part of the wall. (b) Trafﬁc Noise Mixer. Audible noise is

produced in order to indicate the amount and types of vehi-

cle in selected regions. Courtesy of (Thirion et al., 2008).

et al., 2011; Chen et al., 2011) proposed transforma-

tions (compression / distension) of a map of Singa-

pore according to the time demanded for traveling be-

tween origin-destination of points, in a view called

“Isochronic Singapore”. About 290 controlling points

were deﬁned. Selecting an origin point causes the

map to be visually distorted so that the distance from

that origin to all destinations in the network is propor-

tional to the travel time between them. The visualiza-

tion also changes automatically as the observed trafﬁc

ﬂows in the network vary during the day.

The Data Lenses visualization(Cruz, 2012) is part

of the Live Singapore project as well. It was created to

enable precise monitoring of activities of urban pub-

lic transportation (buses) in Singapore. The visual-

ization uses lenses that work like magnifying glasses

(see Figure 2). The users can move the lens or change

its properties (like radius and magniﬁcation level) to

improve perception of a dense information area. It is

possible to adjust the lenses to present different lay-

ers of information, such as bus lines at each bus stop

(in blue), the number of bus passengers, the number

of passengers boarding/exiting at the bus stops (in or-

ange) and the total amounts paid for tickets at each

bus stop (in red).

Furthermore, the LIVE Singapore visualization

called “Trafﬁc Origins” (Kloeckl et al., 2012) aims

to identify road events that usually cause congestion.

In this visualization, information previously available

about road events can be combined with simulation

models for better comprehension of trafﬁc conditions

VisualizingLargeScaleVehicleTrafficNetworkData-ASurveyoftheState-of-the-art

339

Figure 2: The Data Lenses visualization permits the am-

pliﬁcation and suppression of urban trafﬁc data on a map.

Courtesy of (Cruz, 2012).

and to predict the impact of trafﬁc interruption in the

network. Three types of events can be automatically

highlighted in the visualization using colored circles:

accidents (in red), intense trafﬁc (in blue) and broken

vehicles (in orange). After identifying a region of the

map where one of these events occurrs, the visualiza-

tion emphasizes the color in the nearby streets (inside

the circle) to improve perception of their trafﬁc ﬂow

speed. The advantage of this techinique is to focus

the users attention only on the important occurrences,

given by circled events with colored information.

The Web site “Every Death on Every Road in

Great Britain 1999 - 2010” (BBC, 2011), produced

by the British Broadcasting Corporation (BBC) and

released in 2011, uses geometric projection to show

the evolution of trafﬁc accidents over time. It presents

data and facts about accidents with injuries and deaths

on Great Britains roads over a span of eleven years.

The visualization resembles a nightly aerial map view,

with lights indicating road collisions that caused in-

juries and deaths.

Other research has considered technical chal-

lenges involved in generating trafﬁc visualizations

employing geometric projection. For example, the

work of Chen et al. (Chen et al., 2008) considers the

difﬁculty in modeling large meshes of roads. In this

context, they proposed a system to assist the user on

modeling/drawing a road network using stress ﬁelds.

As another example, Sewall et al. (Sewall et al.,

2011) present a hybrid model that combines con-

tinuum (macroscopic) and agent-based (microscopic)

methods for efﬁcient and effective trafﬁc simulation

and visualization of large-scale networks in real time.

In general, macroscopic methods are faster but do not

capture individual vehicle behaviors precisely. On the

other hand, microscopic methods are more ﬂexible for

modeling trafﬁc elements in detail, but demand ex-

tra processing power. In their approach, a large net-

work can be divided into several small regions and

each region can be simulated by either the micro-

scopic or macroscopic method. This hybrid model

allows a ﬁne-grain detail of a network region of in-

terest, while simultaneously keeping the simulation

of the remaining part of the network with the contin-

uum method. Trafﬁc data from the different methods

are interchanged.There are other papers with similar

visualizations (Kim et al., 2009; Mahut and Florian,

2010; Svennerberg, 2010; Buch et al., 2011).

Other research has deeply integrated visualiza-

tion tools into trafﬁc analysis system. The CityMo-

tion Project (Silva et al., 2010) was developed by

researchers of the University of Coimbra, MIT and

the University of Porto, between 2007 and 2010.

Their goal was to build knowledge infrastructure,

computational models and user applications that al-

low access to real-time information about the state of

transportation-related resources as well as to do pre-

dictions regarding their future state.

