Vehicular Communications 2 (2015) 70–79

Contents lists available at ScienceDirect

Vehicular Communications
www.elsevier.com/locate/vehcom

Invited paper

A survey of vehicular communications for traffic signal optimization

Ryan Florin ∗ , Stephan Olariu
Department of Computer Science, Old Dominion University, Norfolk, VA 23529, USA

a r t i c l e i n f o a b s t r a c t

Article history: Traffic congestion is an ever increasing problem on our roadways and city streets. The main contribution
Received 10 November 2014 of this survey is a taxonomy of adaptive traffic signal control strategies achieved through various levels
Received in revised form 27 February 2015 of vehicular communications.
Accepted 5 March 2015
Strategies to optimize traffic signals fall into three categories based on the level of vehicle involvement.
Available online 9 March 2015
The first category involves those strategies that utilize legacy devices with no vehicular involvement. The
Keywords: second category comprises those strategies that utilize vehicles on the road to wirelessly transmit data
Traffic signal optimization about themselves (e.g. location, velocity). The third category involves strategies that utilize the vehicles’
Adaptive traffic signal control on-board computation power to help optimize traffic signals. The bulk of this survey deals with the
Vehicular cloud second category as it appears to be the most prevalent in the research literature. We are however, quick
VANET to point out that the third category seems to be gaining momentum, as the prevalence of smartphones
has suggested supplementing legacy traffic monitoring with traffic-related reports submitted by the
driving public.
© 2015 Elsevier Inc. All rights reserved.

1. Introduction 1. Promote the use of public transportation and carpooling,

In 2011, in the United States there were over 250 million vehi-
cles criss-crossing four million miles of roadways [1]. Since most
US roadways function close to capacity, congestion is a common
occurrence triggered by chance fluctuations in traffic flow. Recent
statistics from the US Department of Transportation (US-DOT) and
National Highway Traffic Safety Administration (NHTSA) have re-
vealed that over half of all roadway congestion is caused by traffic-
related incidents and poor traffic light scheduling, rather than by
recurring rush-hour traffic in big cities [2–4]. In fact, 27% of the
total delay due to congestion is experienced in non-peak times [5].
Worse yet, according to the NHTSA, congested roadways and
city streets are the leading cause of tens of thousands of traffic-
related fatalities [6]. In addition to the loss of life and property
caused by traffic accidents, the US-DOT has shown that congestion
costs the nation, year after year, over 2.9 billion gallons of wasted
fuel, 5.5 billions hours of lost productivity, and releases 56 bil-
lion pounds of carbon dioxide [5]. For example, in 2011 individual
drivers experienced an average of 38 hours of sitting in congested
traffic.
There are four basic approaches to the problem of preventing
congestion or mitigating its effect:
* Corresponding author.
E-mail addresses: rflorin@cs.odu.edu (R. Florin), olariu@cs.odu.edu (S. Olariu).
2. Increase the bandwidth of the current roadways,
3. Increase average speed, and
4. Decrease the travel delay on the existing roadways.
The focus of the first category of approaches is to reduce the
number of vehicles on our roadways by increasing the number of
passengers in each vehicle through the use of buses and carpools,
or by removing them altogether through the use of light rail. This
category is further partitioned into two sub-categories, those that
remove vehicles by placing drivers and passengers into other vehi-
cles, and those that place drivers and passengers into non-roadway
transportation, such as light rail. Carpooling and bus systems both
use existing roadways, but increase the passenger count per vehi-
cle. Alternatively, rail systems remove vehicles from the roadway
completely. In both cases, the number of vehicles is reduced, thus
effectively reducing traffic flow intensity. Both are worthwhile con-
siderations for any municipality, but are not within the scope of
this survey.
The second category of approaches is concerned with adding
more roadway infrastructure to give vehicles alternate paths and
lanes to get to their destination. In effect, this approach proposes
to increase the bandwidth of the roadway system. In the past,
adding more lanes or new roads has been a strategy for dealing
with traffic, but at a great cost. According to the Florida Depart-
ment of Transportation, adding a single lane to an existing road
costs between $1.5 million and $4.75 million per mile [7]. Adding

http://dx.doi.org/10.1016/j.vehcom.2015.03.002
2214-2096/© 2015 Elsevier Inc. All rights reserved.
 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
a new four-lane roadway costs as much as $13.5 million per mile.
In certain circumstances, providing additional roads is necessary;
however, there are other alternatives that must be considered first.
The third category involves approaches focused on increasing
the average vehicular speed. This can be done by increasing the
posted speed limits, by creating a set of dedicated lanes with an
increased speed limit (e.g., High Occupancy Vehicle (HOV) lanes),
or by platooning. Due to safety reasons, increasing speed limits can
only be considered under certain circumstances. HOV lanes are an
additional set of lanes with added cost, and are usually restricted
to highways and, under current practice, only to peak hours. Pla-
tooning was defined in [8] as “a collection of vehicles that travel
together, actively coordinated in formation”. In theory, platooning
offers better roadway utilization, higher throughput, improved gas
mileage, and increased safety [9,10]. While interesting in their own
right, the approaches in this category are not within the scope of
our survey.
Finally, the last category of approaches strives to minimize
travel delay on existing roads, either by routing the vehicle to
the shortest path or by retiming the traffic signals to optimize
flow and, thus, decrease delay. One approach to implementing this
latter strategy is to employ guidance systems that provide the
driving public with information on current traffic conditions, ex-
pected travel times, delays, road construction and the like. The
idea is using this information, the drivers can decide for them-
selves what alternative best suits their needs. Systems that display
travel times on overhead highway signs have been utilized in Eu-
rope and Japan for more than three decades. Recently, the high
penetration of smartphones has made it possible to provide the
drivers with the latest traffic conditions. Google Maps, 511 Traffic
[11], and WAZE [12] are examples of applications that can display
the current traffic conditions on a smart phone. The next step in
these guidance systems is to predict future traffic conditions, and
to offer individualized guidance information to best reduce travel
time of individual drivers. Although guidance is an interesting and
worthwhile topic, the rest of this paper will focus on optimizing
traffic signals.
The remainder of this survey is organized as follows. Section 2
offers basic concepts of traffic signals and vehicular communi-
cations. Of particular interest is the role played by vehicle-to-
infrastructure communication in providing the traffic controller
with real-time information so that it can provide the best timing
for the traffic signals. Specifically, in Section 3 we describe strate-
gies that require no contribution from the vehicles. In Section 4 we
describe strategies that involve vehicular communications where
each vehicle provides information about itself. Further, in Section 5
we describe strategies where the vehicles play an active part in de-
termining the timing of traffic signals. Section 6 offers an in-depth
discussion of the most representative papers surveyed. Section 7
presents our vision of the future of traffic signal optimization. Fi-
nally, the paper concludes with Section 8 that offers concluding
remarks.
2. Background on traffic monitoring, traffic signals and vehicular
communications
The main goal of this section is to present background mate-
rial that may be useful in understanding the technical presentation
in the remainder of the paper. With this in mind, Subsection 2.1
introduces basic concepts of traffic monitoring, Subsection 2.2
presents basics of traffic signal optimization, and Subsection 2.3
presents the rudiments of vehicular communications.
