International Journal of Computer Applications Technology and Research

Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

Techniques for Smart Traffic Control: An In-depth

Review
Roxanne Hawi George Okeyo Michael Kimwele
School of Computing and School of Computing and School of Computing and
Information Technology, Information Technology, Information Technology,
Jomo Kenyatta University of Jomo Kenyatta University of Jomo Kenyatta University of
Agriculture and Technology Agriculture and Technology Agriculture and Technology
Nairobi, Kenya Nairobi, Kenya Nairobi, Kenya

Abstract: Inadequate space and funds for the construction of new roads and the steady increase in number of vehicles has prompted
scholars to investigate other solutions to traffic congestion. One area gaining interest is the use of smart traffic control systems (STCS)
to make traffic routing decisions. These systems use real time data and try to mimic human reasoning thus prove promising in vehicle
traffic control and management. This paper is a review on the motivations behind the emergence of STCS and the different types of
these systems in use today for road traffic management. They include – fuzzy expert systems (FES), artificial neural networks (ANN)
and wireless sensor networks (WSN). We give an in depth study on the design, benefits and limitations of each technique. The paper
cites and analyses a number of successfully tested and implemented STCS. From these reviews we are able to derive comparisons of
the STCS discussed in this paper. For instance, for a learning or adaptive system, ANN is the best approach; for a system that just
routes traffic based on real time data and does not need to derive any data patterns afterwards, then FES is the best approach; for a
cheaper alternative to the FES, then WSN is the least costly approach. All prove effective in traffic control and management with
respect to the context in which each of them is used.

Keywords: smart cities, intelligent traffic systems, artificial intelligent system, WSN, FES, ANN, traffic lights, road traffic
controls and artificial intelligence to make traffic routing
1. INTRODUCTION decisions; a task typically done by traffic officers e.g.
The steady increase in the number of vehicles on the road has
policemen or traffic marshals. Other application areas include:
increased traffic congestion in most urban cities of the world.
surveillance, management of freeway and arterial networks,
One approach most countries are taking to address this issue is
intersection traffic light control, congestion and incident
the expansion of roadways. However, this approach still
management [3].
comes with its share of challenges. Demolition of older roads
can be quite costly. Most urban cities lack the free space 1.1 Background and motivation
required for such a venture. Even with the improvements in Other than inadequate infrastructure developments, other
road infrastructure, it is evident that the rate at which travelers factors that have prompted scholars to further investigate use
buy vehicles has surpassed that of new infrastructure of smart traffic control systems include:
development. Also due to expansions, roads are able to serve
more vehicles; consequently utilizing the additional capacity. 1.1.1 Weakness in current traffic control systems:
This is consistent with the ‘fundamental law of highway
congestion’ suggested by Downs [1] who avers that increasing Almost all urban cities in the world use traffic lights to control
road supply invariably increases vehicle traffic. the traffic on the roads. The lights switch from red, which
means stop, to green, which means move. Over time there has
With inadequate space and funds for the construction of new been developments of different types of traffic light control
roads, and the growing imbalance between traffic demand and systems, the most commonly used being static traffic lights
transportation resources; it is increasingly obvious that and vehicle actuated lights.
countries must move beyond the traditional model of just
building roads to solve traffic problems [2]. This is Static traffic lights’ timing and switching patterns are
demonstrated in a survey done by CBT in Britain. The report predetermined despite prevailing traffic conditions for the
supports that expansion and building of new roads will do different lanes. They do not operate with real time data.
very little to help solve the congestion issue. CBT survey Consequently this means they do not take into account the
found that nearby local roads suffered up to 137% more traffic non-uniform and ever changing nature of traffic conditions. It
after the bypasses opened, and reductions on the roads does not matter whether at a particular period of time route
intended to be relieved were less than expected [18]. one has more cars than route two; the green light allocation
time and pattern still remains the same for all routes. The lack
Therefore, managing of traffic flow needs to be a combination of intelligent strategies in these devices does very little in
of physical infrastructure, new ways of thinking and new improving the road network performance and traffic
technologies. Smarter transport transcends infrastructure [2]. congestion levels.
In light of this, smart traffic control systems have gained a lot
of interest. This was demonstrated in Kenya when the country recently
experienced a setback when the Nairobi County Government
pulled out policemen from the roads to test the newly
These smart traffic control systems use advanced technologies automated traffic lights that had an additional counter feature
such as image processing, computer vision, intelligent – that counts down from one light to the other. This meant

