AI-Based Adaptive Signaling for Traffic Control

Around Roundabouts
Bakheeta Ali Almansoori Salma Saif Almansoori Humian Almansoori
Electrical Engineering Department Electrical Engineering Department Electrical Engineering Department
2022 Advances in Science and Engineering Technology International Conferences (ASET) | 978-1-6654-1801-0/22/$31.00 ©2022 IEEE | DOI: 10.1109/ASET53988.2022.9735009

Higher Colleges of Technology Higher Colleges of Technology Higher Colleges of Technology

Ruwais, United Arab Emirates Ruwais, United Arab Emirates Ruwais, United Arab Emirates
[email protected] [email protected] [email protected]

Raeesa Ahmed Almansoori Irfan Ahmed Khalil Shahid

Electrical Engineering Department Electrical Engineering Department Electrical Engineering Department
Higher Colleges of Technology Higher Colleges of Technology Higher Colleges of Technology
Ruwais, United Arab Emirates Ruwais, United Arab Emirates Ruwais, United Arab Emirates
[email protected] [email protected] [email protected]

Abstract— In the cities, car accidents are rapidly increasing, Timing and switching cycles of the stationary traffic lights
and many car accidents happen on the roundabouts. Usually, are predetermined amid prevailing traffic conditions. There
the driver's negligence in checking the left side before entering are various lanes. They are not dealing with real-time info.
the roundabout causes an accident in the roundabout. Consequently, this means that they should not consider the
Therefore, a robust and reliable system needs to be implemented non-uniform and ever-changing essence of traffic conditions
as an effective solution for this problem. We propose an artificial
intelligence-based roundabout traffic control system (ARTCS ) [5]. In [6], the authors present an image processing solution
to assist drivers near roundabouts. The proposed system is for the traffic congestion problems at the intersection of the
based on Artificial intelligence to detect the car from the left side roads. They use LabView to validate the efficacy of the
of the roundabout and notify the driver. The system controller proposed solutions. Ahmed et al. [7] design a controller that
records a video of the roundabout movements through the takes the input from the field detectors about all the influ x
camera to detect cars using a trained neural network. Once a and outflux traffic and provides satisfactory performance in
vehicle has been detected at a predefined distance, it notifies the highly congested scenarios. The public transport priority
driver if it's safe to enter the roundabouts using signaling lights timing around the roundabout has been investigated in the [8].
that shows red light for stop and a green light for passing. The The current traffic control system uses modern technology,
proposed solution provides many benefits, such as reducing car
accidents, decreasing time delay in a roundabout, increasing such as image recognition, machine vision controls, and
fuel efficiency, and saving public properties. artificial intelligence for traffic routing decisions. The graph
Keywords—machine learning, deep learning, traffic flow neural networks (GNN) have shown great potential traffic
control congestion automation [9]. In this model, the graph represents
the roads, and the deep neural network learns the traffic flow
I. INT RODUCT ION dynamics.
A car accident, also called a traffic collision, None of those mentioned above uses AI techniques to
happens when a vehicle crashes into another vehicle, street assist the traffic at roundabouts. We develop an AI-based
sign, creature, or other fixed obstacles, such as a tree, shaft, system that supports traffic around roundabouts and assists
or building. Car accidents frequently result in injury, death, the drivers. The objectives for designing this system are:
disability, and property damage, which cause money
expenses to both society and the people [1]. Various variables • Detect the vehicle heading towards the roundabout using
add to the danger of crashes, including vehicle configuration, AI techniques.
speed of activity, street plan, street condition, driving • Estimate the distance of the vehicle using AI techniques
abilities, debilitation caused by alcohol or drugs, driver from the roundabout.
behavior, and road racing. About 54 million people around Even though the safety instruction of how to pass through the
the World in 2013, sustained injuries from traffic collisions. roundabout has been taught to the drivers, the main reason for
These traffic collisions resulted in around 1.3 million deaths the accident is similar to what is shown in Figure 1. Car A
each year [2]. Day after day, car accidents are rapidly enters the roundabout without checking if there's no car
increasing. However, in most cities nowadays, the common coming like car B from the left side.
cause for car accidents is mainly on the roundabouts since
they replace some of the traffic lights with a roundabout to
decrease the percentage of traffic jams. But this solution
opens the field to cause traffic collisions in the roundabout
spot.
Almost all urban centers in the World use lights to control
traffic at the intersections and roundabouts [3], [4]. The lights
are flipping from red, which means to stop, to green, which
means to pass. Over time, it's got their developments of
various traffic light management systems; the most widely
used are static traffic lights. And the lighting on the vehicle.
Figure 1: Example of a car accident in the roundabout [2].