71
2.1. Traffic monitoring
Legacy traffic monitoring and incident detection techniques,
that are still in widespread use today, employ Inductive Loop De-
tectors (ILD), video detection systems, acoustic tracking systems
and microwave radar sensors [13,14]. By far the most prevalent are
the ILDs, which are placed on the roadway every mile or so. The
ILDs measure traffic flow by registering a signal each time a vehi-
cle passes over them. Each ILD, including hardware and controllers,
costs around $8200; in addition, adjacent ILDs are connected by
optical fiber that costs $300,000 per mile [15]. It is well docu-
mented [16] that the legacy equipment installed in support of
collecting traffic-related data is expensive and costly to maintain
and repair. Not surprisingly, transportation departments worldwide
are looking for less expensive and more reliable solutions for traf-
fic monitoring and incident detection [10].
To be effective, innovative traffic-event detection systems must
enlist the help of the most recent technological advances.
For example, recent advances in sensor technology have pro-
duced cement-based piezoelectric sensors that do not corrode,
cannot be damaged by thermal expansion of the road, and can
be made of inexpensive materials [17]. These sensors can be em-
bedded in the roadway and detect vehicles like ILDs. They have
been the basis of the NOTICE system [18] that involves embed-
ding intelligent sensor belts in roadways and using these belts to
detect traffic-related events ranging from congestion to lane ob-
structions and potholes. NOTICE has a great deal in common with
ILDs since both systems are intrusive and contribute to weakening
the structural integrity of roadways. Extrapolating from past expe-
rience with ILDs, sensor belts embedded in the roadway are very
likely to suffer from reliability problems and to contribute to the
creation of potholes.
An idea that exploits the prevalence of smartphones is to sup-
plement legacy traffic monitoring with traffic incident reports sub-
mitted by the driving public. A recent implementation of this idea
has lead to 511 Traffic that offers an at-a-glance view of road con-
ditions in a given geographic area [11]. Unfortunately, 511 Traffic
is a centralized system that accumulates and aggregates traffic-
related feeds at Traffic Monitoring Centers and, due to inherent
delays often displays stale traffic information [19].
Finally, the Mobile Millennium project at UC Berkeley exploits
information collected by probe vehicles to infer information about
the traffic [20]. Relying solely on traffic data collected by probe ve-
hicles seems to work best in urban environments that experience
high concentration of vehicles and less well on highways where
there may be no “critical mass” of probe vehicles [21].
2.2. Basics of traffic signal optimization
The reason traffic signals exist is to assign the right-of-way at
intersections. Control of a particular direction is partitioned into
three traffic light phases: green, yellow, and red. The interval from
when the green phase begins and the red phase ends is the cycle
time. The phase and cycle timings for each direction are controlled
by a traffic signal controller.
Most traffic signals in the US run a set of predefined timing
plans that set the signal’s cycle length and green phase length
based on the time of the day. In most cases, the optimization of
the signal systems currently occurs off-line at either the isolated
intersection or corridor level. One of the major disadvantages of
this approach is that it requires data on traffic-turning movements
be regularly collected to develop optimized traffic signal plans off-
line. A second major disadvantage is that the time-of-day based
signal timings do not adapt well to unexpected changes in traf-
fic demand. For example, if an incident on the roadway network
72 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
causes travel patterns to change significantly, the signals often can-
not fully accommodate the changes in flow, resulting in traffic
buildup and congestion. In order to ensure that the signals func-
tion as well as possible, they have to be retimed regularly to reflect
current conditions. A retiming of the traffic signal is suggested at
least every three years [22]. Unfortunately, due to budget or man-
power limitations, transportation agencies often neglect to retime
signals resulting in unnecessary delays to the traveling public. As
recently as 2011, the Federal Highway administration estimated
that more than 75% of the country’s 330,000 traffic signals are op-
erating with outdated or uncoordinated signal timing plans [5].
Traffic signals can be a source of significant delays if cycle and
phase lengths are not suitable for current traffic conditions. Under
current practice, the process of developing optimal signal timing
plans is resource-intensive. First, technicians must go into the field
to manually collect vehicle and pedestrian volumes, types, and
speeds during normal usage and peak usage times. Additionally
collision records are collected and analyzed to determine if varying
the signal can make the intersection safer [22]. Then, this data is
input into commercial signal optimization software packages such
as Synchro or CORSIM [23] to develop optimal timing plans of-
fline.
One solution that has been proposed is the use of Adaptive
Traffic Signal Control (ATSC). ATSC systems use extensive detection
to dynamically optimize flow along a corridor. These systems often
use no fixed cycle of phase lengths and retime signals continually
based on observed traffic flow [24]. There is often a local optimiza-
tion that occurs to minimize delays at an individual intersection
and a secondary global optimization that occurs along a series of
signals on an arterial [21]. While some ATSC systems have been in
existence for over two decades, they have not been adopted on a
wide scale. More cost-effective systems have been developed with
good results on corridor-level deployments [23]. However, due to
their complexity, they have not been deployed beyond the corridor
level [13].
2.3. Basics of vehicular communications
The US Federal Communications Commission has allocated
75 MHz of spectrum between 5.850 GHz and 5.925 GHz for use by
Intelligent Transportation Systems (ITS), and it is known as Ded-
icated Short Range Communications (DSRC) [25,26]. The amount
of bandwidth in this range has been shown to exceed the needs
for its use by traffic safety applications. As such, several appli-
cations have emerged to utilize the additional bandwidth. These
application included, initially, peer to peer (p2p) networking, and
multimedia content delivery. Later, location-based services, on-line
banking, gaming, and mobile entertainment have been included.
In the future we can expect more applications as the supporting
infrastructure matures.
As per DSRC [26] and IEEE 802.11p/WAVE [27,28], DSRC spec-
trum is divided into seven 10 Mhz-wide channels. There is a single
Control Channel (CCH) used for beacon messages, emergency mes-
sages, and service advertisements. There are six additional chan-
nels to be used for non-safety messages; these are Service Chan-
nels (SCH) that can be used by Roadside Units (RSUs). During the
Control Channel Interval (CCHI), the wireless transceiver must be
tuned to the CCH for broadcast updates and notifications from
neighbors and RSUs. During the Service Channel Interval, the wire-
less transceiver can be tuned to one of the six other SCH.
Some vehicles on the road are already equipped with radar,
blind spot indicators, and lane change notification. Cars of the
future can expect to have more sensors, computing and storage
devices to process the real-time data provided by these sensors,
and wireless transceivers to communicate sensor data to the RSUs.
With DSRC, vehicles on the road will organize into networks of
vehicles able to communicate traffic conditions, notify drivers of
traffic safety alerts, and deliver Internet access to the vehicle.
Networks that consist of vehicles on the road communicating
via wireless transceivers are known as Vehicular Ad Hoc Networks
(VANETs). The basic communication in these networks comes in
two flavors: Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure
(V2I). In V2V communications, also known as zero-infrastructure,
each vehicle is responsible for communicating with other vehi-
cles directly, without relying on roadside infrastructure. This is the
most basic form of vehicular communications and has been shown
to suffer from a large array of security problems. On the other
hand, V2I communications are predicated on the existence of pre-
installed roadside communications that replace or supplement the
purely V2V communication regimen.
Systems utilizing VANETs have been found to use the follow-
ing typical architecture, which we refer to as the VANET-based
approach: a Vehicle Agent (VA) transmits important vehicle data to
an Intersection Agent (IA). This IA is connected directly to the traf-
fic signal controller. The IA/traffic signal controller aggregates the
vehicle data to make a decision to change the traffic signal timing.
In most cases, this IA is the RSU as specified above. The main dif-
ferences in these systems are specific pieces of data retrieved from
vehicles, and how it is used to retime the signals. The papers men-
tioned in Sections 4 and 5 include varying degrees of detail of how
the sensors actually collect data. Some specifically detail how the
sensors work, while others leave these details out.