www.ijcat.com 566
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

that the motorists could then know how long to wait before 1.1.3 Sensor Networks:
moving [4]. However, this operation was not successful; there Advancement in very large scale integration (VLSI) and
was a traffic gridlock in most parts of the city with some semiconductor technologies have enabled the development of
motorists spending close to 4 hours in bumper to bumper smaller, tiny, low power, and inexpensive sensors and
traffic. [4] controllers/microprocessors. Furthermore, developments in
wireless technologies have made it possible for the use of
Vehicle-actuated traffic lights were an attempt to enhance the sensors to collect large amounts of environmental data at
static lights. They combine preset time cycles with proximity minimal costs. These networks comprise of many sensors that
sensors. These sensors can activate a change in the cycle time cooperate to monitor and collect data about traffic conditions
or the lights when cars are present. This is due to the on the roads.
assumption that roads with fewer cars may not need a regular
cycle of green lights. However the downside of these traffic Although sensor network technology for highway and traffic
lights is that they are not adaptive. They depend on having management is a relatively new solution to alleviating
some prior knowledge of traffic flow patterns at the highway congestion; they have potential to be one of the
intersection so that signal cycle times and placement of sustainable solutions to road congestion. These networks have
proximity sensors may be customized for the intersection. gained popularity because they provide a cheaper alternative
This means that the signal time/extension is still a fixed value. to that of expansion of roadways and implementation of A.I
Also proximity sensors will only activate a change in signal systems, especially in emerging economies.
light when cars are present, they do not count cars. [10]
Pascale et al. [8], introduce an intelligent transport system
1.1.2 Advancements in the field of Artificial (ITS) that uses a wireless sensor network (WSN) to monitor
Intelligence (A.I): traffic. Their system comprises of a network of traffic sensors
A.I. is the science and engineering of making intelligent deployed throughout the roads that collect and forward
machines, especially intelligent computer programs. It is measurements to a remote server. The server aggregates and
related to the similar task of using computers to understand processes macro-parameters of the traffic flows arising from
human intelligence [5]. The ability of these systems to heterogeneous monitoring systems, then distributes the data to
emulate human intelligence has therefore led to artificial traffic management centers, road control units (RCU) and
intelligent systems pervading our everyday life. For instance, information providers. The macro-parameters can be used for
A.I. is used in a number of scenarios e.g.: banks – automatic traffic analysis and management [8].
check readers, signature verification systems; digital cameras
and mobile phones – automated face/voice detection and In this paper we present the motivations behind emergence
focusing; web – automatic location recognition from your web and use of smart traffic control systems in road traffic
surfing, automatic fraud detection just to mention a few. management; we further describe the design, benefits and the
limitations of the different types of smart traffic control
Consequently scholars; especially in developed countries are system (STCS) in use today. The outline of the paper is as
now interested in investigating the application of A.I in follows. In section I we have described what STCS are and
vehicle traffic management systems. The basic idea behind the factors that influenced their application in road traffic
this being, if these systems can mimic human reasoning, then management. Section II describes the methods used as well as
they can effectively be used to control traffic in place of parameters considered when selecting and analyzing the
traffic officers. related works reviewed in this paper. In section III we present
the different types of STCS in use, their design and benefits.
Developing countries are also catching up with this approach. In section IV, the limitations of each technique (STCS) are
IBM opened a commercial technology research facility in analyzed. Section V we briefly discuss our insights on
Nairobi, the first of its kind in Africa. The lab’s research reviewed work and give some suggestions and
agenda includes the development of cognitive computing recommendation. Finally our concluding remarks are given in
technologies which integrate learning and reasoning section VI.
capabilities enabling experts to make better decisions in
search for solutions to Africa’s most pressing challenges [6]. 2. METHODS
In an effort to tackle the traffic congestion problem, IBM The works reviewed in this paper were selected and analysed
partnered with a Kenyan internet service provider, Access based on the following criteria:
Kenya, to develop a pilot solution to enable Nairobi a) Approaches used to make traffic routing and light
commuters to use their mobile phones to get advice on driving signal allocation decisions. For instance adaptive
routes through the city depending on estimates of traffic (learning) versus non-adaptive strategies; offline
congestion. The project dubbed Twende-Twende is a mobile versus real time strategies; and hybrid strategies.
application that uses specialized algorithms to do image b) Number and types of parameters/variables (input and
processing and interpret visual data received from closed- output) used. We review systems that use single
circuit television (CCTV) cameras positioned around Nairobi. variables (e.g. traffic quantity) and ones that use
Motorists are then able to get information on a) what areas to several variables (e.g. traffic quantity, waiting time,
avoid because of congestion by suggesting alternative routes past and present traffic data knowledge) to make
and b) updates on road conditions to allow them get from traffic routing decisions.
point A to point B safely. This information is retrieved via an c) Traffic data collection methods used (such as sensor
SMS-based query for basic phones and on smart phones the types) and communication methods applied (such as
service is accessed via an application through which users can multi-hop or single-hop) to transmit collected data.
view a map of the city showing route options and potential d) STCS that control traffic at an isolated junction or
traffic hotspots. The project’s main focus is data collection, multiple intersection junction or both.
analysis and information dissemination, it does not actively
control traffic on the roads [7].