One of the CityMotion subprojects is called “Vi-

sualizing the Circulatory Problems of Lisbon” (Cruz

and Machado, 2011). That project resulted in alter-

native visualizations to identify bottlenecks in a road

network based on a metaphor of the circulatory sys-

tem of a living being. The scenario is Lisbon in 2009

with data from 1,534 vehicles. They presented two

types of visualizations inspired by basic biology.

Figure 3: Images (1) and (2) represent clots in Lisbon at

9am and 7pm, respectively. Images (3) and (4) show blood

vessels in Lisbon at 7:04 am and 8:44 am. Courtesy of (Cruz

and Machado, 2011).

In the ﬁrst visualization, shown in images (1) and

(2) of Figure 3, blood clots (as red circles) represent

zones of slow trafﬁc. The diameter of the clots is in-

versely proportional to trafﬁc speed.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

340

In the second visualization, shown in images (3)

and (4), color and the size of blood vessels change

according to the number of vehicles and the average

transit speed in each road.

In both visualizations, trafﬁc in low speed tends

to darken the related blood vessel, representing blood

stagnation. In contrast, if the trafﬁc speed is high, the

blood vessel is painted in lighter color.

The “Pulse of the City” (Reades, 2011; Reades,

2012) is an animated view developed by Jonathan

Reades

that also applies the circulatory system

metaphor. It shows the usage of public railway trans-

portation lines in London, presenting railway lines

over a cartographic map with a dark background. As

time passes and commuters travel in the railway sys-

tem, railway line segments increase in thickness ac-

cording to their use. If a railway segment that had

many passengers before becomes less used, its thick-

ness shrinks to normal size. Since the animation cov-

ers many days and periods of the day, even with rush

hours, the resulting visual effect is of a living circula-

tory system that pulses with blood.

While the circulatory system metaphor is one

speciﬁc approach, a range of projects more gener-

ally apply deformation and color to highlight areas

of concern in trafﬁc visualizations. Another visu-

alization from the CityMotion project is “Morphing

City” (Cruz, 2011), which presents a map whose lines

are deformed (compressed or distended) if the current

ﬂow speed on the road is lesser or greater than the av-

erage global speed. Color assists in the interpretation

of the map as well: deformations that compress the

view are assigned cold colors, while warm colors are

applied for expansions. The Morphing City visualiza-

tion enables rapid identiﬁcation of critical points since

areas with intense trafﬁc are highlighted by warmer

colors and appear to expand in comparison to other

routes.

Seong et al. (Seong et al., 2011) use Geomet-

ric Projection to model and map noise produced by

urban trafﬁc in Fulton County, Georgia, USA. They

created a visualization where decibel ranges are rep-

resented by colors: buildings with low noise level are

shown in green, high levels of noise appear in red,

and there is a color scale indicating intermediate lev-

els. A range of visualizations considering noise and

air pollution have been considered in other research

(Tsai et al., 2009; Elbir et al., 2010; Pamanikabud and

Tansatcha, 2010; Zannin et al., 2013), most employ-

ing some form of geometric projection.

In our review, we also found some projects that

aim to visualize origin-destination matrices using a

combination of geometric projections and other tech-

http://vimeo.com/user11576485/pulseofthecity

niques, including graph-based approaches. The work

of Boyandin et al. (Boyandin et al., 2011) is one

example. It highlights the limitations of geometric

projections alone when applied to temporal origin-

destination datasets. The paper then introduces an in-

teractive visualization called Flowstrates (Figure 4),

which presents two different geometric maps with a

heatmap in the center, using color to represent de-

mand over time. Similar visualizations are presented

in (Wood et al., 2011).

Figure 4: The image represents the annual number of

refugees of Ethiopia. Each column in the heatmap (between

two maps) indicates temporal scale. In this example Soma-

lia is the country which receives more refugees (hot colors

represents a big density, when cool colors is small).

In another project (Wood et al., 2010), the authors

draw OD matrices as colored cells and lines over a

cartographic map, according to the density of trips

between the OD pairs. In addition, Gunay (Gunay,

2009) shows OD matrices using Wavelets transforma-

tions in order to highlight similarity in data.

2.2 Pixel-oriented

In this section we discuss recent work that employs

a pixel-oriented approach. Just as with graph-based

examples, this also is done in a manner that integrates

with a geometric projection.

The work of (Xie and Yan, 2008) presented spa-

tial patterns of trafﬁc accidents in roads. A variation

of the Kernel Density Estimation (KDE) technique

was developed that smooths density surface of spatial

point events over a 2D geographic space. The new

approach was implemented in the ESRI ArcGIS en-

vironment and tested with trafﬁc accident data from

2005 in Bowling Green, Kentucky, USA, provided

by the Kentucky State Police Department. The re-

searchers found that the new KDE was more effective

than the standard method for estimating trafﬁc acci-

dent densities.