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 19,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
 The problem that causes an accident in the
roundabout is the driver's neglect of checking the left side
before entering the roundabout. Therefore, a powerful AI
system needs to be implemented as an effective solution. In a
normal situation, if the driver wants to pass through the
roundabout, he needs to slow down and stop the car
temporarily to check the left side from the car's existence.
With the AI-based embedded system, the vehicle can be
detected, and the drivers could be notified if there is a car on
the left side of the roundabout. This system help reduces
accidents and increases the flow of traffic.

II. SYST EM M ODEL AND PROBLEM FORMULATION Figure 3: Primary roundabout

We choose the primary roundabout, as shown in Fig.2. Our Close-circular intersections are modern roundabouts.
goal is to design the rotor for maximum safety and operational With fewer conflict points and lower speeds, they are an
efficiency. Giving preference to those inside the rotor efficient intersection type, and they provide easier decision-
contributes to the ease of traffic inside it. We consider a making than other types of an intersection. They need less
circular-shaped intersection that helps regulate traffic and maintenance than traffic signals as well. We found well-
reduce traffic density. The benefits of this roundabout are designed roundabouts to reduce accidents (especially fatal and
traffic non-stop regulation and operational efficiency. In this severe injury collisions), traffic delays, fuel consumption, and
roundabout, the priority is for vehicles' movement inside the air pollution. They also have a traffic-calming effect by using
roundabout, and the driver of the car outside the roundabout geometric architecture to decrease vehicle speeds instead of
to wait and give priority to vehicles inside the roundabout. solely traffic control systems. A well-designed roundabout
accomplishes the balance between safety and efficiency. Good
design is a process of creating the smooth curvature,
channeling, and deflection necessary for consistent speeds,
well-marked lane paths, and sufficient distance of sight to be
achieved. Selecting Shape and Positioning is a significant
decision for the roundabout shape, as the condition will impact
design elements that influence the roundabout's safety
efficiency and function.

According to the vehicles on roads and their distances, the

controller decides to turn on the traffic signal's green/red
lights. The decision is sent to the traffic lights that turn on the
required light, i.e., red and green. The entire system is
Figure 2: System Model designed to prevent traffic signal collisions and incidents with
effective traffic control. When the camera detects a vehicle or
A. Primary roundabout: obstacle in any lane at a distance from the threshold, it
switches the red signal on that lane whose traffic has gone
through the block.

III. M ET HODOLOGY
The proposed AI-based roundabout traffic control
- Improved efficiency: Yield-controlled architecture system takes input from the camera. Once the camera detects
means fewer stops, shorter queues, and fewer the car's existence in the roundabout, the signal is transmitted
delays. to the controller to decide the permission to enter in the
- Safer speeds: Encourages lower speeds for cars, roundabout. The controller changes the light sign according
allowing more time for drivers to respond. to the following sequence: if a car has been detected, the
- Aesthetics: Allows for landscaping and distance is measured to see if it's bigger than the threshold. If
beautification. the distance is more significant than the threshold, it turns on
the green light. While if the distance is smaller than the
The only disadvantage is that the primary roundabout threshold, the signal displays the red light. However, if a car
[10] is Not long-term cost-effective: traffic signals mean has not been detected, the signal displays the green light as
higher long-term costs for operations and maintenance. shown in Fig. 4.