www.ijcat.com 567
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

e) Ways to improve overall performance of already Fahmy [10] later presents another system FLATSC that uses
existing intelligent/smart systems (STCS) in use. fuzzy logic controller to manage traffic at a four intersection
roundabout. However, unlike Khiang et al system, his system
employs another input variable, waiting time, to determine
3. SMART TRAFFIC CONTROL green light allocation and extension. FLATSC therefore uses
TECHNIQUES USED BASED ON THEIR traffic quantity and waiting time to determine the priority
SYSTEM ARCHITECTURE degree (output variable) for each lane on the roundabout. The
Smart traffic light controls are dynamic. This means that they output value is the green light time/extension for each lane.
use real time data to make priority based decisions. They use The lane with the highest output value gets allocated the green
advanced communication systems based on sensors and/or light. When cars in one lane move the other lanes stop. The
RFID tags to collect data and provide the system with green light extension was not a fixed value; it was dependent
information on the current situation on the roads (such as on real time data collected from the sensors. The value
number of vehicles on individual roads or how long vehicles changed as the traffic variables fluctuated from cycle to cycle
have been waiting for green light). The smart system then and/or lane to lane. This ensured that traffic was controlled
processes this information and makes decisions; that is, it based on prevailing traffic conditions on the roads.
automatically determines the duration of each traffic light
signal based on prevailing traffic situation on the roads. From a comparison with the fixed controller and vehicle
Commonly used systems include fuzzy expert systems (FES), actuated systems, FLATSC proved to be more effective in
artificial neural networks (ANN) and wireless sensor managing the changing traffic patterns. In addition FLATSC
networks (WSN). attempted to resolve starvation. Starvation describes a
situation whereby some lanes end up always getting last
3.1 Fuzzy expert systems (FES) priority because they usually get the least traffic consequently
ending up always queuing for the longest time during a cycle.
FES is a suitable approach to dynamic traffic signal control
FLATSC addressed this issue by incorporating waiting time
because of the nature of uncertainties on road traffic where the
as a factor in determining green light allocation. For instance
traffic distributions fluctuate non-uniformly. Fuzzy logic is a
a lane that has low traffic but very high waiting time still has a
field started by Zadeh [9]. It is a superset of Boolean logic
chance of getting a high priority degree just as a lane with
that has been extended to handle partial truths between
high traffic but low waiting time; depending on the fuzzy
completely false (0) and completely true (1). This is in an
inference.
attempt to mimic or reflect how humans think, to model our
sense of words when describing certain phenomena as well as
our common sense in decision making. The sensors collect 3.2 Artificial neural networks (ANN)
data from the environment which in turn is fed into the fuzzy The major difference between ANN (learning systems) and
logic controller (FLC) for processing. The inference process FES is that; while an FES uses present knowledge to make
in a FLC is similar to the way traffic officers handle the traffic decisions, in a learning system, the decisions are computed
flow at a typical roundabout [10; 11]. The FLC’s objective is using the accumulated experience or knowledge from
to control operations in systems by making decisions that successfully solved examples. Since ANNs try to mimic the
utilize rules expressed with the uncertainty of human terms human brain they possess an adaptive feature that allows each
such as cool (slightly cold) or warm (slightly hot). Therefore node within the network to modify its state in response to past
FLCs are a suitable approach to traffic signal control because and present knowledge. [12; 3]
it assigns green or red light signal based on urgency or as
traffic fluctuates; and selects the best decision that will Patel et al. present an ANN system used to control traffic. The
minimize congestion at a particular interval. For instance, a input given to the ANN models are the list of data collected
lane could also have low or medium traffic as opposed to just by the sensors which are placed around the traffic lights. The
no traffic (0) or high traffic (1). sensors give the traffic light ANN model all the data which
are related to the past and present traffic parameters. The
Khiang et al [11] present a fuzzy logic traffic light controller. model then processes this input and selects the most suitable
Their system uses two input variables; quantity of traffic on output that suits current traffic situation. These results are then
the arrival side (arrival) and quantity of traffic on the queuing used by the traffic lights to set the timing for the red and green
side (queue) collected from the sensors on the lanes. Their lights. In their ANN approach they evaluate that for the ANN
system controls traffic on multiple lanes simultaneously i.e. to produce accurate decisions it required 83 neural nodes,
North and south lanes move together while east and west their system produced 73% accuracy level for the derived
lanes move together. When North and South have green light, solutions.
East and West stop (queue). The fuzzy controller observes
the density of north and south as one side and east and west as Michael et al. [28] also present a neural networks based traffic
another side. Their system then determines green light light controller called Environment Observation Method
allocation and extension based on the side that has the highest based on Artificial Neural Networks Controller (EOM-ANN)
traffic quantity. From their experiments, they are able to to control urban traffic. Their approach is different from
demonstrate that their fuzzy logic traffic light controller [3;13] because they also incorporate mathematical strategies
performs better than the fixed-time (static) controller. (EOM) to make signal allocation decisions. EOM is a
mathematical methodology for obtaining timing plans for
From a comparison made between the performance of the isolated intersections. It achieves this by calculating the
fuzzy logic controller and that of a fixed-time (static) minimal green time for each phase then to prevent congestion
controller; [11] observed from the results that the fuzzy logic an additional green time is allocated to each lane that still has
controller had a lower average waiting time – a difference of 6 cars even after getting green light. However the downside of
minutes. EOM is that it sets traffic light timing based on averages of
the basic parameters. Due to the fact that these figures are
constants, the EOM doesn’t incorporate the real time nature of