In order to help understand the results, the authors

used a map visualization, as shown in ﬁg 5. Figure

5-(a) highlights the density of accidents by varying

the thickness of roads in the positions where there are

VisualizingLargeScaleVehicleTrafficNetworkData-ASurveyoftheState-of-the-art

341

many accidents. This is carried out a smoothing pro-

cedure that expands and connects points of the net-

work if they are already close to each other and there

is a high number of accidents in them. Figure 5-(b)

presents the same data but uses a 3D perspective to

show trafﬁc accident densities (the higher the density,

the higher the peak over an area).

Figure 5: Visualizing trafﬁc accident densities. The im-

age (a) indicates trafﬁc accident densities by expanding and

connecting nearby points. Other presents this data in a 3D

view using height to indicate density. Courtesy of (Xie and

Yan, 2008).

Other visualization approaches based on varia-

tions of KDE can be found (Ha and Thill, 2011; Liu

et al., 2012; Wang et al., 2009; Mesbah et al., 2012).

2.3 Hierarchical

Wood et al. (Wood and Dykes, 2008) explore the use

of treemaps for variable selection in spatial-temporal

data. Their study focuses on a dataset consisting

of 90 million georeferenced vehicle locations from

a courier ﬂeet. The data was collected during 18

months and each location contains its geographic po-

sition (given by GPS), timestamp, vehicle type (in-

cluding van, large van, motorbike, large motorbike

and bicycle) and travel speed. The goal of that study

was to demonstrate that treemap visualizations can

help users identify useful patterns in the data and to

select information for further analysis.

The researchers present two approaches for visu-

alizing the dataset. In the ﬁrst one shown in Figure 6-

(a), the data is hierarchically organized according to

vehicle type, day of the week (from Monday to Sun-

day) and hour of the day (24 hours in total). In the

lowest level of the hierarchy, each treemap rectangle

has an area proportional to the volume of vehicles (ac-

cording to the corresponding type, day of week and

hour). Color intensity is also used for mapping the

average speed of these vehicles.

In the second approach, illustrated in Figure 6-(b),

a map of the city is divided into a grid of rectangu-

lar regions. Each cell of the grid contains a treemap

Figure 6: (a) Treemap showing trafﬁc data classiﬁed by

transportation mode, day of week and hour of the day. The

rectangles are ordered by size from top-left to bottom-right

for each level of the classiﬁcation. (b) An area of London

divided into cells of 1km square each (left). In each cell

there is a treemap of the hierarchically ordered vehicle data.

The image on the right is a zoomed view of the “Farring”

cell. Lighter colors indicate that vehicles traveled at a lower

average speed. Courtesy of (Wood and Dykes, 2008).

presenting data from that region, hierarchically orga-

nized as before. Color intensity indicates either trafﬁc

volume or average speed in each small rectangle at

the lowest level of the hierarchy.

Despite being used for visualizing data for a

courier ﬂeet, the same approaches can be applied to

other trafﬁc data in a straightforward way.

3 CHARACTERISTICS AND

UNTREATED ASPECTS

As we have seen, current trafﬁc network visualiza-

tion research is heavily oriented toward geometric

projection-based approaches. We can characterize the

visualization techniques we have discussed according

to different criteria so that it becomes easier to com-

pare them as well as to identify as-yet-unexplored av-

enues for trafﬁc visualization. We propose here two

major criteria for vehicle trafﬁc related visualizations:

• Type of data. The visualizations are grouped

according to the main type of trafﬁc data they

present. These types include, but are not lim-

ited to the following: O-D matrices, rigid net-

work infrastructures (such as roads, bridges and

intersections), ﬂexible elements of the network

(that can be changed with little or no infrastruc-

tural intervention, such as lane direction, max-

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

342

imum speed limits, signal settings, turning re-

strictions, parking authorizations, trip and park-

ing tolls, etc.), ﬂow measurements or ﬂow esti-

mations, and meaningful high level data such as

information about car accidents and other events.

One can see that such data can be naturally orga-

nized into levels in a hierarchical multilayer way,

as is usually done in geographic information sys-

tems. Therefore, we can further classify a traf-

ﬁc visualization according to whether it was de-

signed for presenting a given layer of data or a

combination of them.

