SMART SIGNAL TIMING FOR URBAN

INTERSECTIONS USING REAL-TIME VEHICLE

DETECTION AND PCU CALCULATIONS

Project Report Submitted in Partial Fulfillment of Academic Requirement for the

Award of Degree of

BACHELOR OF TECHNOLOGY
IN
CIVIL ENGINEERING

Submitted By
NIRAJ KUMAR (21024117)
NITIN ANAND (21024119)
RAHUL KUMAR VISHVAKARMA (21024124)

Under The Guidance of

DR. UMANK MISHRA
Associate professor

DEPARTMENT OF CIVIL ENGINEERING

SCHOOL OF STUDIES OF ENGINEERING AND TECHNOLOGY,
GURU GHASIDAS VISHWAVIDYALAYA, BILASPUR (C.G.)
(ACentralUniversityEstablishedbytheCentralUniversityAct2009No.25of2009)

2024 - 2025
 DEPARTMENT OF CIVIL ENGINEERING
SCHOOL OF STUDIES OF ENGINEERING AND
TECHNOLOGY,
GURU GHASIDAS VISWAVIDYALAYA, BILASPUR (C.G.)
(ACentralUniversityEstablishedbytheCentralUniversityAct2009No.25of2009)

DECLARATION

I, the undersigned, solemnly declare that this project work, entitled

“SMART SIGNAL TIMING FOR URBAN INTERSECTIONS USING REAL-
TIME VEHICLE DETECTION AND PCU CALCULATIONS”
is my own work, carried out during the course of my study in the 8th Semester
under the guidance of DR. UMANK MISHRA, Associate Professor, Department
of Civil Engineering, School of Studies in Engineering and Technology, Guru
Ghasidas Vishwavidyalaya. I assert that the statements made and conclusions
drawn are the outcome of my work. I further declare that to the best of my
knowledge and belief, this report does not contain any part of any work which has
been submitted for the award of any degree in this or any other university.

Sign……………………. Sign…………………….
NIRAJ KUMAR NITIN ANAND
Roll no. 21024117 Roll no. 21024119
Enrolment (GGV/21/01017) Enrolment (GGV/21/01019)

Sign…………………….
RAHUL KUMAR VISHVAKARMA
Roll no. 21024124
Enrolment (GGV/21/01024)

ii
 DEPARTMENT OF CIVIL ENGINEERING
SCHOOL OF STUDIES OF ENGINEERING AND
TECHNOLOGY,
GURU GHASIDAS VISWAVIDYALAYA, BILASPUR (C.G.)
(ACentralUniversityEstablishedbytheCentralUniversityAct2009No.25of2009)

CERTIFICATE

Certified that the project report entitled “Smart Signal Timing for Urban
Intersections Using Real-Time Vehicle Detection and PCU Calculations”
submitted by NIRAJ KUMAR, NITIN ANAND, RAHUL KUMAR
VISHVAKARMA in partial fulfillment of the requirements for the award of the
degree of Bachelor of Technology in Civil Engineering, Department of Civil
Engineering, School of Studies in Engineering and Technology, Guru Ghasidas
Vishwavidyalaya, Bilaspur, is accorded to the student’s own work, carried out by
him in the Department of Civil Engineering during the session 2024–2025 under
my supervision and guidance.

Sign. ............................ Sign. ......................

DR. UMANK MISHRA EXTERNAL EXAMINER 1
ASSOCIATE PROFESSOR
& GUIDE
Sign. ......................
EXTERNAL EXAMINER 2

Sign. ...........................
Prof. M.Chakradhara Rao
Head of Department
Civil Engineering Department,
School of Studies of Engineering and Technology
Guru GhasidasVishwavidyalaya
Bilaspur, Chhattisgarh

iii
 AKNOWLEDGEMENT

I take this opportunity to express my sincere gratitude to all those who have
helped me throughout the completion of this project titled
“Smart Signal Timing for Urban Intersections Using Real-Time Vehicle Detection
and PCU Calculations.”
First and foremost, I would like to express my deep sense of gratitude to Dr.
Umank Mishra, Associate Professor, Department of Civil Engineering, for his
invaluable guidance, continuous encouragement, and constant support
throughout the course of this project. His expertise and timely suggestions played
a crucial role in shaping the project to its present form.
I would also like to thank the Department of Civil Engineering, School of Studies
in Engineering and Technology, Guru Ghasidas Vishwavidyalaya, for providing the
necessary infrastructure and academic environment to carry out this work.
My heartfelt thanks to all faculty members and staff of the department for their
encouragement and assistance. I also extend my gratitude to my fellow classmates
and friends for their constructive feedback and moral support.
Last but not the least, I am thankful to my family for their unwavering support
and motivation which kept me focused and determined during every phase of the
project.

iv
 ABSTRACT

Managing traffic in busy city areas has become one of the biggest challenges due
to the rapid increase in the number of vehicles. Traditional traffic signals that
work on fixed time cycles often fail to adapt to real-time traffic conditions,
leading to unnecessary delays and congestion. This project, titled
“Smart Signal Timing for Urban Intersections Using Real-Time Vehicle Detection
and PCU Calculations,”
aims to provide a more responsive and practical solution to this problem.
The project is divided into two main parts. In the first part, we used video-based
vehicle detection techniques using YOLOv5 and the SORT tracking algorithm to
identify and count different types of vehicles passing through each lane. These
vehicles were then converted into standard traffic load units using Passenger Car
Units (PCU) to make the data more useful for traffic analysis.
In the second part, we designed a signal timing model that takes these PCU values
as input and calculates the green signal time for each lane dynamically. The
algorithm ensures that every lane gets a minimum signal time and that the total
cycle time adjusts based on overall traffic volume. This method not only improves
traffic flow but also keeps the system fair and efficient.
This work is a step toward smarter, more adaptive traffic management systems
and can be further developed to work with live traffic feeds in real-time
environments.