www.ijcat.com 568
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

traffic which means that the traffic parameter values (data) the green and red periods. TCAMI main objective is
keep changing every time, this is further backed by [13] that coordination and setting of traffic parameters and conditions
traditional mathematical methods have limitations when they on the multiple intersections in general and on the successive
are applied in traffic control. The EOM-ANN is an attempt to intersections in specific, with the objective of minimizing
resolve this issue of real time data, [28] propose use of ANN delays, caused by stopping, waiting and then speeding up
to obtain this traffic data patterns. That way, the green light during road trips (also known as green wave – where drivers
timing and allocation is based on actual/prevailing traffic need not stop on multiple intersections thus achieving, if
conditions rather than analytical calculations. [28] implemented correctly, an open route for the vehicles). When
TCAMI is executed on each intersection it will generate
EOM-ANN uses the feed-forward method with 8 neural nodes traffic information, which in turn represents an input to the
in total for input, hidden and output layers. It is further subsequent intersection, and so on. As such, the traffic flow
divided into two modules; reviser and the neural. The former will be controlled in a flexible manner. [15]
defines correct traffic light timing and the latter provides the
most appropriate value for the current traffic behaviour. The To show efficiency of proposed scheme, [15] compared the
inputs of the ANN are the number of light, medium and heavy system to the traditional traffic light control approach which
vehicles. [28] uses static plans i.e. fixed time control. The results indicate
that the proposed system had a better performance rate in
From a comparison between static time controller and EOM, managing traffic; its AWT was much lower at 2.98 minutes
EOM-ANN reported better traffic flow and congestion compared to 7.87 minutes of the fixed time controller. A low
management. The average traffic flow of the individual AWT means that the flow of traffic is increased hence lower
controllers was as follows: static controller - 82.55, EOM - AQT of 9 cars as opposed to 36 cars per queue in the fixed
68.70 and EOM-ANN registered an average of 53.75. [28] time controller. The dynamic approach was able to handle
queues quickly with less cars accumulating on a lane during
3.3 Wireless Sensor Networks (WSN) the observed time. [15]
In the event WSN is used to not only collect traffic data but
also actively control road traffic, additional functionalities are Bhuvaneswari et al. [16] further support [15] by developing a
incorporated into the network’s controller. An algorithm is traffic congestion control system ATSWSN that is adaptive in
embedded to control the traffic lights – it generates routing nature. However it differentiates from Yousef et al [15]. in
decisions based on sensor data aggregated. Unlike some A.I that the time slots allocated for each route is not only based on
systems, WSN does not require vehicles to have additional traffic density, but also on emergency conditions and speed
systems such as RFID tags to control and manage traffic. As a patterns of incoming traffic. Their system collects real time
result WSN are cost inexpensive and make it a more practical data using IR sensors and the microcontroller’s scheduled
than ANN and FES approaches especially in emerging algorithm processes this data and determines which direction
economies. gets green light priority. The duration of the green light is
dynamically calculated based on the weighted speeds of all
Yousef et al. [15] present an adaptive traffic light control the vehicles in the waiting queue factoring in any
system for single and multiple intersections using WSN. Their emergencies.
system uses the WSN to route traffic based on traffic density
and waiting times. It is composed of: sensors that detect the When compared to the conventional fixed time approach,
presence of vehicles and have a memory that stores their ATSWSN registers a higher traffic flow rate and as a result
waiting times on each road. It also has an intelligent traffic lower average waiting time. This is because the clearance time
controller that processes the sensor data then employs two is inversely proportional to delay factor, slow moving vehicles
algorithms traffic system communication algorithm (TSCA), are allotted more clearance time than fast moving vehicles.
traffic signal time manipulation algorithm (TSTMA) to route Further, by scaling the delay factor by the emergency factor,
traffic based on the traffic variations of all lanes of the the clear route parameter is kept high when an emergency
intersections at a particular time and traffic control algorithm vehicle enters the lane. Thus, both the direction and time to be
on multiple intersections (TCAMI). cleared are chosen optimally. They observed that in the fixed
time approach, the waiting time increased as the number of
TSCA main objective is to enable exchange of information vehicles increased irrespective of their speeds and speed
between the sensors’ base station (BS) and the controller factors [16].
using a direct routing scheme approach. This means all
sensors are within range of the BS and directly communicate
3.4 Hybrids
with it. On the other hand, TSTMA main responsibility is to To overcome the limitations of the individual
set the traffic signal duration in an efficient and dynamic implementations of ANN and FES approaches, such as; lack
manner such that traffic flow is maximized while at the same of learning ability of fuzzy systems and lack of inference
time ensuring minimal average queue length (AQL) and process of ANNs (mentioned in section IV of this paper);
average waiting time (AWT). TSTMA makes use of the Patel et al. [3] developed a hybrid intelligent decision making
traffic information gathered at the traffic BS from the sensors system (IDUTC) for urban traffic control applications.
to calculate in intelligent manner, the expected queue length,
for the next traffic cycle, and then schedule efficient time The sensors (closed loop detectors) placed on the roads collect
setting for the various traffic signals. TSTMA achieves this traffic data; volume (traffic quantity) and occupancy (wait
objective through three main techniques: (a) Dynamic time) of each lane. However IDUTC only computes its
selection and ordering of the traffic phases based on the decisions based on one parameter, the traffic quantity. Using
number of lanes allowed in the intersection; (b) Dynamic the volume data, five traffic parameters are computed that
adaptation to the changes in the arrival and departure rates and describe in more detail the traffic flow of the intersection
thus dynamic decisions about queues’ lengths and their lanes. They are: 1) highest saturation, 2) the cross saturation,
importance; (c) Dynamic control of the traffic cycle timing of 3) the saturation difference of the traffic, 4) the volume

