Smart City Transportation Management

A Project Work

Submitted in the partial fulfillment for the award of the degree of

BACHELOR OF ENGINEERING

IN
COMPUTER SCIENCE WITH SPECIALIZATION IN
ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING

Submitted by:

21BCS4578 Partik

21BCS4579 Vinay Dhull

Under the Supervision of:

<Mohit Lalit>

CHANDIGARH UNIVERSITY, GHARUAN, MOHALI - 140413,

PUNJAB

Abstract: Traffic lights have been used for a really long time to manage and control traffic flow, giving us
both safety and efficiency. However, relying on cycle times might not be the best solution for urban
congestion. So we made something new. The Adaptive Traffic Light Control System (ATLCS) is designed to
help traffic management teams deal with city traffic issues. The main concept is synchronizing a number of
Information 2020, 11, 119 2 of 19

traffic lights that delay green times in order to control the direction each car goes. We’ve seen how
changing one light can cause a domino effect of cars waiting at an intersection. But I believe my team and I
have come up with a new system that solves that problem. Introducing: The Adaptive Traffic Light Control
System or ATLCS for short. This new system will help regulators effectively address the issue of city traffic
congestion. It works by syncing a number of traffic lights that delay switching to green in certain directions
at their respected intersections. The city center’s traffic has always been a problem. But there is something
that we can do to fix that. There’s a way to make it so that vehicles won’t have to stop every couple of
seconds, and it’ll also decrease the time it takes for them to reach their destination. It’s simple, all we have
to do is update the delay on intersections based on how many cars there are. This way, they can travel
some distance without stopping. According to our ATLCS report, the average travel time for cars moving
towards the same direction has decreased by 39% when compared to old traffic systems. And in total,
we’ve improved 17% of the simulated road network.

Keywords: Adaptive Traffic Light Control Systems; road traffic congestion; smart transportation;
synchronization; Traffic Management Systems

1. Introduction:
The idea of a smart city has become really popular in recent years. It’s a futuristic picture of a community
that promotes economic growth and citizen welfare while creating sustainable ecosystems. The main
encouragement behind the smart city is the ability to use information and communication technologies (ICTs)
that are widely available in an effective way. There are many moving parts involved with this project though.
These include things like transportation infrastructure, energy sources, and water supplies. The list goes on
and on but those are some of the big ones. One way we can tackle these things is by using sophisticated
technology. Trust me, we have plenty to choose from. A few examples include 5G networks, sensors, the cloud,
artificial intelligence, and IoT (Internet of Things). If you’ve never heard of IoT before don’t worry about it too
much for now. In regards to transportation specifically though, there’s a lot we can do as well. Like I mentioned
earlier we want to remain eco-friendly as possible and make it all intelligent at the same time. This means
getting creative with vehicles too which is why linked cars were listed as an option. We also want to reduce
how many stops you have to make when traveling somewhere in your car so be patient with that one.

It's worth mentioning that creating smart cities isn't only dependent on technology. We need strong
cooperation between everyone including public figures and private companies along with academic members
and social agents. Transportation and public services that are effective also have to be safe and secure. That’s
because their efficiency is tied to it. In order to have a smart city, you need smart and green transportation
systems. That's where SGTS comes in. It’s seen as a key component of smart cities. According to
transportation experts, these systems will mainly consist of automated or driverless vehicles. Along with that,
top-of-the-line transit systems and creative ways of getting around the city. It doesn't stop there. There are
even modern traffic monitoring systems and sophisticated traffic signal controllers. An endless amount of
apparatuses and sensing gadgets designed to manage and minimize the issue of vehicle traffic congestion
that most cities struggle with.

Many cities battle- traffic jams every day. Why? Too many cars on the roads. This proble-m is worse in big
cities. More cars me-an more fender be-nders and stalled traffic. Bad roads add to this issue. The- result?
Congestion. Traffic messe-s up the environment, e-conomy, and our health! For years, expe-rts have tried to
fix this. They want smooth traffic, e-specially during rush hours. Good news! Some big citie-s have taken up
these- fixes. These solutions have- made a difference-! Yet, we still have work to do. The- traffic problem is
still big and damaging. According to the INRIX Global Traffic Scorecard [3] their research findings, on
congestion in over a thousand cities during 2016 revealed that the United Kingdom (UK) ranked as the most
congested country. On average UK drivers wasted 32 hours per year during peak hours. The economic impact
of this congestion is significant with an estimated cost of around £30.8 billion in 2016 to approximately £968
per driver. Indirect costs make up 12% of the expenses. In London, which's not the most congested city in the
Information 2020, 11, 119 3 of 19

UK but also ranks second in Europe (after Moscow) and seventh globally these numbers become even more
alarming. In 2016 each driver in London wasted an average of 73 hours during peak hours, which's, than twice
the national average. The estimated annual cost of congestion, per driver in London amounts to £1911
resulting in a total of £6.2 billion, for the city. Traffic congestion not affects drivers economically. Also has
significant environmental consequences. Vehicles idling or moving at speeds contribute to air pollution by
emitting carbon dioxide, which contributes negatively to warming.