v
 TABLE OF CONTENTS

TITLE I

DECLARATION II

CERTIFICATE III

ACKNOWLEDGEMENT IV

ABSTRACT V

LIST OF FIGURES IX

LIST OF TABLES X

CHAPTER 1 - INTRODUCTION 11

1.1 Traffic Signal Management Issues 11

1.2 Need For Dynamic Signal Timing Based on

12
Traffic Volume

1.3 Role Of PCU in Traffic engineering 12

1.4 Application Of AI For Vehicle Detection

13
And Classification

1.5 Dynamic Green Time Allocation Logic 14

1.6 Objective Of The Study 15

CHAPTER 2 - REVIEW OF THE

16
LITERATURE

2.1 General Overview 16

VI
2.2 Literature Review 16

2.3 Critical Observation 19

2.4 Research Gap 20

2.5 Scope Of The Work 21

CHAPTER 3 - METHODOLOGY 23

3.1 Overview 23

3.2 Video Acquisition and Preprocessing 23

3.3 Vehicle Detection and Classification Using

24
YOLOv5

3.4 Role Of PCU in Traffic Signal Optimization 25

3.5 Dynamic Green Time Allocation Logic 25

3.6 Data Collection 26

3.7 System Integration and Simulation 31

3.8 Validation and Observation 32

CHAPTER 4 - RESULT AND DISCUSSION 34

VII
4.1 Introduction To Results 34

4.2 PCU Throughput Comparison 34

4.3 Green Time Efficiency 35

4.4 Survey-Based Validation 35

4.5 Real Time Responsiveness 36

4.6 Limitations 37

4.7 Discussion Summary 38

CHAPTER 5 - CONCLUSION 39

CHAPTER 6 - REFERENCES 40

VIII
 LIST OF FIGURES

Fig. No. Description Of Figure Page No.

Comparison Between Conventional Traffic Signal

1 33
vs Dynamic Traffic Signal

IX
 LIST OF TABLES

Table
Description Of Table Page No.
No.

1 PCU Values as per IRC:106-1990 13

2 Survey Report 27

3 Survey Report 27

4 Survey Report 28

5 Average of All Three Surveys 28

Total no. of PCU Crossed During Dynamic

6 30
Green Signal Time

Comparison Between no. of PCUs Passed During

7 31
Green Time

Final Result Showing Overall Improvement in

8 33
Traffic Flow

X
 Chapter - 1
Introduction

Traffic congestion is one of the most pressing challenges faced by urban areas
across the globe. As cities expand and vehicle ownership continues to rise, the
existing traffic infrastructure, especially at intersections, struggles to keep up. In
many Indian cities, signal systems are still based on fixed cycles that operate
irrespective of the real-time traffic load. This leads to inefficient road usage,
unnecessary delays, increased fuel consumption, and avoidable air pollution. Even
during low traffic hours, vehicles often have to wait unnecessarily at red signals,
while high-traffic lanes suffer from insufficient green time.
To solve this problem, traffic management systems must evolve to become smarter
and more responsive. This project—"Smart Signal Timing for Urban Intersections
Using Real-Time Vehicle Detection and PCU Calculations"—proposes a hybrid
solution that blends artificial intelligence and traffic engineering principles. The
primary goal is to optimize signal timings dynamically, depending on the actual
number and type of vehicles approaching an intersection.
The system works in two phases. In the first phase, a real-time video feed from a
traffic camera is analyzed using a deep learning model (YOLOv5), which detects
and classifies each vehicle. The detected vehicles are then converted into their PCU
(Passenger Car Unit) values—a method widely used in traffic engineering to
quantify the space and impact of different vehicle types. For instance, a truck
impacts traffic differently than a motorcycle, and PCU values help standardize
this.
In the second phase, based on the total PCU per lane, the algorithm calculates
dynamic green times. A fixed minimum green time is ensured for each lane to
maintain fairness, and the remaining available cycle time is distributed
proportionally according to the detected traffic load. Furthermore, if the overall
traffic is light, the system intelligently reduces the total cycle time, avoiding
unnecessary delays.
This approach not only brings fairness and efficiency to traffic flow but also lays
the groundwork for future smart city integration. It can be further enhanced with
emergency vehicle detection, automatic input from surveillance systems, and
integration into urban traffic control centers.

1.1 Traffic Signal Management Issues

In India and many developing countries, traffic signal systems typically operate on
fixed-time cycles, regardless of actual traffic flow. This outdated method results in
unnecessary wait times, longer fuel consumption, and increased emissions. Roads

11
that are congested often receive the same green signal time as those with minimal
traffic, leading to inefficient road utilization. Moreover, emergency situations or
unexpected traffic surges cannot be accommodated dynamically. These issues
collectively underline the urgent need for a smarter, data-driven traffic
management approach that adjusts itself based on real-world vehicle flow.