3. Strategies without vehicular involvement
The strategies discussed in this section are those which employ
additional infrastructure to detect vehicles on the road. This new
information is used by the traffic signal controller to make better
decisions in controlling the traffic signals. Note that these strate-
gies involve no contributions from the vehicles, other than their
participation in the roadway traffic; the vehicles are merely by-
standers being counted by a separate system.
The most basic strategy is the actuated traffic signal. A sensor
is embedded in the road surface to indicate to the traffic signal
controller when a vehicle is at the intersection. The traffic sig-
nal controller will make a decision to give green time to that
direction. Actuated traffic signals come in three forms of control,
Semi-Actuated, Full-Actuated, and Volume-Density. Semi-Actuated
is where the sensors are only on the direction of the side streets.
The side streets default to red, and only turn green when a vehi-
cle is detected by the sensor. Full-Actuated is where all directions
have sensors. The traffic signal controller makes decisions based on
the data supplied by the sensors. Volume-Density is similar to Full-
Actuated, but it uses more information from the sensors to make
its decision. For example, the traffic signal controller may choose
to remain green until a variable minimum gap between vehicles is
achieved [29].
In the above scenarios, the most common sensor in use today is
the ILD. Also in use are microloops, magnetic detectors, ultrasonic
sensors, and radar. Some systems work by signaling a simple on or
off. Other systems, as employed by the Volume-Density scenario
above, require more sophistication. By recording the times of when
the sensors are on and off, the distance between vehicles can be
determined as well as other details [14].
The basis for retiming a traffic signal comes from the knowl-
edge of the traffic on the road. Roess et al. [14] mention three ways
this is being done today. The first is by virtual detectors. A vir-
tual detector is a camera system with specialized software able
to count vehicles based on image processing technologies. Using
this approach, vehicle count, velocity, and density, as well as queue
length of the vehicles can be determined. Second, microwave de-
tectors are used to determine when a vehicle passes a particular
 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
point on the road. Third are wireless detectors that behave much
like ILDs, except the information is communicated to a controller
wirelessly. One interesting approach of wireless detectors is pro-
posed by Kwong et al. [30]; they utilize the magnetic detectors
previously mentioned. Seven sensors are embedded within each
lane a foot apart, perpendicular to the direction of the road. The
sensors communicate wirelessly to an access point nearby. A simi-
lar set of sensors is located at a neighboring intersection on the
same stretch of road. The system will record the magnetic sig-
nature and a timestamp for each vehicle as it passes over the
sensors. As the vehicle passes over the other set of these sensors
on the same stretch of road, the magnetic signature is matched
and the timestamp is used to determine the velocity of the ve-
hicle between the two points. Such a system is known as vehicle
re-identification. The traffic information collected by such a sys-
tem can be used by traffic signal controllers to make better deci-
sions.
Prasad [31] makes use of virtual detectors in an actuated traf-
fic signal control system. A video camera is used to detect vehicles
through a software algorithm; however, the interesting part is how
the controller determines the retiming. The decision is based on
the number of vehicles from the camera, a learning algorithm run
to utilize historical data in real-time, and a component that de-
tects events such as traffic jams or accidents. No specific details
are proposed for the learning algorithm or the event detection, but
it makes an effort to reduce processing needs of the traffic sig-
nal controller by moving that cost to the cloud. The term cloud is
used to generally describe many possible services. Here the cloud
is used to describe an Infrastructure as a Service solution; essen-
tially, the processing jobs will be moved to a third party service
provider which offers compute power as a utility to customers. The
computing power offered by this approach is cheaper in compari-
son to the high cost of operating the hardware internally to handle
the same request. This exposes the reader to the need of more
processing power to handle certain algorithms proposed in later
sections of this survey, specifically Section 5. Additionally, Prasad
suggests that the municipality can access the traffic data from the
cloud to help time the signals in a more coordinated fashion. Here
the term cloud describes that the traffic data is saved on third
party that provides data storage as a service.
Each of these strategies is based on using infrastructure to
gather information about the vehicles, but none yet allows vehicles
to participate in the retiming of traffic signals. The next section,
Section 4, considers strategies that utilize the vehicles to supply
information to the roadside infrastructure to aid in the collection
of vehicle data.
4. Strategies where vehicles are passive participants
In Section 2 we described the typical VANET-based approach.
This is the underlying architecture used throughout this section
to enable the traffic signal controller to collect information from
vehicles.
Kari et al. [32] includes some important system bookkeeping
that other papers do not mention; particularly concerning keeping
track of the status of the vehicle in the intersection. When a ve-
hicle is within communication range of the IA it will first check in
if it has not done so already. If the vehicle is leaving the intersec-
tion, it will check out if not already done so. The check-in is done
to inform the traffic signal controller that the vehicle is new and
should be added to the vehicle list. The check-out is done to in-
form the traffic signal controller to remove the vehicle from the
vehicle list.
An alternate approach to wireless transceivers is given by Bhu-
vaneswari et al. [33]. Here, the VA is a passive Radio Frequency
Identification (RFID) tag and the IA is an RFID reader. Such an ap-
73
proach places limits on the solution. Specific limitations include
the range of the communication, communication can only be one
way, and typically only an ID can be stored on a passive RFID tag.
Interestingly, Bhuvaneswari et al. [33] states vehicle information
such as velocity is stored in a remote database. No details are given
into how this information is stored into the remote database.
In the following sections, wireless transceivers utilizing DSRC
will be assumed as the method of communication between VA
and IA. The next subsections investigate further differences in the
VANET-based approach over the traditional approaches.
4.1. Advantages of VANET-based approach
A VANET-based approach offers several advantages over the
infrastructure-based approaches from Section 3. The wireless
transceiver allows the vehicle information to be collected any-
where within the transmission range of the IA. This allows the
traffic signal controller to have a count of all vehicles over a large
area, instead of one single spot on the road. A second advantage
is that the vehicle can now give the IA additional information to
enable the traffic signal controller to make more informed deci-
sions. In most systems surveyed the VA will transmit the location
of the vehicle. With this information the traffic signal controller
can plot the vehicle on the road, instead of just knowing the num-
ber of vehicles on the road. Some systems surveyed [34–36] also
require vehicles to send their velocity. This provides the traffic sig-
nal controller with a rough estimate of when the vehicle will reach
the intersection, and whether it will reach a green phase. In other
systems the vehicles may send information identifying it as an
emergency vehicle, the type of vehicle, or expected carbon emis-
sions depending on what the optimization algorithm requires. The
IA may also send information to the vehicle, an option the tradi-
tional infrastructure-based approaches do not allow. In the strategy
of Kari et al. [32] the intersection geometry is sent to the vehicle.
It is assumed that this additional information is used by the ve-
hicle to aid in calculating the expected carbon emissions. Finally,
the IA may require a small amount of computation to be handled
by the vehicle. The strategy of Kari et al. [32] requires the vehicle
to calculate the expected amount of carbon emissions. Also, some
papers ask the vehicle to determine its own expected time of ar-
rival [32,35]. This alleviates some computing needs from the traffic
signal controller. The computation required by the systems in this
section is limited; however, utilizing vehicles to perform computa-
tions could lead to interesting new approaches in the future. Some
of these will be discussed later in Section 6.7.
4.2. Disadvantages of VANET-based approach
In most papers surveyed that utilize the VANET-based approach,
each vehicle is assumed to have a wireless transceiver. The sys-
tems would degrade if non-enabled vehicles were considered in
the traffic. Few papers surveyed suggested any approaches to han-
dle a vehicle without a wireless transceiver, or a vehicle with a
malfunctioning wireless transceiver.