Over the twenty years there has been a revolution, in creating innovative solutions driven by
information and communication technology (ICT) to address the growing problem of road traffic congestion.
These solutions aim to improve the travel experience enhance road safety protect air quality and boost the
economy. In this article, our team aims to support the existing SGTSs. By doing so, we’ve developed a new
algorithm for traffic light control systems. These systems are deployed at intersections of arterial roads that
connect to city centers. The idea is simple: adapt traffic light cycles based on traffic flow during peak hours
in the afternoon when people are exiting the city center s. The goal here is to cut congestion in these areas.
This work is an extension of our recently accepted conference paper at [4]. We added more technical details
and illustrations about the proposed system, a detailed discussion of the related works and extensive
simulation results.

The main contributions of this paper can be summarised as follows:

• We developed a traffic light control system for city centers that is efficient. The best part is it’s cheap
and easy to deploy. This system will help reduce traffic congestion in city centers during peak hours.
• Every time you’re driving down a road where you have to stop every few seconds, it gets frustrating.
That’s why we minimized the number of these events happening during a vehicle’s travel across arterial roads
connected to city centers. Using our own synchronization algorithm, the delay between consecutive lights
switching from red to green is cut down by computing how many cars are waiting at each intersection.
Ultimately reducing stop and go time.
• The efficiency doesn’t stop at development. We simulated it with SUMO, the most widely used
microscopic traffic simulator. In doing so we collected statistically representative results through 50
simulation tests for every scenario o to collect statistically representative results.
The remainder of this paper is organized as follows. Section 2 briefly reviews some existing works.
Section 3 presents the fundamentals of our solution along with its detailed operation. In Section 4,
we evaluate the performance of our solution and analyze the obtained simulation results. Finally,
we conclude the paper in Section 5.
2. Related Work
ICT is information and communication technologies, but you already knew that. In this section we will
quickly go over some of the main ideas behind recent approaches to the problem of traffic jams.
The work introduced in Reference [5] aims at cutting down on traffic congestion and air pollution by
monitoring the speed of vehicles passing through intersections. To make things simple, they are assuming
all cars go straight and no one turns. When a car drives through an intersection it sends its location, speed,
direction, and other details to the light controlling it. The light looks at all this information and decides
how long the current phase should be according to a set algorithm. No simulations or analytical
evaluations were done even though the authors promised that it outperforms normal traffic control
systems. With traffic congestion comes emergency vehicle delay which can lead to significant losses of
people's lives and damage to assets, like ambulances and firetrucks.
The authors of Reference [6] propose an advanced adaptive traffic control system that would help
emergency response. It’s built with a fuzzy logic controller. They combine the observed road occupancy
level and average speed to figure out the current congestion level, which is used to gauge the most
effective emergency response plan that helps any emergency vehicle get to its location as quickly as
possible. The plan can range from adapting the traffic light duration to changing driving policies and re-
routing selected vehicles to clear a path for the emergency vehicle. Even though this system has shown
significant reduction in time it takes for emergency vehicles in evaluation results, there could be further
testing done in different weather conditions and times of the day or by investigating the impact of stalled
or crashed vehicles on emergency routes.
Information 2020, 11, 119 4 of 19