1.2 Need for Dynamic Signal Timing Based on Traffic Volume

Dynamic traffic signals provide a solution to the shortcomings of fixed-time

systems by adapting green and red signal durations according to real-time traffic
volumes. When signal timings reflect actual vehicle loads, roads clear faster and
smoother. This not only reduces commuter frustration but also helps improve fuel
efficiency and air quality. In this project, the need is addressed using a PCU
(Passenger Car Unit)-based method, which considers the type and number of
vehicles on each lane, giving proportionate green time. It ensures that no road is
unfairly prioritized while maintaining a logical flow of traffic.

1.3 Role of PCU (Passenger Car Unit) in Traffic Engineering

In the diverse and often congested traffic environments found in Indian cities,
simply counting the number of vehicles on a road isn’t enough to understand their
impact on traffic flow. Different types of vehicles—like bikes, buses, cars, and
trucks—occupy different amounts of space, move at different speeds, and behave
differently in traffic. This is where the concept of Passenger Car Unit (PCU)
becomes essential.
The PCU is a standard measure used to equate the impact of various types of
vehicles to that of a standard passenger car. This helps in designing traffic systems
that are fair and efficient by considering not just the number of vehicles but how
much space and time each type consumes on the road. For example, a truck
occupies more road space and moves slower than a car, so it contributes more to
congestion and thus has a higher PCU value.
The Indian Roads Congress (IRC:106-1990) provides recommended PCU values
for various vehicle types under mixed traffic conditions. These values are shown in
the table below:

12
 Vehicle Type PCU Value

Passenger Car 1

Motorcycle / Scooter 0.5

Auto-Rickshaw 1.2

Auto-Rickshaw 3

Truck 3

Light Commercial Vehicle 1.5

Bicycle 0.5

Tractor 4

Table : 1 PCU Value as per IRC:106-1990

In this project, these PCU values are used to convert raw vehicle counts—obtained
from video-based real-time detection—into a standardized traffic load. This allows
for better decision-making while designing green signal timings, ensuring that
larger and slower vehicles are given the appropriate amount of time to clear
intersections safely. Ultimately, using PCU-based calculations helps improve traffic
efficiency and reduce unnecessary delays.

1.4 Application of AI for Vehicle Detection and Classification

The rapid urbanization of cities has led to unpredictable traffic flows, making
manual monitoring and static traffic signal systems insufficient. To address this,
Artificial Intelligence (AI) is being increasingly used to automate the detection,
tracking, and classification of vehicles. In this project, we utilized deep learning-
based object detection models and tracking algorithms to enable real-time traffic
monitoring from video footage. The goal was to build a system that could
accurately identify different vehicle types and help in dynamic signal timing design
using PCU-based calculations.

13
Below are the key components of AI technologies applied in our project:

1.4.1 YOLO (You Only Look Once) Object Detection Algorithm

YOLO is one of the most popular real-time object detection models. Unlike older
methods that required separate stages for region proposal and classification,
YOLO does everything in a single neural network pass. This makes it extremely
fast and efficient—ideal for traffic applications where decisions need to be made in
real time.
In our project, we used YOLOv5s, a lightweight version of the YOLOv5 model. It
was pre-trained on the COCO dataset and capable of detecting 80 object classes,
including vehicles such as cars, buses, motorcycles, and trucks. We fine-tuned it for
our needs by filtering only vehicle classes relevant to Indian roads.
Advantages of using YOLO in our project:
Real-time speed with good accuracy
Single-shot detection: bounding box and class prediction done together
Well-documented and open-source, with PyTorch support

1.4.2 SORT (Simple Online and Realtime Tracking)

While YOLO detects objects frame-by-frame, it does not remember which vehicle
is which over time. This is where SORT comes into play. SORT is a fast and simple
tracking algorithm that links detections across video frames to assign unique IDs
to each vehicle.
We used SORT to:
Track the movement of each vehicle throughout the video
Avoid double-counting the same vehicle in multiple frames
Map vehicle types to unique IDs for PCU conversion

1.5 Dynamic Green Time Allocation Logic

Modern cities experience constant vehicular congestion, especially at intersections.

To handle this growing pressure, traditional fixed-time traffic signals often fall
short. Our project introduces a smarter alternative—dynamic green time allocation
—which adapts signal timings based on actual vehicle presence. By using real-time
vehicle detection and calculating Passenger Car Units (PCU), we assign green
signal durations proportionally, ensuring smoother flow and reduced wait times.
Instead of offering the same green time to every lane regardless of traffic density,
our logic distributes available cycle time dynamically. A fixed minimum time is
allotted to each lane to prevent starvation, while the remaining time is distributed
based on the share of vehicles after threshold adjustment.

14
1.5.1 Fixed vs. Dynamic Timing

In fixed timing systems, each signal gets equal or pre-defined time regardless of
traffic load. In contrast, dynamic systems assess live input (like PCU) to assign
time based on demand. This increases efficiency and reduces idle time at
intersections.

1.5.2 Threshold and Minimum Allocation Concept

We assume that in every cycle, at least 10 PCUs from each lane will clear during a
base green time (e.g., 10 seconds). This base time is reserved, and only the
remaining time is distributed based on the traffic proportion from each lane. This
ensures fairness and avoids extremely short durations.

1.5.3 Cycle Time Adaptation:

Total signal cycle time isn’t static. If total detected PCUs are under 100, a shorter
cycle (e.g., 80 seconds) is used. If it exceeds, we go with 120 seconds. This
flexibility avoids unnecessary delays in low traffic and handles high traffic
efficiently.

1.5.4 PCU-Based Proportional Allocation:

Once the threshold-adjusted PCUs are calculated, we derive the ratio of each
lane's demand to the total and distribute remaining seconds accordingly. All
results are rounded off to whole seconds for practicality.