www.ijcat.com 569
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

difference and 5) the required green time extension. The four 4.2 ANN
parameters from the previous time frame which were stored in While [28;13;3] demonstrated that using ANN to manage
a memory device (the saturation difference was not included) traffic was effective; ANNs training process in most cases is a
and the five parameters for the current time frame become time-consuming task requiring the application of input
inputs to the ANN. This is fed into the IDUTC as crisp values. training patterns in an iterative manner. This was
The ANN then processes all the system data, past (collected experimentally proven by Barbosa and Pinto [17]; they
and stored from previous cycles) and present. The ANN successfully managed to show that by increasing the amount
output provides the input for the FES. The FES performs of data the performance of the ANN system improved. The
inference and assigns a fuzzy label/values to the input error margin was lower when training was extended to 15
received. Then FES fires the rules based on these fuzzy minutes of collecting data as opposed to 5 minutes.
values. The defuzzification unit converts the computed
decisions into crisp values that are used to determine green Also another drawback of ANNs is the lack of rules or guides
light allocation and extension for each traffic light. The cycle to support the decisions to be made; resulting in development
goes on repeating and tries to change the traffic light timings of solutions that are mostly specific or case base problems.
condition so as to ensure that IDUTC self-adjusts according to This means no explanation or guarantee that the solution
the situation. IDUTC integrates the learning abilities of an chosen is the optimum one [17]. This is proven by Patel et al.
ANN and the knowledge-based decision-making ability of the where they demonstrate that the ANN had a correct decision
FES. The back propagation-based ANN allowed the system to rate of 73% as opposed to the IDUTC (hybrid system) and
learn and adapt to the dynamically changing environment and FES which both had a 95% rating. They further realized in
the FES was employed for decision making using the IF- their experiments that the ANN approach had difficulty in
THEN rules. [3] generalizing on the various numbers and the combinations of
traffic parameters and required cycle-time adjustments
A summary of the simulations for the IDUTC system, the (desired outputs) .[3]
ANN, and the FES approaches indicate that: the IDUTC
system provided 95% correct decision rate and an average
waiting time of 2.186 minutes. It relieved intersection 4.3 WSN
congestion better than the ANN approach which provided Owing to the fact that sensors are micro-electric devices, they
73% correct decision rate and an average waiting time of operate on a limited energy budget. For this reason WSNs are
2.958 minutes. While the FES approach correct decision rate faced with the problem of having to regulate their energy
was equal to that of IDUTC, it was observed that the consumption. This can be a daunting task especially when
computed decision did not lead to a better reduction in the dealing with large complex traffic networks with multiple
wait times. The FES had an average wait time of 2.975 intersections. The interdependency of each intersection on its
neighbors makes it extremely important to ensure that the
minutes which is lower than the other two approaches. [3]
different sensors in each intersection are in constant
4. STCS DESIGN CONSTRAINTS communication to ensure real time data processing. This
consequently takes a toll on the energy consumption rate of
Singh. et al [14] aver that most automated traffic control
the sensors especially if the distance between them is wide.
systems not excluding STCS; have the following general
The wider the distance between nodes and BS the higher the
limitations:
attenuation rate; consequently leading to increased power
a) If the position of vehicle does not come in
needed during data transmission/communication. Failure of a
alignment of infrared rays then IR sensor would not
node could lead to massive traffic congestion; also, a
give response.
downside of this is that motorists will avoid this lane and
b) If the vehicle is under faulty condition in the range
move to the lanes with lower AWT. This could lead to an
of IR rays then the response given by IR sensor
increase in congestion levels on the lanes that usually get first
would not be accurate.
priority.
c) If a single camera is used as a vision sensor for
acquiring the image of traffic then it is difficult to
detect the space between two vehicles means 4.4 Hybrid
projection of camera would be crucial factor for The initial implementation cost can be quite costly,
measuring the traffic density. considering that A.I is still considerably a new research area
in traffic control. Also the fact that a hybrid is a combination
Below are the specific limitations with respect to the design of of two or more STCS, the development time can take a long
each of the mentioned STCS: time in an attempt to ensure successful integration of the
different systems.
4.1 FES
Although use of fuzzy expert systems in traffic light control 5. DISCUSION
systems enhances the efficiency of traffic movement in roads, From the review of the different STCS used, some open issues
the downside of such systems is that, they do not have the arise for each technique. Firstly we discuss the general open
ability to learn, they are not adaptive. Meaning, they do not issues that need to be addressed in all the STCS techniques
incorporate past knowledge or experiences to make current reviewed:
decisions; rather they only make decisions based on the
current knowledge they have of the situation. For this reason, We start with starvation. This is one of the general open
it becomes quite challenging to modify the system’s issues that come up in any STCS in an attempt to ensure that
parameters whenever necessary. For instance; changing the no lane at any particular time is neglected for as long as there
green light duration time for a single lane only; based on the are cars queuing. How to solve starvation is still an open issue
discovery that it usually gets more cars on average would be particularly because traffic quantity (TQ) and waiting time
very tasking. (WT) are usually isolated from each other. That is, most
STCS systems use only one of them to determine traffic light