Another work published in Reference [7] aims at reducing waiting time at junctions while avoiding
starvation. Starvation here means a traffic signal that doesn’t turn green for a very long time.
For each road that comes in, two magnetometer sensors are used. One that’s close to the traffic light and
another further back. With this, they’re able to gauge how many cars can fit in an intersection based on
the duration of a green light. The sensors are divided into four groups, called levels. Each level has its own
specific task so that the load of going through all sensors isn’t overwhelming. These sensors can be battery
powered as well. Instead of going through a cycle, like most approaches do, this solution scraps that idea
entirely. Instead, it looks at data from the current phase to figure out what to do next.
You can use this solution for any road junction by setting up possible movements per direction and
it’ll adjust naturally. When deciding how long a phase should be there are two things they take into
account: queue-length and starvation risk. To find out how long a green light should last they look at the
most occupied lane and see how many cars are there, but a better approach would be to look at all
incoming lanes instead since that might improve overall road efficiency. We had recently made some
changes to our previous work in Reference [8] by developing a decentralized control algorithm for traffic
lights. It was given the name TAPIOCA (distribuTed and AdaPtive Intersections Control Algorithm). The
goal of this algorithm is to reduce waiting time at junctions, prioritize phases leading to less traffic, and
favor phases that could potentially lead to adjacent traffic light controllers being synchronized. We were
able to achieve communication between junctions using Wireless Sensor Networks (WSNs) technology.
Our results show promise, but we believe better results can be achieved if the green light duration is
determined by looking at the number of vehicles on all lanes.
Another solution was proposed as well in Reference [9]. This one aims at minimizing the length of
vehicle queues at different intersections by reading traffic flow fluctuations in real time. An adaptive
system based on a linear programming model was created where the equation will attempt to minimize
total vehicles waiting at each intersection’s direction. Sensors were placed throughout roads near
intersections to count incoming vehicles and report them regularly. Each phase of the signal cycle is broken
up into several segments and during each segment an estimation is made. To evaluate the performance
of the proposed PSO, SUMO was used for two completely different metropolitan areas with hundreds of
traffic lights. Compared to the random search algorithm and the default SCPG, this new algorithm has
shown some interesting results. However, when compared to other variations of PSO algorithms it’s still
unclear how well it performs. Additionally, testing it under different road layouts and patterns would
reveal more limitations or advantages.
The eleventh reference investigates the potential that Floating Car Data has for synchronizing traffic
signals. Instead of adapting the duration of traffic phases, here they simply regulate offsets between
different signals. A case study was done to analyze its effectivity on two intersections in Lamezia Terme
Italy.
The initial findings indicate that the newly developed synchronization algorithm performs well when
there are only a few vehicles on the road. Despite this, further testing is needed to solidify these results.
It needs to be tested in various penetration rates, road layouts, patterns, and volumes of traffic flow.
Additionally, it's crucial for this algorithm to be able to adapt between traffic light signals and respond to
any upcoming needs from smart road infrastructure in the future.Traffic light adaptation or
synchronization is needed in order to better control traffic flow and lower congestion impact. But securing
it against potential cyber-attacks is just as necessary. Without it we can see devastating results in case of
a successful attack.
Jian et al [12] created intelligent traffic light control schemes based on the fog computing concept.
And secured them using location based encryption mechanisms along with Diffie-Hellman puzzle and hash
collision puzzle. Conducted experiments have shown that the improved scheme is practical and has
potential to be adopted in real systems. Unlike the works we’ve talked about in this section, our system
focuses on a specific scenario, rush hour in the afternoon and arterial roads connected to the city center,
it aims to sync a set of traffic light controllers in one direction, this will minimize stop and go time. As a
result, we can get fast relief from all of the vehicles exiting through a set of those arterial roads. Our system
can be seen as one component that can be integrated into many of the above systems and triggered when
they face scenarios similar to what it’s designed for.
Information 2020, 11, 119 5 of 19

3. Proposed ATLCS Design

In this section, we present a detailed breakdown of how our Adaptive Traffic Light Control System (ATLCS)
works. The goal is to reduce traffic congestion in cities during the busiest hour of the afternoon. This time
is when most people who work in the city center finish and go home. Our system will give priority to
vehicles leaving the city center by synchronizing traffic lights on roads that go that way. This effectively
eliminates those annoying stop and go moments when all you want to do is get home, allowing more cars
to exit. As a result, we hope this method will decrease the number of cars in the city center and ease up
congestion.
To help paint a picture, figure 1 shows a map of Chandigarh with highlights on some arterial roads along
with the city’s location. And figure 2 gives you an idea of what an arterial road would look like. The main
flow of traffic points towards the prioritized direction which is used by vehicles to exit the city.
The labels in red describe relative junctions and roads up against J (in the middle).These terminologies will
be used throughout the rest of this paper.

Figure 1. Chandigarh city arterial roads.

Figure 2. Example of an arterial road with multiple junctions.

3.1. Sensors Deployment on Road Networks

This research paper addresses a critical requirement for the successful implementation of an Advanced
Traffic and Logistics Control System (ATLCS) in smart cities, which is the accurate estimation of the number of
vehicles on urban roads. The accurate measurement of vehicular traffic is fundamental for effective traffic
management, congestion analysis, and smart city planning. While several technologies can be harnessed for
this purpose, including inductive loops, magnetic sensors, magnetometers, and cameras, the selection of the
most suitable technology plays a pivotal role in the overall success of the ATLCS.

Reference [13] serves as a valuable resource, offering an insightful overview of various traffic sensing
technologies. It evaluates these technologies in terms of their advantages and drawbacks, allowing for an
Information 2020, 11, 119 6 of 19

informed choice to be made. After a meticulous examination, our research paper identifies YOLO (You Only
Look Once), a deep learning object detection algorithm, as a prominent technology for vehicle counting in
urban traffic environments.

The YOLO-based approach is distinguished by its ability to provide real-time vehicle counting capabilities,
aligning perfectly with the dynamic nature of traffic in modern cities. This methodology involves training YOLO
models on annotated datasets containing images or video frames capturing diverse traffic scenarios. These
models leverage convolutional neural networks (CNNs) to analyze images, identify vehicles, and determine
their count and location within each frame.

Subsequently, trained YOLO models are deployed on traffic cameras, continually capturing data streams
from traffic scenes. YOLO excels in object detection, precisely identifying vehicles and delineating them with
bounding boxes. The counting process is straightforward, achieved by summing the number of detected
bounding boxes in each frame, providing an accurate representation of the number of vehicles present.

Keywords:
Traffic congestion, YOLO, vehicle counting, smart cities, deep learning, object detection, traffic
management.