A particularity concerning disadvantage that is not discussed in
detail in the literature is that traditional ILDs offer a glimpse of
traffic at a single point and, consequently, it can determine a count
of traffic per lane. A vehicle attempting to locate itself on the road-
way is unable to determine its location with this accuracy. Kari et
al. [32] and Wenjie et al. [34] suggest that the lane can be de-
termined by GPS or by triangulation, but evidence is not provided
that supports these claims. These strategies rely on knowing the
exact lane; the effectiveness of each is degraded if the vehicle’s
lane is approximated. More discussion on this is provided later in
Section 6.7.
74 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
The final disadvantage is the need for more computing power
from the traffic signal controller. Current traffic signal controllers
make decisions based on fixed timings or from inputs from a small
number of sensor inputs. Many of the algorithms suggested in this
section are only slightly more complex; however, there is a cate-
gory explained later in Section 4.4 which requires a huge amount
of processing power. The computing power required of the traffic
signal controller must be considered.
4.3. Reactive versus predictive traffic signal optimization
One way to handle traffic on the roads is to first determine
the traffic at intersections, and then retime the traffic signal based
on these conditions. Such a system is said to be reactive. In the
infrastructure-based strategies, most systems are reactive because
they adjust phase and cycle lengths based on the number of ve-
hicles waiting at the intersection. VANET-based approaches offer a
bit of intelligence over purely reactive systems. Many require the
VA to send the expected time of arrival (at the intersection), or
information to determine location and velocity. With this informa-
tion, the traffic signal controller can determine when each vehicle
will arrive at the intersection. The algorithms use this information
to reduce times waiting at the intersection. In these cases, the level
of prediction is determined by the range of the IA’s wireless signal
[31–35,37].
McKenney and White [38] proposed a system that offloads pre-
diction to the location of the previous intersection by allowing
communication between intersections. The data for vehicles pass-
ing through an intersection is communicated to the neighboring
intersection along the vehicle’s route. In this approach, the data
collected at each intersection is shared with neighboring intersec-
tions to increase the level of prediction.
In Predictive Microscopic Simulation Algorithm (PMSA) [36],
each of the vehicles on the road communicate their location, head-
ing, and speed to the traffic signal controller. Those within 300 me-
ters of the intersection are mapped into a microscopic simulation
to determine the whereabouts of each vehicle after a horizon of
15 seconds. The algorithm bases whether the vehicle will turn or
go straight on the lane they are in. For example, a vehicle in the
turn lane will likely turn, a vehicle in the straight lane will likely
not turn. For lanes that may go straight or turn, an equal ratio is
used to make a prediction. After 15 seconds another simulation is
run to again predict the vehicle movements. In this algorithm, the
traffic signals are retimed based on the vehicle’s predicted loca-
tion.
CoSIGN [39] is a similar idea as PMSA; however, the authors
offer a pseudo-predictive system. The vehicles on the road are
simulated with generated traffic data. With a few modifications,
CoSIGN can be predictive. At time zero the traffic mapped by the
simulation is set to the current position of the vehicles on the road.
The future paths of the vehicles are predicted based on the current
lane and on historical intersection percentages. For example, at a
given intersection i% of the vehicles turn left, j% of the vehicles
go straight, and k% of the vehicles turn right, such statistics can
be updated in real-time using communications between vehicles
and the infrastructure. The IA transmits a welcome beacon to all
new vehicles within range. The VA of a new in-range vehicle will
transmit its location and current speed. The IA will estimate fu-
ture positions, based on the vehicle’s speed and location, and will
predict the direction the vehicle will take based on historical data.
The IA will regularly share these predictions with its immediate
neighboring intersections and will collectively run short-term sim-
ulations to predict traffic for the next few minutes. The simulations
will be short, localized, and may be able to run in real-time to best
optimize the signals based on prediction.
4.4. Isolated versus coordinated traffic signal optimization
Many of the papers surveyed consider traffic signal optimiza-
tion in the context of an isolated intersection. By extending the
algorithm to coordinate on a city-wide network, the “perfect” traf-
fic timing can be determined for the entire network. Finding this
“perfect” timing involves a complex calculation that requires an
amount of computing power that in unattainable under current
practice. Cheng et al. [39] proposed a network-wide algorithm;
however, because of its complexity several shortcuts were taken.
McKenney and White [38] proposed that vehicle information be
shared with its neighbors; however, it does not attempt to coordi-
nate signal timings with neighbors. Most of the surveyed literature
offer strategies in isolation from other signals [32–35,37,36,40].
These algorithms work in isolation, but still manage to get better
results than traditional fixed-time signals.
Perhaps an effective approach would require neighboring traffic
signal controllers to coordinate their cycles and phases. Without
any coordination, at each signal retiming, the traffic signal con-
troller must consider all incoming roads and create an effective
plan for each. Suppose an intersection with four approaches, North,
South, East, and West. Without coordination, it is possible that
each neighboring intersection is sending vehicles towards this in-
tersection at the same time. The algorithm will run and will either
give the North/South the green light, or the East/West the green
light. In either case, some vehicles will have to be stopped. Now
consider a plan with some coordination. Suppose the traffic signal
controllers at each intersection coordinate their cycles and phases
such that East and West vehicles arrive at the intersection at alter-
nating times than the North and South vehicles. This reduces the
cases where the intersection is causing a group of vehicles wait.
5. Strategies where vehicles are active participants
In the previous section, some systems offloaded certain calcula-
tions to the vehicles on the road. These calculations were limited
to simple ones, e.g. estimation of arrival time to the next intersec-
tion and estimate of carbon emissions. In this section, we present
approaches where the vehicles are responsible for more complex
calculations.
Vehicles on the roadway are being outfitted with more and
more types of sensors, as well as more powerful computers to
aggregate the data collected by these sensors. Additionally, with
the advent of initiatives such as US Department of Transportation
Connected Vehicle Research [41] vehicles of the future will com-
municate with each other and with the roadside infrastructure.
In the previous sections, we covered strategies where the vehicle
is expected to supply information about its location or its speed
to help the traffic signal controller make a decision. In this sec-
tion we explore alternative approaches to enlisting the vehicles
computational capabilities. We believe the future of traffic signal
optimization lies in offloading more of the computing process to
the vehicles on the road.
Maslekar et al. [42] proposed CATS, a system that works in a
way similar to those discussed in the previous section; however,
the density of vehicles on each link is determined by the vehicles
running a clustering algorithm. Clusters are formed based on the
direction the vehicle will take at the upcoming intersection. The
cluster head for each cluster will communicate to the traffic signal
controller the density of the cluster, and the length of the cluster.
The vehicle will exit the cluster once it leaves the intersection. The
interesting work here is that the vehicles themselves determine
the number of vehicles on the road.
Xiang-ya et al. [43] proposed a similar approach where the ve-
hicles are responsible for counting the vehicles on the road. They
offer a traffic information system that can be used to report the
 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
vehicles on a particular stretch of road. The network of roads is
separated into regions by a virtual grid. Streets are split into seg-
ments and each segment is given a level based on the type of road;
expressway, main arterial, minor arterial, collector, or local street.
A cluster head can only be chosen from certain types of street seg-
ments defined as valid. For example, a vehicle on an expressway
may be a poor candidate for cluster head since it will not be in
the cluster for a long time. The vehicle closest to the center of
the grid, and on a valid street segment, is chosen as cluster head
and acts as a location server for a determined duration. Each ve-
hicle on the road will update its local location servers with its ID,
speed, direction, and location upon traveling a set distance thresh-
old. Local location servers are those within an R × R square; all
others are remote location servers. Location servers communicate
with neighboring location servers to send updates on vehicle loca-
tions. The location servers closest to a vehicle will have the most
up-to-date information on the location of that vehicle.