www.ijcat.com 570
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

signal allocation. Also another reason is that naturally, possibly the inference process (which if changed affects
overtime as traffic quantity increases (assuming there is a everything else).
continuous flow of traffic in and out of all lanes at a junction)
the average waiting time of vehicles also increases. Adaptability. Owing to the fact that FES are not learning
systems, it becomes very difficult to ensure that they are
Some scholars like [10] and [15] try to resolve starvation by adaptable. They work for specific data values that have
presenting a system that uses both parameters (TQ and WT) to already been assigned probability factors [24]. This means, for
determine green light allocation and extension. However, out of the box scenario, an FES cannot provide decision as the
overtime our assumption is that the lanes that usually have rules for the exception are not coded [24]. More importantly,
low traffic volumes will end up experiencing some level of because the membership functions and the rules are dependent
starvation especially during peak hours. on each other, a change in one will necessitate a change in the
other [23]. A change in any of the parameters (input/output
The other general open issue is controlling traffic signals in a variables) will require reorganization of part or entire
large scale traffic network (i.e. simultaneously controlling knowledge base depending on the number of parameters being
traffic for multiple intersections). Srinivasan et al. [19] state changed. This can be a daunting task especially if you have
that it is crucial for traffic signal control systems to have the many membership functions. For instance an FES with 3
capability to examine both the microscopic level of the membership functions will yield 9 rules while one with 6
situation (the traffic state of each intersection) as well as the membership functions will yield 36 rules four times more than
macroscopic level of the situation (the overall traffic state of the latter. Consequently a system with many rules can be quite
the traffic network) complex to make adaptable or flexible to changes, yet [22]
and [23] suggest that the more rules a system has the more
Yousef et al [15] implement an adaptive traffic light control accurate the decisions will be because more case scenarios
system that controls traffic across multiple intersections using will be factored in the rules.
WSN. However, for a large-scale traffic management system,
it may be quite difficult to ensure and or even to determine 5.2 ANN
whether the traffic network is flowing smoothly and assess its Optimality of solution. ANN derives solutions based on past
current state. Also due to the non-uniform nature of traffic in and present knowledge that is fed into it. Therefore, in the
traffic networks, predicting the effects of modifying any of the event the data training of the system is improper, it could lead
traffic control parameters is a difficult task [19]. For instance; to incorrect decisions. Further, because ANN does not give
consider the following network arrangement. There are three explanations for its decisions [3; 17; 24]; determining
networked intersections that are coordinated linearly; A, B beforehand whether the new knowledge extracted is the best
and C respectively. Such that, whenever lane 1 in intersection solution can be quite difficult. ANN follows the theory of
A is green, lane 1 in B and C also get green light. In the event ‘garbage in garbage out’ [24]. This remains an open issue
that the STCS needs to control traffic based on the following because an important aspect in intelligent system design is
scenario: the traffic on lane 1 in A is high but that of B decision explanation, which involves supplying a coherent
requires green light for lane 2 instead with lane 1 having no explanation of its decisions [25]. This is required for
cars (least priority) and intersection C requires green light on acceptability of the solution and correctness of the reasoning
lane 3 and its lane 1 requires second priority (next in line); [3].
then it becomes very difficult to determine how to
synchronize the lights across the intersections to ensure Offline vs online methods of learning. Backpropagation being
minimal traffic congestion while at the same time ensuring the most popular approach used to train ANN [2; 26]; an open
that all lanes get green light priority when needed. issue that usually comes up with this approach is, when to use
offline methods (batch leaning) or online methods (single step
The specific open issues pertaining to the individual STCS learning) during the learning process. Offline/batch learning
techniques are described below: requires that all training data is available because learning is
only performed after a full set of training data is presented.
5.1 FES On-line/single-step learning is suitable for when training data
Defining rules. Where do these rules come from? [23]. There is produced by an on-going process meaning training data is
are no specific universal criteria used to derive rules and not available prior to the process. This means that learning for
membership functions (degree to which a variable is a single step is performed after presentation of a single
associated with a term) of this system. Each developer has training pattern. However both methods have their share of
their own criteria. This is because the linguistic variables (set limitations; batch learning can take a bit longer because
of terms expressed in natural language that can represent weights are only updated after a full set of training data is
possible values that a system variable can take [22]) and their presented. Single-step learning can reduce the adaptability
membership functions vary in type and number from one performance of the system because weights are updated based
system to the next. While some applications like fuzzylite [22] on the last training pattern presented; this means the solution
derive the rules for you, it is still quite difficult to determine derived for the next predicted pattern is specific to one single
how the application arrived at those rules. Further, the more previous pattern. On the other hand, the two methods also
variables (input/output) a system has, the more complex it have their strengths when compared to each other; single-step
becomes to define rules. As for their accuracy levels (i.e. how training is faster while batch method yields lower residual
correct the decisions are), this can only be determined during error due to more data information as proved by [17] that
an iterative testing phase; which in our opinion is not ideal increasing the training data of an ANN results in more correct
especially if one is dealing with a large knowledge base of solutions. [27]
rules. In the event the FES fails to meet standards, developer
is left with the difficult task of modularizing the rules and