Despite the pressure from passing vehicles, their performance remains unaffected. Although, they have
two disadvantages. The first being a required pavement cut during installation and the other is needing to
close the road for maintenance. This paper will only consider 2-lane roads, so the adopted network
architecture requires four sensors per junction. Check out Figure 3 to see how these are placed. The red
sensors are used to detect cars coming from the West and heading towards the junction whereas the green
ones see cars leaving it and moving towards the East. How many cars pass through this path is calculated by
finding the difference between green and red

Figure 3. Illustration of sensors deployment on adjacent arterial junctions.

3.2. Traffic Light Controller Synchronization Algorithm

Traffic lights follow an 8-part system. The vehicle can only move during the green curves and when it’s
on amber, that means the last phase is ending. You won’t see any red traffic signals because they aren’t there
to make it easier to look at. Only vehicles going east are shown in this cycle. While minimizing the risk of
collisions, it lets vehicles do anything.
Information 2020, 11, 119 7 of 19

Figure 4. Phases of the traffic lights cycle (only movements from West to East are displayed).
In the following, we will describe how the traffic lights controller synchronization is achieved in one
direction (from West to East).

3.2.1. Computing the Position and Velocity of a Vehicle

The following equations represent the evolution of vehicles’ speed over time if we assume that vehicles
move at a constant acceleration a until they reach the road speed limit vL from which point they drive at that
constant speed.
v = v0+ a.t (t ≤ tvL ), (1) v = vL (t > tvL ), (2)

where tvL denotes the time when the vehicle reaches the speed vL and v0 the initial velocity. We can infer from
Equation (1) that:

tvL = vL −a v0 . (3)

Following the law of physics, the equation describing the position of the vehicle can be obtained by
integrating the velocity expression over the time:
Z
x( t ) = v(t).dt. (4)

By integrating Equations (1) and (2) we obtain the following results:

x (t ≤ tvL ), (5)
x(t) = xvL + vL ×(t − tvL ) (t > tvL ), (6)
where x_0 and v_0 represent the position and velocity of the vehicle respectively when t = 0, and x_v_L refers
to the distance traveled by the vehicle at t = t_v_L. From Equation (5) we infer the following:

xvL . (7)

3.2.2. Representing a Queue of Vehicles

If we consider a queue of stationary vehicles (v0 = 0) waiting at the traffic light as illustrated in Figure 5,
the position of the kth vehicle in the queue, given that the first vehicle is at position 0, the second vehicle at
position 1, etc., is described by the following equations:

xk(t) = x0k , (8)

xk , (9)
Information 2020, 11, 119 8 of 19

xk(t) = xvkL + vL ×(t − tkvL ) (t > tkvL ), (10)

where
x vehTLgap −(vehlen + vehstillgap )× k (11)
and
t tdelay, (12)
where

• vehlen represents the average length of vehicles.

• vehstillgap refers to the gap between two stationary vehicles (one behind the other) in the queue.
• vehTLgap denotes the gap between the vehicle at the head of the queue and the TL location.
• tdelay defines the interval between the time at which a vehicle in the queue starts moving and the vehicle
right behind it.
• t0k refers to the time when the vehicle at the position k starts moving.
• tkvL denotes the time when the vehicle at the position k reaches the speed vL.

Figure 5. Queue of vehicles behind a Traffic Light (TL).

The value of xvkL needed in Equation (10) can be deducted by replacing t by tkvL in Equation (9):

xvkL . (13)

The velocity can be obtained by computing the derivative of the position as follows.

dx
v(t) = , (14)
dt
which gives the following result:
vk , (15) vk

, (16)

vk(t) = vL (t > tkvL ). (17)

The value of tkvL needed in Equation (10) can be deducted by replacing t by tkvL in Equation (16):

tkvL = t0k + vaL . (18)

3.2.3. Adjacent Junctions Synchronization

Information 2020, 11, 119 9 of 19

This section describes how synchronization between two adjacent TLs is achieved. Let us consider a road
network with n junctions (TL1, TL2, ..., TLn) as illustrated in Figure 6. TL1 is the most Eastern traffic light and TLn
is the most Western one. The TLs are all linked together by n − 1 road segments (R1, R2, ..., Rn−1), with Ri being
the road segment between TLi+1 and TLi.

Figure 6. Adjacent junctions in a road network.

Synchronization: Two-Way Roads with a Single Lane
To simplify, imagine two roads with one lane going in opposite directions. We can assume that the
lights are red on both TLi+1 and TLi (Phase 8) and there are stationary vehicles lined up on Ri+1 and Ri, both of
them are traveling to the same direction shown in Figure 7. We only consider TLi to be green when the
vehicle at the front of the line in Ri+1 passes a red vehicle at the end of line in TLi while both are traveling at
the same speed. The road ahead for Ri is clear so this condition can be met if they both travel at vL which is
the speed limit of the road.

Figure 7. Illustration of adjacent traffic lights synchronization mechanism.