Since each vehicle on the road is responsible to update its lo-
cation server, these location servers will contain all the vehicle
locations within its region of the grid. Each location server will
aggregate vehicle information into a data structure that contains
the street ID, segment ID, and list of vehicles.
The authors explain a traffic information system, but it does not
apply it to any specific application. In the interest of this survey
paper, we apply it to a traffic retiming system. In our proposed
system, the traffic signal controller keeps track of each location
server, and will regularly request a report of incoming traffic. Ad-
ditionally, traffic signal controllers will also report upcoming signal
times to the location server. The location server will contain data
for each vehicle on the road and the upcoming signal times. This
should allow for better predictions of vehicles’ expected waiting
times at each intersection. For such a system to work, the desti-
nation of each vehicle needs to be known or estimated. Assuming
they are not known, intersection statistics can be used to estimate
how many vehicles go straight, turn left, or turn right. If vehicles
on the road reported this information to the traffic signal con-
troller, then these statistics can be kept up to date in real-time.
Future traffic conditions can be predicted by starting with the cur-
rent vehicles locations, estimating their routes using intersection
statistics, and running these locations and routes through a simu-
lation, such as done in CoSIGN [39].
One can even utilize a system such as CoSIGN [39] to retime
the traffic signals across the entire network utilizing this data. As
stated previously, this cannot be done in real-time unless the run-
ning time of the algorithm can be reduced. In addition to the pro-
posed changes to CoSIGN, perhaps we utilize the compute power
of each of vehicles on the road to run a parallel retiming algo-
rithm. Such a system may provide the compute power needed to
run such an algorithm in real-time. This is future work to be ex-
plored by the authors of this survey paper.
6. In-depth discussion of some representative papers
This section provides a summarized explanation of a select few
papers that offer a general insight into the rest of the literature.
6.1. Early wireless sensor strategy
Wenjie et al. [34] proposed an early strategy in using wireless
sensor networks to retime traffic signals. The authors suggest three
types of nodes. The vehicle nodes transmit data to the detector
nodes. The detector nodes collect the data from the vehicle nodes.
The control nodes aggregate the data from each detector node and
use it to optimize the traffic signals. Each vehicle will transmit its
distance from the next intersection and its current velocity. Using
75
this information, the control node can estimate if the vehicle will
be stopped at the signal or will pass through.
The traffic signal controller will cycle through each phase in
turn. If there are no vehicles recognized by the system for a par-
ticular direction, that phase will be skipped. If there are vehicles
waiting at the intersection for a direction, then the traffic signal
controller will calculate how long it will take the queued vehicles
to leave the intersection and will remain green for that amount
of time. Within this time, more vehicles may join the queued ve-
hicles. The traffic signal controller will adjust the green phase to
account for these additional queued vehicles. This process can con-
tinue for a maximum of 90 seconds. The point of this is to keep
the light green as long as there is a vehicle to utilize the intersec-
tion, and turn red as soon as it is no longer being utilized.
Wenjie et al.’s system depends on a pre-determined amount of
time for each vehicle to leave the intersection. Consider a vehi-
cle at the front of the line who notices the green signal late. The
signal will only stay green for the initial set amount of time. The
signal will not be extended even though there are vehicles in the
queue. Even worse, it is not clear if the system will consider these
vehicles in the next cycle. Alternatively, if all the vehicles react
quickly, and get through the intersection quickly, the signal will
not end any sooner. Both of these situations can cause further traf-
fic buildup. Additionally, the system will set timings based on each
lane. If a vehicle is in a turn lane, the paper assumes it can de-
termine this based on the position of the vehicle. The accuracy of
judging position based on a wireless signal is not sufficiently accu-
rate to pinpoint the lane of the vehicle. A better way to determine
location is required for this approach to work as intended.
6.2. Adaptive traffic light controller
Kwatirayo et al. [35] proposed a system that applies a unique
weight to the locations of the vehicles on the road. It has a Traf-
fic Light Controller that receives wireless messages from each of
the vehicles within range. The Traffic Light Controller will be fur-
ther referred to as IA and the vehicle as VA. Each VA will send a
vehicle identifier, the expected arrival time to the intersection, rel-
ative position to the intersection, the lane of the vehicle, and the
velocity of the vehicle. In order to determine what direction re-
ceives the next green phase, the IA will apply a weighting system.
Vehicles closest to the intersection will receive a higher weight,
while vehicles furthest from the intersection will receive a lower
weight. The weighting scheme is linear based on distance from
the intersection. The weights for each approach to the intersection
are calculated and the direction with the largest weight is given
the green phase. In each traffic signal cycle, each direction will be
given a turn before the cycle restarts.
There is some confusion regarding the length of the green
phase. It is our understanding that the signal will remain green
for at least the minimum green time; additionally, it will remain
green until the minimum of the following: decongestion thresh-
old is met, or maximum green time has occurred. In the paper the
decongestion threshold is not well articulated.
The weighting system requires the distance from the intersec-
tion and the lane of each vehicle to be known; however, the au-
thors offer no suggestions as to how this is determined. Also, like
many papers surveyed, vehicles without wireless transceivers, or
those whose wireless transceivers are malfunctioning are not con-
sidered. Finally, the weighting system causes the order of the green
cycles to differ. In certain cases, the first direction to go in one cy-
cle could potentially be the last direction to go in the next cycle.
Also, a similar case can be found where the last direction to go in
one cycle is the first direction to go in the next cycle. This could
potentially lead to unduly high waiting times for some of the ve-
hicles stopped.
76 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
6.3. Penetration rate study
Guler et al. [44] proposed a system which does not require
all vehicles to be equipped with wireless transceivers. When a
vehicle enters the zone of interest, an event is recorded at the
traffic signal controller. The size of this zone is not described in
the paper, but should be within the transmission range of the
wireless transceivers, and should be larger than the largest pos-
sible queue length. This event is used to determine the time the
vehicle will reach the intersection assuming no existing queue,
free-flow velocity, and no deceleration. This value is referred to
as the virtual arrival time. Once the vehicle stops at the intersec-
tion, the vehicle sends its location to the traffic signal controller.
This event is recorded at the traffic signal controller, and is used
to calculate the number of vehicles in the queue and the num-
ber of unequipped vehicles. No details are given as to how this
is done Also, the paper assumes the vehicle location can be de-
termined by the wireless communication infrastructure. Again, the
details are not provided as to how this is done. The number of
unequipped vehicles is calculated by subtracting the number of
equipped vehicles in the queue by the calculated total of vehi-
cles in the queue. There are two approaches to determine the
virtual arrival time of unequipped vehicles. For unequipped vehi-
cles between two equipped vehicles, the virtual arrival time will
be uniformly distributed between the virtual arrival times of the
equipped vehicles. Unequipped vehicles which arrive after the last
equipped vehicle in the queue will be estimated based on histori-
cal data.
The traffic signal controller makes its decisions based on the
virtual arrival times of vehicles from each approach. It will dis-
patch vehicles one at a time based on an objective function. Es-
sentially, it estimates the time it will take a single vehicle to leave
the intersection; to mimic a real traffic signal, a penalty function is
given to keep the algorithm from swapping between access to the
intersection too often. The traffic signal controller sets an amount
of green time for a single vehicle to pass through the intersection;
however, this is based on virtual times. The traffic signal controller
does not actually know if the vehicle passes through within this
time. As with the Early wireless sensor strategy in Subsection 6.1,
drivers taking too long or too short through the intersection may
cause problems. There are some cases where the vehicle may be
given a green signal to go through the intersection before the ve-
hicle even arrives; the longer a vehicle arrives at the queue from
the virtual arrival time has the potential to cause problems with
this approach.