1.6 Objective of the Study

The main objective of this project, titled "Smart Signal Timing for Urban
Intersections Using Real-Time Vehicle Detection and PCU Calculations", is to
design an intelligent traffic signal management system that adapts to real-time
traffic conditions. The system aims to detect and classify vehicles using artificial
intelligence and compute Passenger Car Units (PCUs) to reflect actual traffic
density at intersections. Based on this data, the goal is to dynamically allocate
green signal time to each lane in a fair and optimized manner, ensuring smooth
vehicle movement, minimizing idle time, and reducing congestion. The study also
intends to make this system scalable for future integration, where video input can
automatically drive signal logic, enhancing traffic control efficiency, especially in
densely populated urban areas.

15
 Chapter - 2
Literature Review
2.1 General Overview

To develop this system effectively, we reviewed various approaches used globally

and locally for traffic control — including fixed-time models, sensor-based
actuated systems, and intelligent systems using AI and computer vision. We
specifically focused on how traffic density can be evaluated through vehicle
classification and how PCU (Passenger Car Unit) values can guide signal timing.
Several studies have utilized AI models like YOLO (You Only Look Once) for
object detection in traffic scenes, with promising results in vehicle classification.
However, very few have connected this detection output to actual traffic signal
design using PCU-based dynamic logic — especially tailored for Indian traffic
diversity, where auto-rickshaws, bikes, and buses all interact differently with the
road.
Hence, our project bridges this gap by integrating AI-based vehicle detection (via
YOLOv5) with a PCU-calculated dynamic green time logic. The approach not
only provides a more responsive signal timing system but also holds potential for
future integration with real-time surveillance systems and urban traffic
management platforms.

2.2 Literature Review

Christofel Rio Goenawan and Haar-Dong Soo (2024)

The literature review explores the integration of AI in smart traffic management

systems, highlighting how Convolutional Neural Networks (CNNs) and Recurrent
Neural Networks with LSTM can optimize vehicle detection and traffic
prediction. It emphasizes the evolution of AI, its application in computer vision
for object detection, and the use of predictive models for congestion forecasting.
Smart systems, evaluated using CARLA simulation, demonstrate significant
improvements in traffic flow and vehicle delay, showcasing AI’s potential in
enhancing urban mobility infrastructure.

Sangeetha R.G et al. (2024)

Previous research on traffic management explored sensor-based and vision-based

systems, but many lacked accuracy or practicality, especially in chaotic urban
settings like India. IR, acoustic, and RFID sensors faced limitations in range and
real-time responsiveness. Vision approaches improved detection but were
computationally heavy.

16
 Some IoT and machine learning models predicted traffic flow but did not
prioritize emergency vehicles effectively. Existing systems often assumed ideal
conditions, like lane discipline or widespread onboard units. This study stands out
by combining KNN-based traffic density estimation and YOLO-based emergency
vehicle detection, offering a dynamic, real-time solution suitable for smart cities
with high traffic congestion.

Nitin Sakhare et al. (2024)

Recent studies in intelligent traffic management focus on leveraging IoT, image

processing, and machine learning to address congestion and inefficiencies in
traditional systems. Techniques such as YOLO-based vehicle detection, adaptive
signal control, and Raspberry Pi integration are commonly employed. These
systems prioritize emergency vehicles, dynamically adjust signal timings, and
enhance traffic flow through real-time data analysis. This paper builds upon such
approaches by combining lane-specific vehicle detection and adaptive control to
reduce congestion and environmental impact effectively.

Mahalinga V. Mandi et al. (2023)

Traditional traffic systems relying on manual or fixed-timing signals are

inadequate for growing urban traffic needs. Prior research highlights the use of
IoT, sensors, and automation to enhance real-time traffic control. Existing systems
often fail to dynamically adjust signals or prioritize emergency vehicles. Recent
advancements integrate data from sensors and video feeds with adaptive
algorithms to optimize traffic flow. This paper builds upon such approaches,
proposing a real-time, density-based system using IoT for smarter urban traffic
management.

Bhuvan S T et al. (2022)

Recent studies on smart traffic management emphasize the integration of IoT, AI,
and RFID to address congestion and optimize urban mobility. Traditional traffic
systems based on fixed timings are inefficient in dense urban areas. Advanced
models use neural networks, video processing, and real-time sensors to estimate
traffic flow and adapt signal timings dynamically. Some works also include
emergency vehicle prioritization and environmental monitoring. These approaches
show potential in reducing congestion, improving safety, and enabling data-driven
urban planning.

17
Mohammed sarrab et al. (2020)

Several studies have explored IoT and AI for traffic management, mainly focusing
on highways and urban roads. Traditional systems rely heavily on smartphones
and vehicle sensors, limiting accessibility. Recent works have used ultrasonic,
magnetic, and video sensors for real-time traffic updates. However, limited
attention has been given to collector roads and non-smart environments. This
study addresses the gap by proposing an IoT-based system using magnetic sensors
and roadside message units to deliver real-time traffic information without
requiring user devices.

Manoj G et al. (2018)

Several studies have addressed the limitations of fixed-time traffic systems using
technologies like inductive loops, ultrasonic, infrared, and acoustic sensors. Recent
advancements include image processing, RF detectors, fuzzy logic, IoT, and cloud-
based systems for dynamic traffic control. However, many rely on a single
technology. This study integrates ultrasonic sensors and image processing with
cloud storage via Raspberry Pi, offering real-time, reliable traffic density
estimation. The hybrid approach enhances accuracy and introduces fault tolerance
for sensor failures.