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 19,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
 Input = 0 (no vehicle or vehicle is far from threshold)
Input = 1 (Vehicle has crossed threshold distance)
Output = 0 (Red light)
Output = 1 (Green light)

Table 1: Functional table

Input Output

A B C D w x y z
Figure 4: AI-based proposed design for traffic control
0 0 0 0 1 1 1 1
A. Design Concept 0 0 0 1 0 1 1 1
0 0 1 0 1 1 1 0
0 0 1 1 0 1 1 0
0 1 0 0 1 1 0 1
0 1 0 1 0 1 0 1
0 1 1 0 1 1 0 0
0 1 1 1 0 1 0 0
1 0 0 0 1 0 1 1
1 0 0 1 0 0 1 1
1 0 1 0 1 0 1 0
The idea of our design is that it is an accessible roundabout.
It is easy for them to enter the roundabout. It will reduce the 1 0 1 1 0 0 1 0
time required to slow down near the roundabout to make sure 1 1 0 0 1 0 0 1
there is no car from the left side of the roundabout because 1 1 0 1 0 0 0 1
we use camera and machine learning techniques to detect the 1 1 1 0 1 0 0 0
cars near the roundabout. Once the camera detects the car's 1 1 1 1 1 0 0 0
existence in the roundabout, the signal is transmitted to the IV. SIMULAT ION SCENARIO AND RESULT S
controller, and the red light turns on. The controller changes When there is no incoming car from any intersection (𝐴𝐴 =
the light sign according to the following sequence: if a car has 𝐵𝐵 = 𝐶𝐶 = 𝐷𝐷 = 0), then all signals are green (𝑤𝑤 = 𝑥𝑥 = 𝑦𝑦 =
been detected, the camera will measure the distance to see if 𝑧𝑧 = 1). Similarly, when there is an incoming car only fro m
it's bigger than the threshold. If space is more significant than intersection D ( 𝐴𝐴 = 𝐵𝐵 = 𝐶𝐶 = 0, 𝐷𝐷 = 1 ), then 𝑤𝑤 = 0 (red
the threshold, the signal will display the green light. While if signal), and so on.
the distance is smaller than the threshold, the signal will We use a deep learning toolbox in MATLAB. After
indicate the red light. However, if a car has not been detected, developing the vehicle detection and distance estimation deep
the signal will display the green light. neural network, we check its performance in terms of
percentage accuracy. The input dataset is divided into 70%,
B. Design Evaluation
15%, and 15% for the training, validation, and testing phases.
Roundabouts are circular, not index intersections Because of the real-time nature of decision-making with
where all traffic moves counterclockwise around a central video streaming, we employ faster Regional Convolution
island, but we add the artificial intelligence-based roundabout Neural Networks (faster-RCNN), and You Look Only Once
traffic control system in our design. Traffic slows down before (YOLO).
approaching the roundabout and returns to traffic already
inside the roundabout. With one or more rotating lanes,
roundabouts can be built. Design options allow channelization
of the correct turns to bypass the circulating roads. The block diagram clarifies the principle of YOLO 2:
Fig. 5 shows the evaluation model for the proposed
ARTCS. This is a 4-way roundabout intersection with inputs
of A, B, C, D, and w, x, y, and z as outputs.

Figure 6: Block diagram clarifies the principle of YOLO 2

1. Computer vision toolbox.

2. Deep learning toolbox.

Figure 5: Roundabout inputs and outputs

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 19,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
 A. Video labeler detect:
We use a ground truth labeler to create the training dataset
from the recorded video. The stepwise process of the creation
of the training dataset is given below.

Step 1
• Select video labeler.

Step 2
After finishing all the cars labels and detecting the
cross threshold, click on automate and then export to
save the work.

V. RESULT S AND VERIFICAT ION

The entire system is designed to prevent traffic
signal collisions and incidents with effective traffic control.
When the camera detects a vehicle or obstacle in any lane at
a distance from the threshold, it switches the red signal on
that lane whose traffic has gone through the block.

Table 2: Traffic signal system

Vehicle in Vehicle in Vehicle in Vehicle in Signal Signal Signal Signal

Lane A Lane B Lane C Lane D Light Light Light Light
ON in ON in ON in ON in
Figure 8: Far Label. lane A lane B lane C lane D

Step 3 Not Not Not Not Green Green Green Green

Add to only one ROI as crossThres. Select Algorithm Detected Detected Detected Detected