www.ijcat.com 571
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

Despite the benefits of each method, considering that the

nature of traffic is uncertain and keeps fluctuating, the Perhaps then the most sustainable approach to energy
question of when to use these methods when it comes to smart consumption currently is regenerative power sources for the
transport systems is still debatable. Scholars like [27] suggest nodes such as solar charged batteries. However rechargeable
use of on-line methods is a better approach to transport traffic batteries are more expensive than disposable ones. Therefore
management. However, Pinto et al [17] insist that for an ANN designers are forced to make the difficult decision of a trade-
to provide more accurate solutions the training time has to be off between cost and energy consumption.
extended to incorporate more data and that includes more past
data not just single training pattern. With that in mind, then 5.4 Hybrid
perhaps the best solution is to incorporate both methods in one While this area of research is still fairly new, there still exist
system. Start with single-step learning to fasten the learning few studies done. However one major issue with these
process and then batch learning to reduce error rate and systems is cost. This is in the context of both design and their
implementation particularly the implementation cost
improve accuracy of final result [27].
considering that A.I is still a new open ground research area.
Patel et al [3] come up with a hybrid IDUTC that actually
5.3 WSN strives to reduce the design components of a hybrid system
Techniques for regulating energy consumption in large traffic while still at the same time improving the system’s
WSN networks (multiple intersections). It is evident that an performance. In the IDUTC we see that they successfully
increase in the distance between a node and a base station will manage to reduce the design of the system with regards to the
trigger the need to use more transmission power. Thus, in the individual approaches. Their system has 55 nodes that
design of WSN, short range transmission (multi-hop) should produces an overall correct decision rate of 95% as opposed to
be considered, in order to reduce attenuation as well as ANN individual approach which requires 83 nodes in order to
minimize power consumption during transmission. In multi- produce a correct decision rate of 73% overall. The ANN
hop communication the sensor nodes serve as relays for other approach required more neural nodes than the ANN in
sensor nodes, and must co-operate with each other to find the IDUTC, which led to slower training and a higher
most efficient route to transmit sensor data towards the BS. implementation cost [3].
However, in an attempt to make the network more energy
efficient by splitting up large distances between nodes into 6. CONCLUSION
several shorter distances, WSN designers are faced by yet A review on the use of technology to control and manage
another challenge, routing [20]. traffic was presented in this paper. It is observed that the
implementation of smart technology in transportation systems
This routing problem, that is, the task of finding a multi-hop has a substantial impact on traffic levels. While the static
path from a sensor node to the base station, is one of the most systems provide a simpler method of automatically
important challenges and has received immense attention controlling traffic; they do not have the flexibility needed on
from the research community [20]. This challenge is most urban junctions which serve non uniform traffic from the
especially experienced in networks that use switching various approaches/roads. Advancement in AI has further led
techniques. To conserve power, transceivers are designed to to the development of intelligent traffic control systems. The
have switching states: active, idle and sleep states. Where main objective of these smart systems is to have the traffic
active is when the nodes are receiving and transmitting, idle is lights mimic the human intelligence thus eliminating the need
when the sensor is on but not transmitting or receiving any of having traffic officers control traffic on the roads. These
data and sleep state is when the sensor is off. Designers thus intelligent systems provide a way for the lights to change from
have the task of deciding how and when it is appropriate to red to green based on current traffic conditions. Though these
implement each state in order to conserve energy and still systems provide substantial benefits to management of traffic,
maintain network efficiency FES and ANN are a branch of A.I. that is still an emerging
field in IT; hence the implementation of such systems as stand
In such networks, the sensor nodes are switched off when not alone is still quite costly, especially in the developing
in operation. In [21] it is observed that most transceivers countries. Another STCS alternative to using the A.I systems
operating in idle mode have power consumption almost equal is sensor networks. These networks have gained popularity
to the power consumed in receive mode. Thus, it is better to especially due to the low cost of implementation compared to
completely shut down the transceiver rather than leave it in the A.I based systems. The network is comprised of many
the idle mode when it is not transmitting or receiving. As a sensors that cooperate to monitor and collect data about traffic
consequence, during these down-times, the sensor node conditions on the roads. This information is then forwarded to
cannot receive messages from its neighbors nor can it serve as a controller that processes the data into meaningful
a relay for other sensors [20]. Consequently if node 1 wants to information. Using an algorithm the controller is able to make
send data to node 2, but node 2 is in sleep state this might routing decisions based on current traffic conditions.
cause some communication/network.
Although smart traffic control systems still have some
Scholars have suggested some strategies to resolve this limitations to what they can achieve intelligently, the future
routing problem, they include: 1) wake-up on demand where a still holds a lot of promise for these systems. Researchers
node switches to active state only when needed and 2) especially in the field of A.I are working hard to find ways to
adaptive duty cycling strategy, when not all nodes are allowed overcome these limitations in order to make them completely
to sleep at the same time; instead, a subset of the nodes in a efficient. From this paper it is evident that smart systems are
network remains active to form a network backbone. the way forward for road traffic control.
However, a significant amount of power is consumed when
switching from sleep mode to transmit mode in order to
transmit a packet. [21]

www.ijcat.com 572
 International Journal of Computer Applications Technology and Research
Volume 4– Issue 7, 566 - 573, 2015, ISSN: 2319–8656