• Target task. In this case, the visualizations are

classiﬁed according to the kind of task(s) they are

intended to support. We envision three types of

tasks. The ﬁrst is simply to present trafﬁc data

for overall understanding and for pattern identiﬁ-

cation. This involves showing the data and pos-

sibly allowing interaction to control the visual-

ization attributes, or to apply ﬁlters. The second

type of task is to input data. Although user in-

terface components like dialog boxes, tables and

text ﬁelds could be used for inputting data into a

trafﬁc system, we mean here a more intuitive and

visually-oriented way of entering information by

direct manipulation of the elements of a visual-

ization. The third type of task is to simulate and

optimize trafﬁc behavior through the exploration

of multiple scenarios of the network and trafﬁc

conditions. This task is very common in decision

support systems for trafﬁc engineering. We can

say that it involves an aggregation of the previous

two tasks in order to allow the user to pursue a

more complex goal. Similarly to the ﬁrst classi-

ﬁcation criterion, this second one is not mutually

exclusive in its options, allowing the presence of

two or all three target tasks simultaneously.

We analyzed the visualizations reviewed in Sec-

tion 2 according to these criteria and presented the

result in Table 1.

Note that it is possible to characterize the views

according to the type of visualization approach too

(2D, 3D, treemap, etc.) and by the interaction tech-

niques used. We consider these other criteria useful,

but mainly to further differentiate visualizations that

have already been classiﬁed by the other means.

One of the main observations from our survey of

the ﬁeld of visualizing vehicle trafﬁc data and traf-

ﬁc problems is that there are many combinations of

the two criteria proposed above for which no visual-

ization technique is currently available. Furthermore,

there has been no systematic study that demonstrates

how to build good visualizations for variations of data

type and target task.

Table 1: Characterization of the reviewed visualizations.

Type of Data Target Task

Projects

OD Matrix

Network Infrastrucure

Flexible network elements

Flow

High Level Meaningful Data

Data Presentation

Inputting Data

Simulation and Exploration

(Waze, 2013) x x x x x x

(Zannin et al., 2013) x x x x x

(Cruz, 2012) x x x x x

(Reades, 2012) II x x x x

(Liu et al., 2012) x x x

(Mesbah et al., 2012) x x x x x

(Seong et al., 2011) x x x x

(Sewall et al., 2011) x x x x x

(Reades, 2011) x x x x

(Cruz, 2011) x x x x x

(Kloeckl et al., 2011) x x x x x

(Chen, 2011) x x x x x

(Boyandin et al., 2011) x x x

(BBC, 2011) x x x x x

(Ha and Thill, 2011) x x x x

(Elbir et al., 2010) x x x x x

(Wood et al., 2010) x x x x

(Silva et al., 2010) x x x x x x x

(Kim et al., 2009) x x x x x

(Wang et al., 2009) x x x x x

(Outram et al., 2009) x x x x x

(Tsai et al., 2009) x x x x

(Wood and Dykes, 2008) x x

(Jradi et al., 2009) x x x x x x x

(Xie and Yan, 2008) x x x x x

(Thirion et al., 2008) x x x x

(Chen et al., 2008) x x x

(Madsen et al., 2007) x x x x x

(Tang and Wang, 2007) x x x x x x

(Li et al., 2007) x x x x x

(Wang, 2005) x x x x x x

(Elbir, 2004) x x x x

(Borho, 2003) x x x x x x

(Brandt et al., 2003) x x x x

(Claramunt et al., 2000) x x x x x x

Another important observation is that, for many

tasks and data types for which supposedly reason-

able visualization techniques have been applied, the

visualizations are revealed to be ineffective or inef-

ﬁcient in some common cases. For instance, a tra-

ditional way of denoting ﬂow orientation on streets

and turning permissions/restrictions on intersections

is by drawing arrows that indicate the ﬂow directions.

However, when the network is large, it becomes prac-

tically impossible to visualize these details in a car-

tographic map. Interactive zoom can solve this prob-

lem, but at the expenses of compromising perception

of the complete network. Also, it does not allow

effective simultaneous visualization of multiple de-

tailed regions of the network. Innovative pieces of re-

search as the work done by Sewall et al. (Sewall et al.,

2011), that uses augmented reality over a 2D network

map to present animations of vehicles moving, would

offer ﬂow direction perception while preserving full-

picture zoom level. However, this approach demands

high processing power, does not show the network

characteristics in regions where there is absence of

ﬂow, and does not provide a good visualization of

VisualizingLargeScaleVehicleTrafficNetworkData-ASurveyoftheState-of-the-art

343

ﬂow orientation at intersections with many crossings

and turning permissions.