(Point Tracker) then detect the cars manually. Not

Detected
Not
Detected
Not
Detected
Detected Red Green Green Green

Not Not Detected Not Green Green Green Red

Detected Detected Detected
Not Not Detected Detected Red Green Green Red
Detected Detected
Not Detected Not Not Green Green Red Green
Detected Detected Detected
Not Detected Not Detected Red Green Red Green
Detected Detected
Not Detected Detected Not Green Green Red Red
Detected Detected
Not Detected Detected Detected Red Green Red Red
Detected
Detected Not Not Not Green Red Green Green
Detected Detected Detected
Detected Not Not Detected Red Red Green Green
Detected Detected
Detected Not Detected Not Green Red Green Red
Detected Detected
Detected Not Detected Detected Red Red Green Red
Detected

Detected Detected Not Not Green Red Red Green

Detected Detected
Detected Detected Not Detected Red Red Red Green
Detected
Detected Detected Detected Not Green Red Red Red
Detected
Detected Detected Detected Detected Green Red Red Red

In addition to verifying the functionality with table 2, we also

do the verification at different times of the day. We train our
model for daytime as well as nighttime because car headlights
ON/OFF are two different scenarios for the machine learning
model.

Training with car's headlights were OFF

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 19,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
 • Car's headlights were ON

Training with car's headlights were ON VI. CONCLUSIONS

In this paper, we propose an artificial intelligence-based
roundabout traffic control system that assists the drivers
crossing the roundabout intersections. The training dataset is
obtained from the real-world video recording near a
roundabout, and then the MATLAB video labeler app is used
to produce the training dataset. Because of the real-time
nature of decision-making with video streaming, we employ
faster Regional Convolution Neural Networks (faster-
RCNN), and You Look Only Once (YOLO). Finally, the
trained model is deployed and tested with 98.53% accuracy.

REFERENCES
[1] A. Flannery and T . K. Datta, “ Modern roundabouts and traffic
crash experience in United States,” Transp. Res. Rec., no. 1553,
pp. 103–109, 1997.
[2] WORLD HEALT H ORGANIZATION, “ Road traffic injuries.”
The results are shown below: [Online]. Available: https://www.who.int/news-room/fact-
sheets/detail/road-traffic-injuries. [Accessed: 06-Jan-2022].
[3] R. L. Reid, “ All About the Roundabout,” Civ. Eng. Mag., vol. 91,
no. 2, pp. 34–39, Mar. 2021.
[4] M. Ebrahim Fouladvand, Z. Sadjadi, and M. Reza Shaebani,
“ Characteristics of vehicular traffic flow at a roundabout,” Phys.
• Car's headlights were OFF Rev. E - Stat. Physics, Plasmas, Fluids, Relat. Interdiscip. Top.,
vol. 70, no. 4, p. 8, 2004.
[5] M. Owais, O. Abulwafa, and Y. A. Abbas, “ When to Decide to
Convert a Roundabout to a Signalized Intersection: Simulation
Approach for Case Studies in Jeddah and Al-Madinah,” Arab. J.
Sci. Eng., vol. 45, no. 10, pp. 7897–7914, Oct. 2020.
[6] P. Majumder and S. Chaudhury, “ Vehicle detection and counting
the number of vehicles for intelligent traffic control using
labview—an image processing approach,” Lect. Notes Electr.
Eng., vol. 664, pp. 503–513, 2020.
[7] F. Ahmed and Y. E. Hawas, “ An integrated real-time traffic signal
system for transit signal priority, incident detection and congestion
management,” Transp. Res. Part C Emerg. Technol., vol. 60, pp.
52–76, Nov. 2015.
[8] O. De Keyser, M. Hillewaere, P. Audenaert, and B. Maenhout,
“ Optimising the public transport priority at road intersections,”
IET Intell. Transp. Syst., vol. 12, no. 8, pp. 986–994, Oct. 2018.
[9] C. Englund, E. E. Aksoy, F. Alonso-Fernandez, M. D. Cooney, S.
Pashami, and B. Åstrand, “ AI Perspectives in Smart Cities and
Communities to Enable Road Vehicle Automation and Smart
T raffic Control,” Smart Cities, vol. 4, no. 2, pp. 783–802, 2021.
[10] B. Zachariah, P. O. Odion, and R. I. Saidu, “ Roundabouts
Modeling and Vehicular T raffic Control T echniques : A Survey,”
IUP J. Inf. Technol., vol. XVI, no. 1, pp. 1–17, 2020.

Authorized licensed use limited to: Higher College of Technology. Downloaded on March 19,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.