Mishra et al (2017)

Numerous studies have addressed the estimation of Passenger Car Unit (PCU)
values under heterogeneous traffic conditions using parameters such as speed,
density, and delay. However, these approaches often yield inconsistent results due
to varying roadway and traffic compositions. Static PCU values recommended by
the Indian Roads Congress (IRC:106-1990) are based on limited empirical data
and may not reflect real-world mixed-traffic dynamics. Microsimulation models
like VISSIM and HeteroSim have been employed to derive dynamic PCU values,
but speed-based methods often struggle under non-lane-disciplined traffic. In this
context, the concept of area occupancy, introduced by Mallikarjuna and Rao,
offers a more reliable and field-measurable parameter. It accounts for the actual
space occupied by vehicles, making it a promising basis for improved PCU
estimation.

18
2.3 Critical Observation

2.3.1Technological Integration

Most studies integrate AI models like Yolo for object recognition and LSTM for
traffic prediction. These systems demonstrate the possibility that AI can
adaptively manage traffic. However, integration into infrastructure remains
limited, and response to live real-time data needs to be further refined for robust
urban use.

2.3.2. IoT and Sensor Utilization

IoT-based transport systems are based on sensors such as Raspberry PI, IR, and
RFID, and collect data such as vehicle number and type. These allow for dynamic
signal control and emergency vehicle prioritization. Nevertheless, their cover is
often limited to intersections, and syncing in urban networks is still
underdeveloped.

2.3.3. Environmental Systems

Many studies consider environmental benefits, such as reduced emissions and fuel
consumption due to optimized signaling and reduced idle times. Minimizing
unnecessary outages and delays contributes to sustainability. However, detailed
indicators to reduce pollution are rarely quantified, and long-term reviews of
environmental impacts are largely lacking.

2.3.4. Simulations for Real Tests

Simulation environments such as Carla provide useful tests for intelligent traffic
systems. However, it cannot replicate actual complexity such as weather changes,
driver behavior, signal failures, and more. The real attempts to offer you are rare,
making it difficult to assess how the system works in a real city under traffic
conditions mixed with chaotic.

2.3.5. Issues for urban centers in India

Papers like Mishra et al. Discuss unstructured traffic in India and attach a PCU
estimation model based on surface loads. These are more suitable for Indian roads
where lane truck discipline is not available. However, most AI models used
worldwide take on structured truck behavior that limits their effectiveness in
Indian or similarly chaotic urban environments.

19
2.3.6. Data Collection Gap

Many systems rely on static data records or APIs from third party providers, such
as Google Maps. This limits adaptability to dynamic conditions. Real-time multi-
source data integration is limited (for example, combining video, sensors, and V2X
inputs). Without accurate and comprehensive data, sophisticated algorithms
cannot consistently make optimal traffic decisions.

2.4 Research Gap

2.4.1 Lack of uniform architecture

Existing frameworks need a standardized, secluded engineering that consistently

coordinating AI calculations, IoT-based detecting, and control components. Most
models are custom-built, making them difficult to scale or imitate. A bound
together system would permit plug-and-play components, simpler overhauls, and
interoperability between innovations, which is basic for real-world arrangement
over changing urban frameworks.

2.4.2. Restricted Real-world Deployment

While simulation-based results are promising, few systems have undergone large-
scale, real-world validation. Challenges like sensor calibration, weather variability,
network latency, and infrastructure limitations remain untested. Without
deployment in diverse traffic conditions, including rural and highly congested
urban zones, it’s difficult to measure true effectiveness, reliability, and user
acceptance of these solutions.

2.4.3. Emergency Vehicle Handling

Although some systems prioritize ambulances or fire trucks using RFID or object
detection, real-time dynamic rerouting and signal coordination for emergency
vehicles are still rudimentary. Systems often fail to simulate complex city scenarios
where multiple emergency vehicles must be prioritized simultaneously, especially
during peak traffic or in multi-intersection zones.

2.4.4. Neglect of Non-motorized Traffic

Most models are designed for motorized vehicles and overlook bicycles,
pedestrians, and informal transport (e.g., rickshaws). This omission limits accuracy
and inclusiveness, particularly in countries with mixed-traffic environments.

20
Systems must evolve to detect and predict interactions across all road users to
improve safety, signal accuracy, and fair space allocation.

2.4.5. Scalability and cost-effective analysis

Few studies address the cost of scaling traffic systems across entire cities or
regions. Resource-intensive models requiring high-end computing or expensive
sensors limit affordability. A research gap exists in developing lightweight, energy-
efficient, and low-cost systems that can be feasibly adopted by small municipalities
or developing urban centers with constrained budgets.

2.4.6. Insufficient Real-time Data Fusion

Traffic control relies on accurate, up-to-date information, yet current systems

struggle to merge inputs from various sources—like sensors, cameras, mobile GPS,
and V2X communication—into a cohesive model. Delays in data transmission or
poor data quality reduce the system’s responsiveness. Real-time fusion methods are
needed to ensure optimal and timely decision-making.

2.5 Scope Of The Work

2.5.1. Development of Hybrid AI-IoT-V2X Systems

There is significant potential in developing integrated systems combining Artificial

Intelligence (for decision-making), IoT (for sensing and control), and V2X (for
communication). This integration would allow vehicles to interact with
infrastructure in real-time, enabling highly adaptive traffic control, efficient
emergency routing, and cooperative behavior between autonomous vehicles and
traffic signals, enhancing both safety and flow.