6.4. Cumulative travel-time responsive (CTR) real-time control
algorithm
The CTR algorithm, proposed by Lee et al. [45] attempts to min-
imize the cumulative travel time of vehicles within a segment. For
each vehicle in the segment, the residency time is added to get
the cumulative travel time for the segment. To determine residency
time, each vehicle will send its travel time information in a here-i-
am message sent every 50 ns; meaning, the traffic signal controller
will have to update each vehicles travel time information every
50 ns. This can be reduced if the vehicle sends a check-in message
upon entering the segment. The traffic signal controller associates
a timestamp to specify the time the vehicle entered the segment,
and determines the travel time without the need for constant up-
dates. When the vehicle leaves the intersection, it can check-out
to be removed from the travel time calculation. The traffic signal
controller chooses the green phase such that the cumulative travel
time is minimized.
Like the Penetration rate study in Section 6.3, this paper
also assumes that all vehicles are not equipped with wireless
transceivers. As such, existing infrastructure is assumed to be able
to determine the total vehicle counts for the intersection. The con-
nected vehicle infrastructure is able to determine the number of
connected vehicles. The number of non-connected vehicles is de-
termined by subtracting the connected vehicles from the number
of total vehicles. Estimates are made for non-connected vehicles
based on Kalman filtering.
This research also finds that all vehicles on the road do not
need to be equipped with wireless transceivers. Non-equipped ve-
hicle data can be estimated. Using the proposed algorithm, the
authors find that only a 30% penetration rate is needed to see ben-
efits over traditional traffic signals.
6.5. Coordinated signals (CoSIGN)
The goal of CoSIGN [39] is to reduce the average travel time
for each vehicle in the network. It does this by running several
traffic simulations with different parameters in order to find the
best cycle timings. Such a strategy proves to be too complex com-
putationally, so the problem must be simplified. Normally, this is
solved by utilizing a pre-determined pattern or by limiting the size
of the network. Instead, CoSIGN solves the full network by utilizing
a Sampled Fictitious Play algorithm. Each iteration of the algorithm
is run through a simulation to determine how well the algorithm
performs. Furthermore, the algorithm is parallelized to make it
more efficient. In spite of all these optimizations, the algorithm
takes far too long to run, even when run on a multi-processor clus-
ter.
In the algorithm the cycle time is chosen based on the level of
congestion. For more congestion, the length of the cycles is longer
to reduce the time the intersection is unused. For less congestion,
the length of the cycles is shorter to reduce the time any partic-
ular vehicle will wait at the signal. The range of the cycle time is
limited between a minimum and maximum value. The length of
the green phase given to a link is the cycle time multiplied by the
ratio between the congestion in the given link to the congestion in
all links.
CoSIGN is not a real-time system ready to be used on the roads
today. It does not use actual Origin-Destination (OD) data from ve-
hicles on the road; OD data is a list of origin and destination for
each vehicle on the road. Sometimes the route is also included.
Even a parallel version of the algorithm running on a cluster does
not run sufficiently fast to be used as a real-time solution.
It is possible, perhaps, that with a few additional modifications,
CoSIGN can turn into a usable real-time system. First, the vehicle
locations should match actual vehicles on the road. Second, the ve-
hicle destinations should be known, or estimated. Third, the length
of time simulated can be decreased such that only the next minute
or so is simulated. Finally, since the time span is decreased, the
system can be run between local intersections, or along a corridor
to restrict the number of intersections in the network. This reduces
the size of the problem, and may allow it to be a more real-time
system, given an appropriate level of resources.
6.6. Predictive microscopic simulation algorithm (PMSA)
An approach similar in style to CoSIGN is Predictive Microscopic
Simulation Algorithm (PMSA), proposed by Goodall et al. [36]. Each
vehicle must know its location, heading, and speed. This informa-
tion is sent to the traffic signal controller via wireless connection.
The traffic signal controller plots each vehicle within 300 meters
of the intersection and runs a microscopic simulation to deter-
mine where the vehicles will be in 15 seconds. The simulations
place priority on keeping red signal times to less than 120 sec-
onds and keeping vehicles from blocking intersections. If these are
 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
both satisfied, the algorithm will choose the next phase based on
an objective function.
The algorithm predicts whether the vehicle will turn or not
based on the lane of the vehicle. A vehicle in a turn lane is likely to
turn; whereas, a vehicle in the straight lane is likely to go straight.
In cases where the turn lane and the straight lane are the same,
it will split the directions evenly. With concern the privacy, the al-
gorithm will not track vehicles from one location to another, so it
is unable to determine if a vehicle turns or goes straight. If the
algorithm does track the location of the vehicles, it can track in
real-time the percentages of vehicles that turn and those that go
straight in real-time to make better predictions.
The length of time each simulation takes is not reported in the
paper, so it is unable to determine if this algorithm is a viable op-
tion to be used in a traffic signal controller. It is expected that an
algorithm running microscopic simulations for a large number of
vehicles will be too complex. Unless there are additional comput-
ing resources that can be utilized, algorithms need to be simple
and efficient.
6.7. Common issues and solutions
Common issues found in the surveyed literature include the fol-
lowing:
1. Each vehicle must have a wireless transceiver,
2. The lane and location of the vehicle must be known,
3. Cycle times may not be fair, and
4. The system handles isolated intersections.
Many papers surveyed assume that all vehicles have a wireless
transceiver to communicate with the infrastructure, and do not in-
clude a backup system in case a wireless transceiver fails. Each
system should include a fall back plan for vehicles with malfunc-
tioning wireless transceivers or those vehicles with no wireless
transceivers. A simplistic approach to solve this issue could be to
ensure each direction at an intersection gets a green cycle at least
once every n cycles. Another possible solution could be to utilize
legacy equipment: video cameras, ILDs, etc. In yet another solu-
tion, vehicles on the road can contain some sensing equipment
to help find neighbors that do not have the appropriate wireless
equipment. In such a system, exact number of vehicles may not
be possible; instead estimated numbers would need to be consid-
ered. Comert and Cetin [46] proposed a solution in which probe
vehicles estimate the number of vehicles at an intersection using
only its own vehicle location. Lee et al. [45] and Guler et al. [44]
also makes use of estimations to include the unequipped vehicles
in calculations.
Additionally, in many of the systems surveyed, GPS or triangu-
lation with wireless signals is suggested to determine the exact
location and lane of vehicles. GPS can be expected to provide an
accuracy of 7.8 meters at 95% confidence level [47]. With this ac-
curacy, GPS can provide an estimated time of arrival, but cannot
be expected to provide an exact time, or to provide the lane of the
vehicles. Instead an alternate solution to determine the lane must
be found. Solutions using cameras mounted on the vehicle [48,49]
and using GPS data over time to probabilistically estimate the lane
[50] have been proposed. Perhaps the lane can be determined by
utilizing lane change sensors on the vehicle. Suppose a vehicle en-
ters the roadway into the right lane. By counting subsequent lane
changes, the vehicle can determine its current lane position.
Certain papers in the literature suggest a dynamic order to the
traffic signal cycle. In these approaches the direction that receives
the green light is chosen based on some weighting system. The
issue is that some vehicles may wait at the intersection for an un-
fair amount of time. Many systems offer a max green time for each
77
signal to ensure fairness across each direction; however, it should
also include a maximum amount of time any signal can remain
red. Of course, the maximum red time can be overridden in cer-
tain cases; specifically, if an emergency vehicle needs to pass from
a different direction.