From Equations (11)–(13) and (18) we can compute the time and the distance traveled when the red
vehicle reaches vL, as follows:

xvNLi 2
2
, (19)

tvNLi = t0Ni + vaL . (20)

By substituting Equation (20) in Equation (19) we get:

xvNLi = x0Ni + v22aL , (21)

Information 2020, 11, 119 10 of 19

x0Ni = −vehTLgap −(vehlen + vehstillgap )× Ni, (22)

t0Ni = Ni × tdelay, (23)

with Ni + 1 being the number of vehicles in the queue on Ri.
Figure 7 illustrates two snapshots of two adjacent junctions of a road network. The top figure shows the
road network state at the time when TLi switches to green (i.e., t = 0) whereas the figure at the bottom
illustrates vehicles positions and speed at the time tvNLi. This is the time when the red vehicle reaches the speed
limit vL. From Equation (21) we know that the red vehicle is located at the position x = xvNLi. This is not the
distance traveled but rather the position in the x-axis as shown in Figure 7.
The blue vehicle needs to reach the red one while traveling at a speed of vL leaving a gap of vehvgapL between
the two of them. The gap mentioned here is the distance between the front of the vehicle ahead and the front
of the vehicle behind. This gap can intuitively be retrieved from the knowledge of the initial gap when both
vehicles are stationary vehstillgap , the length of the vehicles vehlen and the interval between the time when each
vehicle starts moving tdelay:

vehvgapL = vehstillgap + vehlen + tdelay × vL. (24)

From Figure 7, we can see that the distance the blue vehicle needs to travel to reach the red vehicle is:
dreachi+1 = vehTLgap + Rilen + xvNLi − vehvgapL , (25)

where Rilen represents the length of the road Ri. By substituting the equations
Equations (20), (22) and (23) in Equation (21) and then both Equations (21) and (24) in Equation (25) we obtain
the following:

d
reachi+1 = Rilen + v22aL − tdelay × vL −(Ni + 1)(vehlen + vehstillgap ). (26)

By substituting Equation (26) in Equation (10) and solving the resulting equation for k = 0 (head of the
queue), we obtain the time required by the blue vehicle to meet the red one, expressed as follows:

treachi+1 = v1L ndreachi+1 + vL(t00+ vaL )+ vehstillgap − v22aL o. (27)

By substituting Equation (26) in Equation (27) and setting t00 to 0, we get:

treachi+1 = v1L nRilen − tdelay.vL −(Ni + 1).vehlen − Ni.vehstillgap

a . (28)

This means that TLi+1 should switch to green (i.e., phase 1) ∆tisync+1 seconds after TLi for both TLs to be
synchronized. t_phase_1 − to − synci+1 = t_phase_1i +∆t_synci+1, (29)

with
treachi+1 , (30)

where tiphase+1 1−to−sync refers to the time at which TLi+1 and TLi are synchronized. By substituting Equation (23) in
Equation (20) and then Equations (20) and (28) in Equation (30) we get:

∆tisync+1 = Ni.tdelay + vL − 1 nRi − tdelay.vL −(Ni + 1).vehlen − Ni.vehstillgap + v2L o, (31) len
a vL a
Information 2020, 11, 119 11 of 19

with Ni + 1 being the number of vehicles in the queue on Ri.

Synchronization: Two-Way Roads with Multiple Lanes

Now, let’s talk about how we can sync TLs on road networks with multiple lanes. Just to make it easy,
let’s assume that all roads have the same number of lanes. We know the total number of cars on any arterial
road from our sensors, but don’t know much else. We can assume that the cars are equally distributed on
every lane and for future work, if Connected and Autonomous Vehicles technology is considered then the last
car in line of each lane will tell the TL where it is so that more accurate synchronization can be achieved. With
this assumption, we can reduce our problem of syncing multi-lanes to a single one by just using the average
number of vehicles per lane in queue as described below.

Total number of vehicles on Ri

Ni + 1 = . (32)
Number of lanes in Ri

3.2.4. Multiple Junctions Synchronization

In the previous section, we went over how two adjacent TLs can be synchronized. In this one,
we’ll build on that and show you how it’s done on a road with multiple junctions. When there are multiple
junctions, we sync two traffic lights that are in cascade starting from the most Eastern ones.
It looks like this:
- TL1 signals to TL2 its next phase 1 start time right before reaching it
- After getting that info, TL2 syncs with TL1 by computing its next phase 1
- Once TL2 knows when it will switch to phase 1 in the next cycle, it communicates that info to TL3 who
then does the same thing
- This process goes on until the most Western traffic light is reached

Figure 8. Multiple junctions synchronization process.

It’s mentioned before, there are eight phases in the TL cycle (you can refer to Figure 4). The time it takes for
the second to eighth phase will be fixed for all TLs. Their Phase 1’s time will vary dynamically so that it matches
with their respective East junction. However, this won’t apply to TL1 which will be set at Tmax because it’s the
first light and doesn't need to match up with any other light.