The Flowstrates approach (Boyandin et al., 2011),

as another example, improves the understanding of

Origin-Destination matrices, but has several limita-

tions. There is a lot of visual overlapping when the

number of selected OD pairs increases. Furthermore,

it does not show how OD demands translate into real

trafﬁc ﬂow, particularly when considering an urban

trafﬁc network, with many possible ﬂow paths.

In general, the existing visualizations suffer from

two main problems: visual overlapping as the size

of the data set steadily increases, and an inability to

present different types of data in an integrated way

(network structures, OD demands, trafﬁc ﬂow, etc.).

It is also difﬁcult to develop a “robust” visualization

that produces visually acceptable results for all traf-

ﬁc scenarios, since we are dealing with very dynamic

systems in which small changes of the network struc-

ture may result in totally different ﬂow behavior.

Finally, the majority of trafﬁc-related visualiza-

tions and interaction techniques considered to date

were designed with traditional mouse+keyboard or

touchscreen conﬁgurations in mind. Few researchers

are investigating the use of novel HCI approaches, go-

ing beyond the simple dualities of large/small screens

and touch/indirect interfaces to build on the frontier

and explore how techniques from augmented reality,

tangible, mobile and ubiquitous computing can play a

role.

Some interesting examples are emerging in the

processes of collecting trafﬁc data. For example, the

Copenhagen Wheel Project (Outram et al., 2009) con-

sists of an electric wheel extension (called “hub”)

that transforms any ordinary bike into a hybrid e-

bike with a motor, batteries, GPRS, and sensors for

torque, moisture, carbon dioxide, temperature and

noise. While a bicyclist follows their route, the sen-

sors collect information and send them to an iPhone

every two seconds via Bluetooth.

Copenhagen Wheel is one example of a larger

trend toward Crowdsourcing trafﬁc network data

(Howe, 2006; Weissman and Villalobos, 2012). Waze

(Waze, 2013) is an application for mobile devices that

allows trafﬁc information to be shared, such as loca-

tions with congestion, accidents, objects on a track,

as well as tips and warnings. It has been reported that

the use of such software can help in better choices of

route (Chen et al., 2012; Talele et al., 2012).

In future, systems that exploit crowdsourcing,

open data, and heterogeneous data sets will need to

manage information that is not always complete, cor-

rect or provided at the same level of granularity. This

will also impact on how to best visualize and interact

with the data.

4 CONCLUSION

We have reviewed the research on information visual-

ization applied to vehicle trafﬁc problems and data in

order to show the state-of-the-art in this ﬁeld.

We then proposed a characterization approach for

trafﬁc-related visualization methods. Based on our re-

view and on the characterization criteria, one can see

that there still lacks much investigation of interactive

visualization methods for certain tasks and types of

trafﬁc data. Furthermore, a range of emerging HCI

techniques has not generally been employed by re-

searchers working in this domain. We have also iden-

tiﬁed two main limitations with current approaches:

visual overlapping and poor integration of different

forms of trafﬁc network data. Our paper is a step to-

wards better understanding these challenges, by re-

viewing the state-of-the-art in this ﬁeld and by situ-

ating the existing research within our categorization

scheme. Nevertheless, there is still much to investi-

gate and to deﬁne in this area.

For future research, we suggest:

• more exploration of techniques for integrating

multiple views, each optimized for speciﬁc types

of network data;

• consideration of 3D interaction techniques to mit-

igate the problem of visual overlapping;

• expanding the characterization scheme to include

interactive techniques and conceptual approaches

for inputting data;

• further exploring interactive visualizations for the

task of inputting different types of trafﬁc-related

data; and

• adapting emerging HCI techniques to trafﬁc-

related visualization in decision support systems.

REFERENCES

Agrawal, R., Imielinski, T., and Swami, A. (1993).

Database Mining: A Performance Perspective.

Knowledge and Data Engineering, IEEE Transactions

on, 5(6):914–925.

Ashok, K. and Ben-Akiva, M. E. (1993). Dynamic Origin-

destination Matrix Estimation and Prediction for Real-

time Trafﬁc Management Systems. In International

Symposium on the Theory of Trafﬁc Flow and Trans-

portation (12th: 1993: Berkeley, Calif.). T ransporta-

tion and Trafﬁc Theory.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

344

Baskar, L., De Schutter, B., and Hellendoorn, J. (2009).