2.5.2 Field Trials and Public Policy Integration

To transition from prototype to practical use, future research should focus on

piloting smart traffic systems in diverse urban settings. Simultaneously,
collaboration with government agencies is crucial to integrate such systems with
existing transport policies, standards, and regulations. These trials would also offer
insight into public response, legal concerns, and long-term sustainability.

2.5.3. Incorporation of Multimodal Traffic Behavior

Future systems should account for all traffic participants—pedestrians, cyclists,

public transport, and informal vehicles. Including multimodal behavior in traffic
models will lead to more inclusive and efficient urban traffic management.

21
This will require object detection algorithms that can classify different road users
and adaptive signal strategies that prioritize safety and equity.

2.5.4. Design of Low-cost, Scalable Infrastructure

Developing countries require solutions that are both technically effective and
economically feasible. The scope includes designing energy-efficient, low-cost
sensor nodes, open-source platforms, and edge computing solutions that minimize
dependency on expensive cloud services. This would enable wider adoption of
smart traffic management in small towns and resource-constrained municipalities.

2.5.5. Predictive and Preventive Traffic Control Models

Current models react to congestion; future work should focus on predictive

analytics using AI to foresee traffic buildup before it occurs. Combined with
preventive strategies like rerouting or adjusting signal timings in advance, these
systems can significantly reduce congestion and emissions, improving the overall
efficiency of the transportation network.

2.5.6. Integration of Real-time Data Fusion Systems

There is scope to create robust real-time data fusion frameworks that merge data
from sensors, GPS, cameras, social feeds, and vehicle telemetry. Such systems
would offer comprehensive situational awareness, allowing traffic controllers to
make accurate and timely decisions. Standardizing protocols and ensuring data
quality across sources will be critical for reliability.

22
 Chapter - 3
Methodology
3.1 Overview

The project “Smart Signal Timing for Urban Intersections Using Real-Time
Vehicle Detection and PCU Calculations” adopts a practical and modular strategy
to improve traffic flow at signalized intersections. The approach combines modern
artificial intelligence (AI) techniques with established traffic engineering concepts,
aiming to create a smart system that can adapt to real-time road conditions
without requiring extensive hardware infrastructure.
At the core, the methodology is divided into two main stages. The first stage
involves detecting and classifying vehicles in a video feed using a deep learning
model called YOLOv5. These vehicles are tracked across frames with the help of a
lightweight and efficient algorithm known as SORT (Simple Online and Realtime
Tracking). Based on the class of each vehicle — such as two-wheeler, car, bus, or
truck — a standard PCU (Passenger Car Unit) value is assigned, and the total
PCU count for each lane is calculated over a fixed duration.
The second stage focuses on dynamically designing green signal durations based
on these PCU inputs. A logical formula is applied that ensures a base minimum
time is allocated to every lane, ensuring fairness. The remaining time in the traffic
signal cycle is distributed in proportion to the PCU load across lanes. To make the
system adaptive, the total cycle time is also varied based on the overall traffic
volume — shorter cycles for low traffic, and extended ones when congestion is
high.
This step-by-step logic forms the foundation of a system that doesn't just operate
on fixed timing but adapts intelligently, which is crucial for densely populated
urban settings in countries like India. The proposed solution also considers future
scalability, where real-time camera input can be fed directly into the system,
allowing for fully automated signal control.

3.2 Video Acquisition and Preprocessing

To design a traffic signal system that truly reflects on-ground realities, it was
important to base the model on realistic traffic scenarios. For this reason, we used
a combination of self-recorded videos from nearby intersections and publicly
available traffic footage relevant to Indian road conditions. Videos showing foreign
traffic systems were specifically avoided, as lane discipline, vehicle types, and road
behaviors in those setups differ significantly from those seen in India. Using
Indian-context videos made the detection and logic more generalized and relatable
to real urban traffic in our country.

23
After video collection, preprocessing became essential to make the raw footage
compatible with the AI detection model. Each video was standardized in terms of
frame size and resolution, typically resized to either 640×480 or 1280×720 pixels.
This step ensured a good balance between image quality and computational
efficiency. Frame rates were also adjusted to allow near-real-time detection
performance.
To enhance accuracy, Regions of Interest (ROI) were manually defined in the
video frames. This helped focus the AI model only on areas where vehicle
movement occurred, filtering out irrelevant parts like sidewalks, buildings, or the
sky. Additionally, basic brightness and contrast adjustments were performed when
lighting inconsistencies or shadows affected visibility.
This careful curation and preprocessing of the video data ensured that the vehicle
detection and classification model (YOLOv5) received consistent, India-specific,
and clean inputs — laying a solid foundation for accurate PCU estimation and
signal timing in later stages.

3.3 Vehicle Detection and Classification Using YOLOv5

Accurate vehicle detection is the backbone of any intelligent traffic management

system. In our project, we implemented YOLOv5 (You Only Look Once version 5),
a widely adopted deep learning model known for its real-time object detection
capabilities. YOLOv5 stands out for its speed and precision, making it ideal for
dynamic environments like busy intersections.
The YOLOv5 model was trained and tested using traffic videos containing diverse
vehicle types commonly found in Indian urban areas—such as motorcycles, auto-
rickshaws, cars, buses, and trucks. Unlike rule-based image processing methods,
YOLOv5 learns patterns through thousands of image samples, allowing it to
detect vehicles even in partially occluded or poorly lit scenes.
For our application, we used the pre-trained COCO model as the base and
customized it to focus on the classes relevant to traffic. The model was integrated
with a frame-by-frame video processing pipeline. Each detected vehicle was given a
bounding box, label (vehicle class), and confidence score, which served as inputs
for further tracking and classification.
To ensure the system could identify whether a vehicle had already passed or not,
we implemented a lightweight object tracking method called SORT (Simple Online
and Realtime Tracking). This enabled consistent counting without duplication,
even as vehicles moved across frames.
What makes this setup more robust is that it doesn’t just count vehicles—it
classifies them into categories. This classification is essential for our Passenger Car
Unit (PCU) calculations, as different vehicle types have different impacts on traffic
flow. For instance, a truck occupies more space and time than a bike and hence
contributes a higher PCU value.