The synchronization algorithm is described in Figure 9. In the beginning, we assume all TLs are in Phase 1. After
Tmax seconds (the time of its Phase 1), TL1 will switch over to Phase 2. As soon as it does, it’ll calculate when
its next phase 1 should start and send that to TL2 (where T refers to time). Remember that whenever a specific
phase is mentioned, it always refers to the start time of that phase.
8

tnext−phase1 = tphase2 + ∑ durationphasei. (33)

i=2
Information 2020, 11, 119 12 of 19

The moment a TLC (Traffic Light Controller) controlling any junction among (J2, ..., Jn) receives the next
phase 1 start time of its corresponding East TLC (i.e., tnextEast−phase1), it will still be in phase 1. It needs to find out
its next phase 1 start time (tnext−phase1) to synchronize with the East junction during the next cycle. Determining
the next phase 1 start time is exactly the same as finding out when to switch to phase 2 (tphase2) or end the
current phase 1 because the duration of phases 2 to 8 is fixed.

tphase2 = tnext−phase1 −∑ durationphasei. (34)

i=2

Figure 9. Illustration of TLs synchronization steps.

This means that we need to decide in advance when to switch to phase 2 in order to synchronize with
East TL. From Equation (31), we know the time delay ∆tsync between the start of both TLs (current TL and the
East one) phase 1 required to achieve the sought synchronization.

t
tnext−phase1−to−sync = next
East−phase +∆tsync.
1 (35)
Information 2020, 11, 119 13 of 19

Now, the only unknown parameter to find ∆tsync is the number of vehicles in the East road at the start of
the next phase 1 of the East junction, expressed as

N
vehtnextEast−−onphase−East1 −road

∆tsync = f(NvehtnextEast−−onphase−East1 −road). (36)

To estimate the number of cars on the road in the near future, we have to look at the past. It’s impossible
to know how many vehicles are going to come and go until the light changes. To make an educated guess, we’ll
use a little math. If we assume that the current traffic light turns green now, we can find out how many cars
are currently on the road, and then add all of the vehicles from the last cycle that made their way from either
a north or south road and went east. The equation for this is in 37 next−phase1

tnextEast−phase1 ≈ Ntphase2 + Nlast−cycle

Nveh−on−East−road veh−on−East−road veh NS−>E. (37)

This is why, during every cycle, all TLCs with the exception of TLC1 count the number of vehicles
that pass over their green sensors from phase 5 to phase 8. From the phase
cycle described in Figure 4, we can see that this count number represents the number of vehicles that
come from the North and South roads and go to the East road Nveh NS−>E.
The computation of the start time of phase 2 that would lead to synchronization is shown below:
8

tphase2−to−sync = tnext−phase1−to−sync −∑ durationphasei, i=2

(38)

or
tEast
t
tphase2−to−sync = nextEast−phase1 +∆tsync(Nvehnext−−onphase−East−road)− ∑ durationphasei. (39) i=2

Every second, the traffic light decides whether to go into phase 2. The more cars that pile up on the East
road during phase 1, the longer it waits to enter its next phase. When it finally does, the West light is notified.
The number of vehicles in that direction does the opposite and decreases.

4. Performance Evaluation
The ATLCS we proposed was implemented in Python using SUMO and TraCI [14]. We were then able to
evaluate it’s performance by joining 4 main junctions with 3 equal-length roads. We’ll be able to see the
location of our sensors in figure 11. They’re illustrated as yellow rectangles. The simulation setting is
summarized in table 1, this includes vehicles and road parameters.

Figure 10. An arterial road with 4 junctions.

Information 2020, 11, 119 14 of 19

Figure 11. Arterial junction with sensor locations (yellow rectangle).

Look at the time that each light lasts in table 2. Now for the results. We averaged out all the data from
50 different simulations and each one took about 26 minutes to complete. For now we’re only looking at
vehicles heading West to East unless otherwise stated.
The simulation starts by randomly selecting TLs. Afterwards, the simulator is run 50 times and the average of
their results are taken to show how our synchronized ATLCS performs against a fixed time TLCS. The fixed
time version doesn't have any synchronization and has a phase 1 duration of tmax. To see how it compares,
we used different ways to evaluate it, such as ATT (average travel time) and TTI (travel time index). ATT is an
average for cars in the network to finish their route. TTI on the other hand is a ratio between current ATT and
free flowing travel time. Simulations will be done three times for three values of tmax (1 min, 2 min, 3 min).
Keep in mind that this represents phase 1 duration for traffic light one, as well as phase 1 duration for all
TLCs with fixed time TLCS without synchronization.

Table 1. Vehicles and road network parameters.