Optimal Routing for Intelligent Vehicle Highway Sys-

tems Using a Macroscopic Trafﬁc Flow Model. In In-

telligent Transportation Systems, 2009. ITSC’09. 12th

International IEEE Conference on, pages 1–6. IEEE.

BBC (2011). Every death on every road in Great

Britain 1999-2010. http://www.bbc.co.uk/news/

uk-15975564. [Online; acessed 1-July-2013].

Borho, S. (2003). ATLAS: A Language For Modeling

And Simulating Urban Trafﬁc Visualization Of Trafﬁc

Models. PhD thesis, Faculty of Engineering Carleton

University.

Boyandin, I., Bertini, E., Bak, P., and Lalanne, D. (2011).

Flowstrates: An Approach for Visual Exploration

of Temporal Origin-destination Data. In Computer

Graphics Forum, volume 30, pages 971–980. Wiley

Online Library.

Brandt, J., Christensen, J., Frohn, L., and Berkowicz, R.

(2003). Air Pollution Forecasting from Regional to

Urban Street Scale Implementation and Validation for

Two Cities in D enmark. Physics and Chemistry of

the Earth, Parts A/B/C, 28(8):335 – 344.

Buch, N., Velastin, S., and Orwell, J. (2011). Intelli-

gent Transportation Systems, IEEE Transactions on,

title=A Review of Computer Vision Techniques for the

A nalysis of Urban Trafﬁc, 12(3):920–939.

Button, K. J. (2001). Handbook of Transport Systems and

Trafﬁc Control, volume 3. Emerald Group Publishing.

Caliper (2012). TransCad Transportation Planning Soft-

ware. http://www.caliper.com/tcovu.htm. [On-

line; acessed 1-July-2013].

Chen, C. (2005). Top 10 Unsolved Information Visualiza-

tion Problems. Computer Graphics and Applications,

IEEE, 25(4):12–16.

Chen, G., Esch, G., Wonka, P., Mulller, P., and Zhang, E.

(2008). Interactive Procedural Street Modeling. In

ACM Transactions on Graphics (TOG), volume 27,

page 103. ACM.

Chen, X. (2011). Isochronic Singapore.

http://xiaoji-chen.com/blog/2011/

isochronic-singapore/. [Online; acessed

1-July-2013].

Chen, X. et al. (2011). Seeing Differently: Cartography for

Subjective Maps Based on Dynamic Urban Data. PhD

thesis, Massachusetts Institute of Technology.

Chen, X., Santos-Neto, E., and Ripeanu, M. (2012). Crowd-

sourcing for On-street Smart Parking. In Proceedings

of the Second ACM International Symposium on De-

sign and Analysis of Intelligent Vehicular Networks

and Applications, pages 1–8. ACM.

Claramunt, C., Jiang, B., and Bargiela, A. (2000). A New

Framework for the Integration, Analysis and Visual-

isation of Urban Trafﬁc Data Within Geographic I

nformation Systems. Transportation Research Part

C: Emerging Technologies, 8:167 – 184.

Cruz, P. and Machado, P. (2011). Visualizing the Circula-

tory Problems of Lisbon. In ACM SIGGRAPH 2011

Posters, page 92. ACM.

Cruz, P. M. (2011). Morphing City. http://fmachado.

dei.uc.pt/featured/visualizing-traffic.

[Online; acessed 1-July-2013].

Cruz, P. M. (2012). Data Lenses. http://pmcruz.com/

information-visualization/data-lenses. [On-

line; acessed 1-July-2013].

Dargay, J., Gately, D., and Sommer, M. (2007). Vehicle

Ownership and Income Growth, Worldwide: 1960-

2030. The Energy Journal, pages 143–170.

Elbir, T. (2004). A GIS Based Decision Support System

for Estimation, visualization and Analysis of Air Pol-

lution for Large T urkish cities. Atmospheric Environ-

ment, 38(27):4509 – 4517.

Elbir, T., Mangir, N., Kara, M., Simsir, S., Eren, T., and

Ozdemir, S. (2010). Development of a GIS-based De-

cision Support System for Urban Air Quality Manage-

ment in the City of Istanbul. Atmospheric Environ-

ment, 44(4):441 – 454.

Fayyad, U. M., Piatetsky-Shapiro, G., Smyth, P., and Uthu-

rusamy, R. (1996). Advances in Knowledge Discov-

ery and Data Mining.

Gershon, N., Eick, S. G., and Card, S. (1998). Information

Visualization. Interactions, 5(2):9–15.

Gunay, B. (2009). Using Wavelet Transforms for Better In-

terpretation of Trafﬁc Simulation. Trafﬁc Engineering

and Control, 50(10).