7. REFERENCES [15] Yousef, K. M., Al-Karaki, J. M., & Shatnawi, A. M.

(2010). Intelligent Traffic Light Flow Control System Using Wireless
[1] Downs, A. (1962). “The law of peak hour expressway
Sensor Networks. Journal of Information Science and Enginering, 26,
congestion”. Traffic Quarterly 16(3):393–409. Retrieved from
753-768. Retrieved from
http://worldcat.org/issn/00410713
http://www.iis.sinica.edu.tw/page/jise/2010/201005_02.pdf
[2] IBM (2013). “A Vision for a smarter city – How Nairobi
[16] Bhuvaneswari, P.T.V., Arun raj, G.V., Balaji, R.,
can lead way into a prosperous and sustainable future.” Retrieved
Kanagasabai, S. (2012). “Adaptive Traffic Signal Flow Control using
from http://www-05.ibm.com/za/office/pdf/IBM - A Vision for a
Wireless Sensor Networks”. IEEE Computer Society, 85-89. Paper
Smarter City – Nairobi.pdf
presented at 4th International Conference on Computer Intelligence
[3] Patel, M., & Ranganathan, N. (2001). “IDUTC: An
and Communication Networks. Retrieved from
Intelligent Decision-Making System for Urban Traffic-Control
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6375077&url
Applications”. IEEE Transactions on Vehicular Technology, 50(3),
=https%3A%2F%2Ffanyv88.com%3A443%2Fhttp%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Far
816-829. Retrieved from
number%3D6375077
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=933315&url=
[17] Barbosa, M. R., & Pinto, G. C. (n.d). “Exploring the use of
https%3A%2F%2Ffanyv88.com%3A443%2Fhttp%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farn
Neural Networks in urban traffic management”. Business
umber%3D933315
Sustainability I, 287-292. Retrieved from
[4] Ndonga, S. (2014). “Gridlock in Nairobi as Traffic Lights
http://labve.dps.uminho.pt/bs11/CD/PDF/47%20-%20pp%20287-
put to test” Retrieved from
292%20-%20Pinto%20G.C.,%20Barbosa%20M.R..pdf
http://www.capitalfm.co.ke/news/2014/02/gridlock-in-nairobi-as-
[18] CBT. (2012). “Going Backwards the new road
traffic-lights-put-to-test/
programme.” Retrieved from
[5] “Basic Question”. Retrieved from http://www-
http://bettertransport.org.uk/sites/default/files/research-
formal.stanford.edu/jmc/whatisai/node1.html
files/Roads_to_Nowhere_October2012_web_spreads_0.pdf
[6] IBM (2013). “IBM Opens Doors to First African Research
[19] Srinivasan, D., Choy, M. C., & Cheu, R. L. (2006).
lab – Continent’s Grand Challenges in its Sights”. IBM News Room.
“Neural Networks for Real-time Traffic Signal Control”. IEEE
Retrieved from http://www-
Transactions on Intelligent Transportation Systems, 7 (3), 261-272.
03.ibm.com/press/us/en/pressrelease/42409.wss
Retrieved from http://www.jhuapl.edu/spsa/PDF-
[7] Deedrick, T. (2014). “IBM’s New Research Center in
SPSA/Srinivasan_etal_IEEETITS06.pdf
Kenya Is Poised to Impact the Region’s Growth”. Retrieved from
[20] Dargie, W. & Poellabauer, C. (2010). Fundamentals of
http://www.ibmsystemsmag.com/power/trends/ibmresearch/IBM%E2
Wireless Sensor Networks Theory and Practice (1st ed.). John Wiley
%80%99s-New-Research-Center-in-Kenya-Is-Poised-to-Im/?page=2
and Sons Ltd.
“PDCA12-70 data sheet,” Opto Speed SA, Mezzovico, Switzerland.
[8] Pascale, A., Nicoli, M., Deflorio, F., Dalla Chiara, B., & [21] Wikipedia. (n.d). “Sensor Node”. Retrieved from
Spagnolini, U. (2012). Wireless sensor networks for traffic http://en.wikipedia.org/wiki/Sensor_node.
management and road safety. IET Intelligent Transport Systems, 6(1), [22] Rada-Vilela, J. (2014). “Fuzzylite-paper 3.1 a fuzzy
67-77. Retrieved from control library. Retrieved from http://www.fuzzylite.com
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6157101&url [23] Watts, M. J. (n.d). "Fuzzy Systems". Retrieved from
=https%3A%2F%2Ffanyv88.com%3A443%2Fhttp%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Far http://mike.watts.net.nz/Teaching/IIS/Lecture6.pdf
number%3D6157101 [24] Nagori, V. & Trivedi, B. (Find out). "Which type of
[9] Zadeh, L.A. (1965). Fuzzy sets. Information and Control, Expert system? Rule Base, Fuzzy or Neural is most suited for
8(3), 338–353. Retrieved from evaluating motivational strategies on human resources: - An
http://www.cs.berkeley.edu/~zadeh/papers/Fuzzy%20Sets- analytical case study." International Journal of Business Research
Information%20and%20Control-1965.pdf and Management (IJBRM), 3, (5), 249-254. Retrieved from
[10] Fahmy, M. M. M. (2007). “An adaptive Traffic Signaling http://www.cscjournals.org/manuscript/Journals/IJBRM/volume3/Issu
for Roundabout with Four Approach Intersections Based on Fuzzy e5/IJBRM-113.pdf
Logic.” Journal of Computing and Information Technology (CIT), 15 [25] Davis, R., Buchanam, B. & Shortcliffe, E. (1977).
(1), 33-45. Retrieved from “Production rules as a representation for a knowledge-based
http://cit.srce.unizg.hr/index.php/CIT/article/view/1625/1329 consultation program.” Artificial Intelligence, 8, 15–45. Retrieved
[11] Khiang, T. K., Khalid, M., & Yusof, R. (1997). Intelligent from
Traffic Lights Control by Fuzzy Logic. Malaysian Journal of http://www.inf.ufpr.br/alexd/ARTIGOS_IA/Davis_Buchanan_Shortlif
Computer Science, 9 (2), 29-35.. Retrieved from fe_1977.pdf
http://www.researchgate.net/publication/229029935_intelligent_traffi [26] Annunziato, M. & Pizzuti, S. (n.d). “A Smart-Adaptive-
c_lights_control_by_fuzzy_logic System based on Evolutionary Computation and Neural Networks for
[12] Beattie, A. (2011). “What is the difference between the on-line short-term urban traffic prediction”. Retrieved from
artificial intelligence and neural networks?” Technopedia. Retrieved http://www.researchgate.net/publication/228578769_A_Smart-
from http://www-technopedia.com/2/27888/programming/what-is- Adaptive-
the-difference-between-artificial-intelligence-and-neural-networks System_based_on_Evolutionary_Computation_and_Neural_Network
[13] Dai, Y., Hu, J., Zhao, D., & Zhu, F. (2011). “Neural s_for_the_on-line_short-term_urban_traffic_prediction
network based online traffic signal controller design with [27] Goerke, N. (n.d). “Which one is better between online and
reinforcement training”. Intelligent Transportation Systems (ITSC), offline trained neural network?” Retrieved from
1045-1050. Paper presented at 14th International IEEE Conference, http://www.researchgate.net/post/Which_one_is_better_between_onli
Washington, DC. Retrieved from ne_and_offline_trained_neural_network
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=6083027&url [28] Michael, B.W. & Areolino de Almeida Neto. (2014).
=https%3A%2F%2Ffanyv88.com%3A443%2Fhttp%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Far “Optimization of Traffic Light Timing based on Artificial Neural
number%3D6083027 Networks”. Paper presented at 17th IEEE International Conference
[14] Singh, Y. P., & Mittal, P. K. (2013). “Analysis and on Intelligent Transportation Systems. Retrieved from
Designing of Proposed Intelligent Road Traffic Congestion Control http://scl.hanyang.ac.kr/scl/database/papers/ITSC/ITSC14_HTML/me
System with Image Mosaicking Technique”. International Journal of dia/files/0179.pdf
IT, Engineering and Applied Sciences Research (IJIEASR), 2 (4), 27-
31. Retrieved from http://www.irjcjournals.org/ijieasr/apr2013/7.pdf

www.ijcat.com 573