24
By combining YOLOv5's fast detection with SORT’s tracking, our system achieves
reliable, real-time identification and classification of traffic—setting the stage for
intelligent signal timing based on actual road conditions.

3.4 Role of PCU in Traffic Signal Optimization

In traffic engineering, particularly within complex and heterogeneous systems like

those seen on Indian roads, the concept of Passenger Car Units (PCU) plays a
pivotal role. Rather than simply counting vehicles, PCU allows for the
standardization of different vehicle types based on the space they occupy and the
time they take to cross intersections. This ensures that the signal timings are
determined by actual road usage demand, not just vehicle count.
In our project, we utilized PCU to convert the count of detected vehicles into a
weighted total, reflecting the true traffic load. For instance, a truck occupies
significantly more road space than a motorcycle; thus, treating both equally would
distort traffic estimates and result in inefficient signal timing. By applying PCU
values provided by the Indian Roads Congress (IRC), we accounted for this
variability effectively.
The standard PCU values adopted in our system are summarized in Table no. 1,
which includes common vehicle categories such as two-wheelers, passenger cars,
buses, trucks, and non-motorized vehicles. These values are integral to computing
the effective traffic volume per lane.
After vehicle detection and classification using AI, we assign each class its
corresponding PCU weight and sum them per lane. This total PCU count per lane
then serves as the foundation for our dynamic green signal time calculation. This
approach ensures that signals adapt in real-time to actual traffic demand, rather
than relying on fixed or arbitrary durations.
Overall, incorporating PCU into our model enhances fairness, efficiency, and
adaptability, making our system suitable for real-world urban traffic intersections
with mixed traffic flow.

3.5 Dynamic Green Time Allocation Logic

An essential part of this project is the method used for dynamically allocating
green signal times based on real-time traffic load at an intersection. Unlike
conventional fixed-time signals, which assign equal or preset green durations
regardless of actual traffic, our system adjusts green times proportionally using
vehicle counts (converted to PCU) as input. This approach helps in minimizing
delays and improving traffic flow across all lanes.

25
3.5.1 Minimum Green Time Allocation

To ensure that no lane is neglected, each approach is first given a base minimum
green time. In our model, we allocate 10 seconds to each of the four directions,
accounting for 40 seconds of the total cycle time. This ensures that even lanes with
very few vehicles still get an opportunity to clear.

3.5.2 Dynamic Allocation Using PCU Ratios

Once the minimum time is allocated, the remaining green time (i.e., 40 or 80
seconds depending on total cycle time) is distributed based on the remaining PCU.
We subtract 10 PCU per lane (assuming approx. 1 PCU clears per second during
the minimum time) from each lane’s total. Then, the remaining PCUs are used to
calculate the proportional ratio of traffic left to be cleared. This remaining time is
then divided among the lanes based on their respective PCU ratios, ensuring that
lanes with higher vehicle pressure get longer green time.

3.5.3 Dual Cycle Time Logic

To make the system adaptable to overall traffic load, we used two fixed cycle time
settings:
80 seconds, when total PCU detected at the intersection is ≤100
120 seconds, when total PCU is >100
This dual-cycle system was implemented to ensure simplicity in design and
analysis. However, the same proportional time allocation logic could easily be
extended to a fully dynamic cycle time model, where the total cycle time changes
continuously based on real-time PCU load. This flexibility makes the system
scalable and suitable for future upgrades.

3.6 Data Collection

3.6.1 Traffic Survey and Comparative Analysis

Three traffic surveys were conducted at Satyam Chowk, Bilaspur (Chhattisgarh) to

assess the performance of the proposed PCU-based dynamic traffic signal system
under varying traffic conditions. The first survey was conducted on Sunday at 8:00
AM, representing a low-traffic scenario. The second survey took place on Monday
at 12:00 PM, reflecting moderate traffic, and the third survey was conducted on
Monday at 6:00 PM, capturing peak-hour conditions. Vehicle counts from all three
surveys were categorized by type and converted into Passenger Car Units (PCUs)
using standard values as listed in Table 1. This conversion enabled a more accurate
comparison of traffic volume and lane-wise demand across different time periods.