Ha, H.-H. and Thill, J.-C. (2011). Analysis of Trafﬁc Haz-

ard Intensity: A Spatial Epidemiology Case Study of

Urban Pedestrians. Computers, Environment and Ur-

ban Systems, 35(3):230 – 240.

Howe, J. (2006). The Rise of Crowdsourcing. Wired Mag-

azine, 14(6):1–4.

Huang, B. and Pan, X. (2007). GIS Coupled with Traf-

ﬁc Simulation and Optimization for Incident Re-

sponse. Computers, Environment and Urban Systems,

31(2):116–132.

Jradi, W., do Nascimento, H. A. D., Longo, H., and Hall,

B. R. (2009). Simulation and analysis of urban traf-

ﬁc the architecture of a web-based interactive decision

support system. pages 1–6.

Keim, D. A. and Kriegel, H.-P. (1996). Visualization Tech-

niques for Mining Large Databases: A Comparison.

Knowledge and Data Engineering, IEEE Transactions

on, 8(6):923–938.

Kim, K., Oh, S., Lee, J., and Essa, I. (2009). Augment-

ing Aerial Earth Maps with Dynamic Information. In

Mixed and Augmented Reality, 2009. ISMAR 2009.

8th IEEE International Symposium on, pages 35–38.

IEEE.

Kloeckl, K., A., A., R., B., and P., C. (2012). Traf-

ﬁc Origins. http://senseable.mit.edu/

visual-explorations-urban-mobility/

traffic-origins.html. [Online; acessed 1-

July-2013].

Kloeckl, K., Di Lorenzo, G., SENN, O., and Ratti, C.

(2011). LIVE Singapore!-an Urban Platform for

Real-time Data to Program the City. Proceedings of

CUPUM.

Krajzewicz, D., Hertkorn, G., Rossel, C., and Wagner,

P. (2013). Simulation of Urban Mobility - SUMO.

VisualizingLargeScaleVehicleTrafficNetworkData-ASurveyoftheState-of-the-art

345

http://sumo.sourceforge.net/. [Online; acessed

1-July-2013].

Li, L., Zhu, L., and Sui, D. Z. (2007). A GIS-based

Bayesian Approach for Analyzing Spatial Temporal

Patterns of Intra-city Motor Vehicle Crashes. Journal

of Transport Geography, 15(4):274 – 285.

Liu, R., Vilet, D. V., and Watling, D. P. (2008). Dy-

namic Route Assignment Combining User Learning

and microsimulation - DRACULA . http://www.

its.leeds.ac.uk/software/dracula/. [Online;

acessed 1-July-2013].

Liu, Y., Wang, F., Xiao, Y., and Gao, S. (2012). Urban

Land Uses and Trafﬁc Source-sink Areas: Evidence

from GPS-enabled Taxi Data in Shanghai. Landscape

and Urban Planning, 106(1):73 – 87.

Madsen, C., Carlsen, K. C., Hoek, G., Oftedal, B., Nafstad,

P., Meliefste, K., Jacobsen, R., Nystad, W., Carlsen,

K.-H., and Brunekreef, B. (2007). Modeling the Intra-

urban Variability of Outdoor Trafﬁc Pollution in Oslo.

Atmospheric Environment, 41(35):7500 – 7511.

Mahut, M. and Florian, M. (2010). Trafﬁc Simulation With

Dynameq. Fundamentals of Trafﬁc Simulation, pages

323–361.

McNally, M. G. (2000). The Four-step Model. Handbook

of Transport Modelling, 1:35–41.

Mesbah, M., Currie, G., Lennon, C., and Northcott, T.

(2012). Spatial and Temporal Visualization of Tran-

sit Operations Performance Data at a Network Level.

Journal of Transport Geography, 25(0):15 – 26.

Meyer, M. D. and Miller, E. J. (2001). Urban Transporta-

tion Planning: A Decision-oriented Approach.

MitSim (2013). MitSim. http://sourceforge.net/

projects/mitsim/. [Online; acessed 1-July-2013].

Outram, C., Ratti, C., and Biderman, A. (2009).

Copenhagen Wheel. http://senseable.mit.edu/

copenhagenwheel/. [Online; acessed 1-July-2013].

Pamanikabud, P. and Tansatcha, M. (2010). Virtual Visual-

ization of Trafﬁc Noise Impact in Geospatial Platform.

In Geoinformatics, 2010 18th International Confer-

ence on, pages 1–6.