Most systems found in the literature offer strategies to isolated
intersections. In doing so the timing plans are better than those
from traffic signals with fixed-time traffic signal cycles, but they
can be optimized further. If signals are able to coordinate, they
can ensure that a wave a vehicles from one intersection will meet
a green light at another intersection. It is the assumption of the
authors that such a system will provide more optimal traffic strat-
egy than one that works in isolation.
7. Our vision of future developments
Quite recently, the prevalence of smartphones has suggested
supplementing legacy traffic monitoring with incident reports sub-
mitted by the driving public [51]. An implementation of this idea
has lead to the emergence of 511 Traffic that offers an at-a-glance
view of road conditions in a given geographic area [11]. However,
511 Traffic is a centralized system that accumulates and aggregates
traffic-related feeds at Traffic Monitoring Centers and, due to inher-
ent delays, often displays stale traffic information [19].
Thus, there is a critical need for secure and privacy-aware detec-
tion of premonitory signs of imminent congestion and other trends
in traffic flow. Such a system will allow drivers to make more in-
formed decisions about their travel, therefore reducing congestion
and travel times. Ultimately, the system will not only make our
roads safer by reducing the number of traffic accidents, it will also
reduce carbon emissions, lost productivity and wasted fuel. In ad-
dition, the system can be augmented to provide other services to
drivers, including advertising services available at exits and assist-
ing with emergencies and planned evacuations.
While the transportation agencies have devoted substantial ef-
fort to optimizing traffic signals at the corridor level, to the best of
our knowledge, the problem of rescheduling the timing of traffic
lights at the scale of a wider urban area is still very much un-
charted territory. The principle reason behind this state of affairs
is the combinatorial explosion inherent in the process of transiting
from small-scale to large scale, complex problems. This is clearly
the case when the traffic flows from several corridors compete
for unshareable resources – time and road bandwidth. In an ideal
world, the municipal Traffic Management Center (TMC) would have
at its disposal a number of supercomputers that could be used to
compute in near real-time optimal timing plans for all the traffic
signals under its jurisdiction. In reality, no municipality can afford
the huge expenditure involved in purchasing and maintaining a
huge computational resource solely dedicated to optimizing traffic
flow. One alternative would be for the TMC to outsource this huge
computational task to one of the existing cloud service providers.
This approach, however, would be not only be costly but, due to
the overheads involved, would not guarantee traffic signal retiming
in useful time, say, to mitigate the effects of a pop-up congestion
event.
Under present-day practices, the vehicles on our roadways and
city streets are mere spectators that witness traffic-related events
without being able to participate in the mitigation of their ef-
fect. We anticipate that in the near future the on-board resources
present in the vehicles stuck in urban congestion will be harnessed
in order to assist transportation agencies with dissipating conges-
tion through large-scale signal retiming. One approach to achieve
this could be based on vehicular crowdsourcing wherein the ve-
hicles have the potential to cooperate with various transportation
authorities to solve problems that otherwise would either take an
78 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
inordinate amount of time to solve or cannot be solved for lack for
adequate municipal resources.
Instead of retiming signals at the corridor level only, vehicular
crowdsourcing offers the opportunity to optimize traffic flow at
the municipality level by making dynamic use of vehicular network
probe data to retime signals. This will enable traffic signals to be
more responsive to actual conditions, rather than being based on
historic volume counts.
Vehicular crowdsourcing would offer direct benefits to both
drivers and the municipality. By developing timing plans that re-
spond to current traffic conditions, overall traffic flow will improve,
carbon emissions will be reduced and the economic impact of con-
gestion, in terms of wasted fuel and lost productivity hours, will be
lessened. It is expected that drivers will be willing to participate in
vehicular crowdsourcing to develop improved signal timing plans
in return for the direct benefits of time savings and reduced fuel
consumption costs. This will allow the municipality to dynamically
respond to traffic conditions while simultaneously reducing invest-
ment in the computational resources that would be required for
traditional adaptive traffic signal control systems. These ideas are
currently receiving attention by the authors of this survey.
8. Concluding remarks
In this survey, we have identified a taxonomy of congestion
mitigation strategies: public transportation and carpooling, increas-
ing the bandwidth of the current roads, increasing average speed,
and minimizing delay, among others. Minimizing delay by opti-
mally retiming traffic signals has been surveyed in this paper. First
infrastructure-based strategies that utilize ILDs, microwave, cam-
eras, etc. were described. A common characteristic of this first
category of strategies is that vehicles do not play any role in in-
forming the infrastructure about traffic parameters. Next, strate-
gies in which the vehicles provide wirelessly some information to
the system were surveyed. These strategies have the advantage of
providing location information for all vehicles within the transmis-
sion range of a base station. One particular disadvantage that is
commonly sidestepped throughout the literature is that the tradi-
tional systems like ILDs can specify counts per lane, whereas the
location determination by wireless systems cannot pinpoint the
lane of a vehicle. Finally, strategies that utilize the on-board com-
pute power of the vehicles on the road were investigated. These
strategies are not yet as widespread but are gaining momentum
fast. The trend is for vehicles to be outfitted with more sensors
and more compute power to handle the data collected by these
sensors. We expect that researchers will soon see the compute
power available and will find new and exciting ways to utilize
it. Future work for the authors of this survey is to utilize the
compute power of the vehicles on the road to minimize traffic
delay.
References
[1] Pocket guide to transportation, http://www.rita.dot.gov/bts/bts/sites/rita.dot.gov.
bts/files/publications/pocket_guide_to_transportation/2014, September 2014.
[2] NHTSA National Highway Traffic Safety Administration, Traffic safety facts
– preliminary 2009 report, http://www-nrd.nhtsa.dot.gov/Pubs/811255.pdf,
March 2010.
[3] US Department of Transportation, National transportation statistics, 2008.
[4] US Department of Transportation, Research and Innovative Technology As-
sociation, National transportation statistics, http://www.bts.gov/publications/
national_transportation_statistics/, 2011.
[5] Texas Transportation Institute, 2012 urban mobility report, http://mobility.
tamu.edu/ums/, December 2013.
[6] US Federal Highway Administration, Congestion pricing: a primer, http://www.
ops.fhwaa.dot.gov/publications/congestionpricing/congestionpricing.pdf, 2006.
[7] Highway construction costs district 7, http://www.dot.state.fl.us/planning/
policy/costs/costs-D7.pdf, September 2014.
[8] M. Selinger, Hitting all the greens, Roads Bridges (2010) 52–55.
[9] S. Olariu, M.C. Weigle (Eds.), Vehicular Networks: From Theory to Practice, CRC
Press/Taylor & Francis, 2009.
[10] R. Sengupta, S. Rezaei, S.E. Shlavoder, D. Cody, S. Dickey, H. Krishnan, Coopera-
tive collision warning systems: concept definition and experimental implemen-
tation, California PATH Technical Report UCB-ITS-PRR-2006-6, May 2006.
[11] 511 Traffic, Latest traffic news, http://traffic.511.org/LatestNews.
[12] Waze, https://www.waze.com, October 2014.
[13] J. de Dios Ortúzar, L.G. Willumsen, Modelling Transport, Wiley, New York, 2002.
[14] R. Roess, E. Prassas, W. McShane, Traffic Engineering, 4th ed., Pearson, Boston,
2011.
[15] I. Sreedevi, J. Black, Loop detectors, California Center for Innovative Transporta-
tion, February 2001, http://www.calccit.org/itsdecision/serv_and_tech/Traffic_
Surveillance/road-based/in-road/loop_report.html.
[16] M. Fontaine, Traffic monitoring, in: Vehicular Networks: From Theory to Prac-
tice, 2009, pp. 1.1–1.28, x–14.