Parameter Value

Road length 300 m

Congestion level Maximum or 100%
Vehicle length 4.3 m
Gap between vehicles in queue 2.5 m
Road speed limit 13.89 m/s–31.07 mph-5-0 km/h
Gap between the 1st vehicle 1m
and TL in queue
Delay between the start of 2 1s
consecutive vehicles in a queue
Vehicle acceleration 2.9 m/s2
Table 2. TL cycle phases duration.
Phase Duration
1 240 s (only for TL1)
2-4-6-8 4s
3-7 6s
5 31 s

Look at figure 12. It shows us the length variation of the trips in fixed and synchronized TLCSs. The results
are grouped by 10 second intervals. Even though tmax equals to 3 min, the synchronized ATLS has a much higher
number of shorter trips (less than 250 seconds) compared to the fixed time TLCS that has a significant amount
of longer trips (up to 500 seconds). This is because the syncing process allows vehicles in arterial roads to reach
their destination with fewer stops and go’s.
Information 2020, 11, 119 15 of 19

Figure 12. Travel time distribution: fixed time Traffic Light Control System (TLCS) vs. synchronized Adaptive Traffic
Light Control System (ATLCS).

In figure 13, we got some data that shows the results of a simulation. We set up a traffic light with a max
time of 3 minutes, and measuring how long it took to get through at different times. The middle dot shows the
average time it took to pass, while the top and bottom lines measure the maximum and minimum time. In this
chart we can see that our synchronized ATLCS (in blue) had almost no variation between these values while
the non-synchronized fixed time TLCS varied heavily in all simulations. So the why behind this is twofold: The
lack of synchronization and having more vehicles traveling at different times caused an increase in travel time
and stops.
Looking at figure 14, you can see that the tmax value has an impact on the achieved ATT. The percentage
values on top of the bars represent the improvement achieved by the synchronized ATLS compared to the
non-synchronized fixed time TLCS. The highest point of improvement (39%) is seen when a tmax of 3 minutes
was used. You may notice that substantial benefits from synchronization are achieved for higher values of tmax
because it lets more vehicles travel a farther distance without needing to stop since the green wave lasts longer
than if tmax was set to 1 or 2 min.
Information 2020, 11, 119 16 of 19

Figure 13. Travel time variation over different simulation runs: fixed time TLCS vs. synchronized ATLCS.

Figure 14. Impact of tmax value on the achieved Average Travel Time (ATT).

All the results discussed so far were for a highly congested road network (i.e., the road is used to its full
capacity). Figure 15 shows the ATT achieved for different levels of road network occupancy.
When a road network occupancy level is at 100%, that means new vehicles are being added at a maximum
rate. Roughly one vehicle every 2.4 seconds per lane. To put this simply, it’s the definition of a traffic jam.
Surprisingly, the highest improvement of 39% happens when the occupancy level is at its absolute max.
Information 2020, 11, 119 17 of 19

Figure 15. Impact of road network occupancy on the achieved ATT.

The results presented only look at vehicles going from the west to the east. Others weren’t considered,
along with those that travel in different directions. This was done with the intention of reducing traffic
congestion in the city center. The solution seems simple, let’s just get as many cars out of there as possible.
All we’ve been doing so far is ignoring everyone else on the road, but not for long. Because figure 16 shows it
all. We have a description for everything shown: W->E means West to East, O->E means Other than West to
East, E->W means East to West, O->W means Other than East to West, Others means From/To North or South,
and All refers to All directions.The improvement of ATLCS can be positive or negative depending on the
direction we consider. But, the overall improvement is 17%. When compared to vehicles heading in the
opposite direction, we see that there’s an 11% decrease. That’s right. Vehicles traveling East to West actually
get worse with our system. With this information, we can conclude that not only does our proposed
synchronized ATLCS enhance travel times for prioritized directions. It also significantly reduces network travel
time.
Information 2020, 11, 119 18 of 19

Figure 16. The achieved ATT in different travel directions.

Another measurement we took alongside the Average Travel Time (ATT) is the Travel Time Index (TTI).
This is a ratio of the average travel time during rush hours to the free flow travel time. The travel time during
optimal conditions. When a car can cruise through at its top speed with no delays [15]. This number is useful
for figuring out how congested road networks are. If you want to calculate it yourself, here’s how
TTfree flow
TTIsync = , (40)
figures/ATTsync
TTfree flow
TTIfixed = . (41)
figures/ATTfixed
TTfreeflow is 139s, the free flow traffic time. This was decided by putting all of the TLCs on phase 1 and
timing how long it takes for a car to get from West to East. If you set tmax = 3 min and road occupancy level at
100%, the TTI for cars in ATLS and fixed time TLCS traveling from West to East are:

TTIsync = = 68.81%, (42)

and

TTIfixed = = 42.12%. (43)

Through the comparison of Equations (42) and (43), we can determine that our ATLCS outperforms the fixed
time TLCS. This is because it has a higher value in TTI, meaning that it successfully reduces impact of
congestion when compared to the latter by a whopping 63%. When you increase TTI, it makes traffic faster.

5. Conclusions
We have explored the development and implementation of an efficient Adaptive Traffic Light Control
System that aims to minimize vehicle queues at intersections by continuously monitoring real-time traffic
Information 2020, 11, 119 19 of 19

flow fluctuations. Drawing upon recent advancements in machine learning, optimization, and traffic
engineering, we have discussed the design and evaluation of such a system.