PTV (2013). PTV Vissim . http://vision-traffic.

ptvgroup.com/en-uk/. [Online; acessed 1-July-

2013].

Reades, J. (2011). Pulse of the City. http:

//simulacra.blogs.casa.ucl.ac.uk/2011/

08/pulse-of-the-city/. [Online; acessed

1-July-2013].

Reades, J. (2012). Pulse of the City Reboot.

http://simulacra.blogs.casa.ucl.ac.uk/

2012/05/pulse-of-the-city-reboot/. [Online;

acessed 1-July-2013].

Seong, J. C., Park, T. H., Ko, J. H., Chang, S. I., Kim, M.,

Holt, J. B., and Mehdi, M. R. (2011). Modeling of

Road Trafﬁc Noise and Estimated Human Exposure

in Fulton County, Georgia, USA. Environment Inter-

national, 37(8):1336–1341.

Sewall, J., Wilkie, D., and Lin, M. C. (2011). Interactive hy-

brid simulation of large-scale trafﬁc. In ACM Transac-

tions on Graphics (TOG), volume 30, page 135. ACM.

Silva, J. A., R., M., J., V., and A., S.

(2010). CityMotion Project. http://www.

mitportugal.org/transportation-systems/

intelligent-transportation-systems.html.

[Online; acessed 1-July-2013].

Svennerberg, G. (2010). Beginning Google Maps API 3.

Apress.

Talele, T., Pandit, G., and Deshmukh, P. (2012). Dynamic

Ridesharing Using Social Media. International Jour-

nal.

Tang, U. and Wang, Z. (2007). Inﬂuences of Urban Forms

on Trafﬁc-induced Noise and Air pollution: Results

From a Modelling System. Environmental Modelling

and Software, 22(12):1750 – 1764.

Taylor, M. A., Woolley, J. E., and Zito, R. (2000). Integra-

tion of the Global Positioning System and Geographi-

cal Information Systems for Trafﬁc Congestion Stud-

ies. Transportation Research Part C: Emerging Tech-

nologies, 8(1):257–285.

Thirion, S., C., C., J., H., and R., M. (2008). Cas-

cade on Wheels. http://medialab-prado.es/

article/visualizar_muestra_de_resultados_

del_taller. [Online; acessed 1-July-2013].

Tsai, K.-T., Lin, M.-D., and Chen, Y.-H. (2009). Noise

Mapping in Urban Environments: A Taiwan Study.

Applied Acoustics, 70(7):964 – 972.

Van Vliet, D. (2013). Simulation and Assignment of Trafﬁc

to Urban Road Networks - SATURN . http://www.

saturnsoftware.co.uk/. [Online; acessed 1-July-

2013].

Wang, H., Zou, H., Yue, Y., and Li, Q. (2009). Visualizing

Hot Spot Analysis Result Based on Mashup. In Pro-

ceedings of the 2009 International Workshop on Loca-

tion Based Social Networks, LBSN ’09, pages 45–48,

New York, NY, USA. ACM.

Wang, X. (2005). Integrating GIS, Simulation Models, and

Visualization in Trafﬁc Impact Analysis. Computers,

Environment and Urban Systems, 29(4):471 – 496.

Waze (2013). Waze. http://www.waze.com/. [Online;

acessed 1-July-2013].

Weissman, D. and Villalobos, M. (2012). Mobility Apps.

Wood, Jo Slingsby, A. and Dykes (2008). Using Treemaps

for Variable Selection in Spatio-temporal Visualisa-

tion. Information Visualization, 7(3-4):210–224.

Wood, J., Dykes, J., and Slingsby, A. (2010). Visualisation

of Origins, Destinations and Flows with OD Maps.

Cartographic Journal, The, 47(2):117–129.

Wood, J., Slingsby, A., and Dykes, J. (2011). Vi-

sualizing the Dynamics of London’s Bicycle-Hire

Scheme. Cartographica: The International Jour-

nal for Geographic Information and Geovisualization,

46(4):239–251.

Xie, Z. and Yan, J. (2008). Kernel Density Estimation of

Trafﬁc Accidents in a Network Space. Computers,

Environment and Urban Systems, 32(5):396–406.

Zannin, P. H. T., Engel, M. S., Fiedler, P. E. K., and Bunn,

F. (2013). Characterization of Environmental Noise

Based on Noise Measurements, Noise Mapping and

Interviews: A Case Study at a University Campus in

Brazil. Cities, 31(0):317 – 327.

IVAPP2014-InternationalConferenceonInformationVisualizationTheoryandApplications

346