[17] Z.-X. Li, X.-M. Yang, Z. Li, Application of cement-based piezoelectric sensors for
monitoring traffic flows, J. Transp. Eng. 132 (7) (2006) 565–573.
[18] G. Yan, S. Olariu, D.C. Popescu, NOTICE: an architecture for the notification of
traffic incidents, IEEE Intell. Transp. Syst. Mag. 4 (4) (2012) 6–16.
[19] NHTSA National Highway Traffic Safety Administration, An analysis of re-
cent improvements to vehicle safety, http://www-nrd.nhtsa.dot.gov/Pubs/
811572.pdf, June 2012.
[20] A. Bayen, J. Butler, A. Patire, Mobile millenium final report, Technical Report
UCB-ITS-CWP-2011-6, California Center for Innovative Transportation, Septem-
ber 2011.
[21] A. Hofleitner, R. Herring, P. Abbeel, A. Bayen, Learning the dynamics of arte-
rial traffic from probe data using dynamic Bayesian network, IEEE Trans. Intell.
Transp. Syst. 13 (4) (2012) 1679–1693.
[22] S. Sunkari, The benefits of retiming traffic signals, ITE J. (2004) 26–29.
[23] US Department of Transportation, Federal Highway Administration, Cor-
ridor simulator (CORSIM/TSIS), http://ops.fhwa.dot.gov/trafficanalysistools/
corsim.htm, 2006.
[24] A. Stevanovic, Adaptive traffic control systems: domestic and foreign state of
practice, Technical Report Synthesis 403, National Cooperative Highway Re-
search Program, 2010.
[25] J.B. Kenney, Dedicated Short-Range Communications (DSRC) standards in the
United States, Proc. IEEE 99 (7) (2011) 1162–1182.
[26] US Federal, Communications Commission (FCC), Standard specification for
telecommunications and information exchange between roadside and vehi-
cle systems – 5 GHz band Dedicated Short Range Communications, (DSRC)
medium access control (MAC) and physical layer (PHY) specifications.
[27] US Department of Transportation and IEEE, IEEE 1609 – family of stan-
dards for wireless access in vehicular environments (WAVE). Intelligent
transportation systems standards fact sheet, http://www.standards.its.dot.gov/
fact_sheet.asp?f=80, January 2006.
[28] IEEE, Status of project IEEE 802.11p, http://grouper.ieee.org/groups/802/11/
Reports/tgp_update.htm, November 2006.
[29] P. Koonce, L. Rodegerdts, K. Lee, S. Quayle, Traffic signal timing handbook, Final
Report, FHWA Contract No. DTFH61-98-C-00075, Kittelson and Associates Inc,
Portland, OR, 2008.
[30] K. Kwong, R. Kavaler, R. Rojagopal, P. Varaiya, A Practical Scheme for Arterial
Travel Time Estimation Based on Vehicle Re-identification Using Wireless Sen-
sors, Transportation Research Board, Washington, DC, 2009.
[31] D. Prasad, Adaptive traffic signal control system with cloud computing based
online learning, in: Proc. 8th International Conference on Communications and
Signal Processing, ICICS-2011, 2011, pp. 1–5.
[32] D. Kari, W. Guoyuan, M. Barth, Eco-friendly freight signal priority using con-
nected vehicle technology: a multi-agent systems approach, in: Proc. IEEE In-
telligent Vehicles Symposium, 2014, pp. 1187–1192.
[33] P. Bhuvaneswari, G. Raj, R. Balaji, S. Kanagasabai, Adaptive traffic signal flow
control using wireless sensor networks, in: Proc. Fourth International Confer-
ence on Computational Intelligence and Communication Networks, CICN-2012,
2012, pp. 85–89.
[34] C. Wenjie, C. Lifeng, C. Zhanglong, T. Shiliang, A realtime dynamic traffic control
system based on wireless sensor network, in: Proc. International Conference
Workshops on Parallel Processing, ICPP’2005, 2005, pp. 258–264.
[35] S. Kwatirayo, J. Almhana, Z. Liu, Optimizing intersection traffic flow using
Vanet, Sensor, Mesh and Ad Hoc Communications and Networks (SECON), in:
Proc. 10-th IEEE Conference on Communications and Networks, SECON’2013,
2013, pp. 260–262.
[36] N. Goodall, B. Smith, B. Park, Traffic signal control with connected vehicles,
Transp. Res. Rec., J. Transp. Res. Board 2381 (2013) 65–72.
[37] M. Khamis, W. Gomaa, Adaptive multi-objective reinforcement learning with
hybrid exploration for traffic signal control based on cooperative multi-agent
framework, Eng. Appl. Artif. Intell. 29 (2014) 134–151.
 R. Florin, S. Olariu / Vehicular Communications 2 (2015) 70–79
[38] D. McKenney, T. White, Distributed and adaptive traffic signal control within a
realistic traffic simulation, Eng. Appl. Artif. Intell. 26 (2013) 574–583.
[39] S. Cheng, M. Epelman, R. Smith, CoSIGN: a parallel algorithm for coordinated
traffic signal control, IEEE Trans. Intell. Transp. Syst. 7 (4) (2006) 551–564.
[40] M. Abbas, M. Karsiti, M. Napiah, B. Samir, M. Al-Jemeli, High accuracy traffic
light controller for increasing the given green time utilization, Comput. Electr.
Eng. 41 (2015) 40–51.
[41] Connected vehicle research, http://www.its.dot.gov/connected_vehicle/
connected_vehicle.htm, October 2014, pp. 35–24.
[42] N. Maslekar, J. Mouzna, M. Boussedjra, H. Labiod, Cats: an adaptive traffic sig-
nal system based on car-to-car communication, J. Netw. Comput. Appl. 36 (5)
(2013) 1308–1315.
[43] B. Xiang-yu, Y. Xin-min, J. Hai, L. Jun, A novel traffic information system for
VANET based on location service, in: Proc. 16th IEEE International Conference
on Networks, ICON-2008, 2008, pp. 1–6.
[44] S. Guler, M. Menendez, L. Meier, Using connected vehicle technology to im-
prove the efficiency of intersections, Transp. Res., Part C, Emerg. Technol. 46
(2014) 121–131.
79
[45] J. Lee, B. Park, I. Yun, Cumulative travel-time responsive real-time intersec-
tion control algorithm in the connected vehicle environment, J. Transp. Eng.
139 (10) (2013) 1020–1029.
[46] G. Comert, M. Cetin, Queue length estimation from probe vehicle location and
the impacts of sample size, Eur. J. Oper. Res. 197 (1) (2009) 196–202.
[47] National coordination office for space based positioning, navigation, and tim-
ing, http://www.gps.gov/systems/gps/performance/accuracy, November 2006.
[48] W. Yifei, D. Naim, A. Alin, A novel system for robust lane detection and track-
ing, Signal Process. 92 (2) (2012) 319–334.
[49] K. Chang-Hoon, C. Dong-Chan, R. Moon-Soo, K. Whoi-Yul, Lane detection sys-
tem with around view monitoring for intelligent vehicle, in: Proc. International
SoC Design Conference, ISOCC’2013, 2013, pp. 215–218.
[50] D. Thanh-Son, K. Leung, C. Clark, J. Huissoon, Markov-based lane positioning
using intervehicle communication, IEEE Trans. Intell. Transp. Syst. 8 (4) (2007)
641–650.
[51] M. Abdullah, A. Negara, M. Sayeed, D. Choi, K. Muthu, Classification algo-
rithms in human activity recognition using smartphones, Int. J. Comput. Inf.
Eng. (2012) 415–432.