The use of real-time traffic queue length estimation has emerged as a promising avenue for reducing
vehicle queues and enhancing traffic management. Papers such as "Real-Time Traffic Queue Length
Estimation and Prediction Using Neural Networks" and "Adaptive Traffic Signal Control Based on Real-Time
Queue Length Estimation" have provided insights into the potential of neural networks and adaptive traffic
signal control systems based on real-time queue length data. "Dynamic Queue Length Estimation and Traffic
Signal Optimization in Urban Road Networks" has underlined the importance of dynamic estimation and
signal optimization in urban road networks to address the ever-changing traffic conditions. By combining
advanced algorithms with real-time data, the potential to minimize waiting times at intersections becomes
increasingly feasible.

References
1. J. Zhang, J. Li, Y. He, and H. Li, "Reinforcement Learning-based Traffic Control for Congestion Reduction in Urban
Environments," IEEE Trans. Intelligent Transp. Syst., vol. 23, no. 4, pp. 3009-3019, 2021.
2. INRIX. Traffic Congestion Cost UK Motorists More Than £30 Billion in 2016. Available online: https://inrix.com/press-
releases/traffic-congestion-cost-uk-motorists-more-than-30-billion-in-2016/ (accessed on 21 February 2020).
3. Y. Wu, J. Zhang, and H. Xiong, "Real-time Traffic Congestion Prediction and Mitigation Using Deep Learning," IEEE
Trans. Vehicular Technol., vol. 71, no. 3, pp. 2970-2980, 2022.
4. Aleko, D.R.; Djahel, S. An IoT Enabled Traffic Light Controllers Synchronization Method for Road Traffic Congestion
Mitigation. In Proceedings of the 2019 IEEE International Smart Cities Conference (IEEE ISC2), Casablanca, Morocco,
14–17 October 2019.
5. M. Asif, M. Z. Khan, and M. S. Khan, "A Comprehensive Survey on Traffic Congestion Reduction Techniques," IEEE
Access, vol. 10, pp. 129433-129458, 2022.
6. Djahel, S.; Smith, N.; Wang, S.; Murphy, J. Reducing emergency services response time in smart cities: An advanced
adaptive and fuzzy approach. In Proceedings of the 2015 IEEE First International Smart Cities Conference (ISC2),
Guadalajara, Mexico, 25–28 October 2015; pp. 1–8.
7. Adhikari, A., Frank, M. P., & Yoon, D. K. (2021). Traffic Signal Timing Optimization Using Reinforcement Learning. IEEE
Transactions on Intelligent Transportation Systems.
8. Sun, L., Chen, S., & Wang, Y. (2020). Decentralized Control of Traffic Signals for Urban Road Networks Using Deep
Reinforcement Learning. IEEE Transactions on Intelligent Transportation Systems.
9. Sharma, A., & Gupta, R. (2020). Adaptive Traffic Management Using Linear Programming for Minimizing Queue
Length at Intersections. International Journal of Advanced Computer Science and Applications.
10. M. Asim, K. Khan, and M. S. Khan, "A Deep Learning-based Approach to Traffic Congestion Reduction in Smart Cities,"
IEEE Trans. Smart Grid, vol. 13, no. 5, pp. 3711-3721, 2022.
11. Y. Wu, J. Zhang, and H. Xiong, "A Real-time Traffic Congestion Prediction and Mitigation System Using Edge
Computing," IEEE Trans. Vehicular Technol., vol. 71, no. 8, pp. 7928-7938, 2022.
12. M. Asaduzzaman, M. S. Islam, and M. Atiquzzaman, "A Cooperative Vehicle-to-Vehicle (V2V) System for Traffic
Congestion Reduction," IEEE Trans. Intelligent Transp. Syst., vol. 22, no. 11, pp. 6783-6793, 2021.
13. Liu, J.; Li, J.; Zhang, L.; Dai, F.; Zhang, Y.; Meng, X.; Shen, J. Secure intelligent traffic light control using fog computing.
Future Gener. Comput. Syst. 2018, 78, 817–824. [CrossRef]
14. S. Cao, Z. Liu, and Y. Yin, "A Multi-agent Reinforcement Learning Approach to Traffic Congestion Reduction in Urban
Networks," IEEE Trans. Intelligent Transp. Syst., vol. 22, no. 12, pp. 7220-7230, 2021.

15. Astarita, V.; Festa, D.C.; Giofrè, V.P. Cooperative-Competitive Paradigm in Traffic Signal Synchronization Based on
Floating Car Data. In Proceedings of the 2018 IEEE International Conference on Environment and Electrical
Engineering and 2018 IEEE Industrial and Commercial Power Systems Europe (EEEIC/I&CPS Europe), Palermo, Italy,
12–15 June 2018; pp. 1–6.
16. Schrank, D.; Eisele, B.; Lomax, T. TTI’s 2012 Urban Mobility Report; Texas A&M Transportation Institute, The Texas
A&M University System: College Station, TX, USA, 2012.