26
 Green
PCU Crossed
Lane Signal
(Available) (PCU)
Time (sec)

A 38 30 30

B 14 30 14

C 12 30 12

D 35 30 28

Total 101 120 84

Table 2. Survey on Sunday at 8:00 AM, representing a low-traffic

scenario

Green
PCU Crossed
Lane Signal
(Available) (PCU)
Time

A 43 30 28

B 20 30 20

C 23 30 23

D 57 30 29

Total 133 120 100

Table 3. Survey on Monday at 12:00 PM, reflecting moderate traffic

27
 Green
PCU Crossed
Lane Signal
(Available) (PCU)
Time

A 84 30 30

B 35 30 28

C 26 30 26

D 102 30 29

Total 247 120 113

Table 4. Survey on Monday at 06:00 PM, reflecting heavy traffic

Green
PCU Crossed
Lane Signal
(Available) (PCU)
Time

A 55 30 30

B 23 30 21

C 21 30 21

D 65 30 29

Total 164 120 101

Table 5. Average of all three survey ( Rounded off to upper limit)

28
3.6.2 Calculation for Green Signal Time

Lane A: 55
Lane B: 23
Lane C: 21
Lane D: 65

Step 1

Total PCU = 55 + 23 + 21 + 65 = 16

Step 2: Cycle Time Selection

Total PCU > 100 → Cycle_time = 120

Step 3: Minimum Green Time

Min green time = 10 → total min = 4 × 10 = 40

Remaining time = 120 - 40 = 80

Step 4: Deduct 10 crossing PCU from each lane

Lane Original PCU Deducted Remaining PCU

Lane A 55 10 45

Lane B 23 10 13

Lane C 21 10 11

Lane D 65 10 55

Step 5: Total Remaining PCU

Total remaining PCU = 45 + 13 + 11 + 55 = 124

29
Step 6: Dynamic Green Time Calculation

Lane Remaining PCU Green Time Formula Green Time Formula

A 45 10 + (45 / 124) × 80 = 39.03 s

B 13 10 + (13 / 124) × 80 = 18.38 s

C 11 10 + (11 / 124) × 80 = 1 17.10 s

D 55 10 + (55 / 124) × 80 = 45.48 s

Final Output (Rounded to Nearest Second)

Signal Timing Plan (Cycle Time: 120 sec)

Lane A: 55 PCU → 39 sec
Lane B: 23 PCU → 18 sec
Lane C: 21 PCU → 17 sec
Lane D: 65 PCU → 45 sec

Green
PCU Crossed
Lane Signal
(Available) (PCU)
Time

A 55 39 39

B 23 18 18

C 21 17 17

D 65 45 45

119
Total 164 (Approx 119
120 sec.)

Table 6. Total no. Of PCU Crossed During Dynamic Green Signal Time

30
3.6.2 PCU-Based Throughput Comparison

Model Type PCUs Passed During Green Time

Traditional Signal Model 101

Proposed Dynamic Model 119

Table 7. Comparison b/w PCU Passed in Green Time

Improvement in Traffic Throughput

To quantify the performance gain, percentage improvement is calculated using the

formula:
Improvement %= ( Dynamic - Traditional/Traditional) × 100
Substituting values:

Improvement %= ( 119 - 101/101) × 100= 17.82%

Thus, the proposed dynamic signal timing logic demonstrates a 17.82%
improvement in vehicle throughput during peak traffic hours compared to the
traditional fixed-time system. This highlights the potential of PCU-based dynamic
signal allocation in improving traffic flow and reducing congestion at urban
intersections.

3.7 System Integration and Simulation

Once the individual components—vehicle detection, object tracking, PCU

calculation, and signal timing—were developed, they were integrated into a single
Python-based system designed to simulate real-time traffic signal control using real
intersection footage.

The system workflow is as follows:

Video Input: Recorded footage from an intersection is used as input.
Vehicle Detection: YOLOv5 identifies vehicles in each frame.
Vehicle Tracking: The SORT algorithm tracks each detected vehicle across
frames using unique IDs to avoid double counting.
PCU Estimation: Based on detected vehicle classes (car, bike, auto, truck, bus),
appropriate PCU values are assigned as per IRC guidelines.
Signal Time Computation: Each lane's total PCU is input to a dynamic
allocation module that calculates the optimal green signal duration.
Simulated Output: The system outputs green signal timings for each lane,
mimicking a live traffic signal environment.

31
Execution was performed through a Command Line Interface (CLI), allowing the
user to supply traffic video inputs and view output in real time. The project
utilized Python libraries including PyTorch, OpenCV, NumPy, and Matplotlib for
implementation.
This integrated setup demonstrates that the proposed model can adapt signal
timing based on current traffic demand, showing promise for practical deployment
at urban intersections.

3.8 Validation and Observations

To assess the performance and reliability of the proposed PCU-based dynamic

signal timing system, a validation study was conducted using empirical data
collected from a real intersection—Satyam Chowk, Bilaspur, Chhattisgarh. The
validation aimed to compare the throughput efficiency of the traditional fixed-
time signal method with the developed dynamic model.

3.8.1 Survey-Based Data Collection

Traffic video data was collected during three distinct time slots to represent
varying traffic conditions:
Sunday, 8:00 AM (Low Traffic)
Monday, 12:00 PM (Moderate Traffic)
Monday, 6:00 PM (Peak Traffic)
Lane-wise vehicle counts were converted to PCU values using the standard PCU
conversion factors defined by the Indian Roads Congress (IRC). This standardized
traffic flow data served as input for both the fixed-time model and the proposed
dynamic model.

3.8.2 Method of Validation

Under the traditional fixed-time system, a uniform green time of 30 seconds

per lane (total 120 seconds per cycle) was assumed.
In contrast, the proposed dynamic model computed lane-wise green time
based on PCU inputs, with a cycle time dynamically selected as either 80 or
120 seconds based on total demand.
For both methods, the number of PCUs able to pass during the allocated
green time was estimated based on typical saturation flow conditions and
average vehicle discharge rates per second.

32
3.8.3 Results and Observation

Method Cycle Time (sec) Total PCU PCUs Passed Improvement

Fixed-Time 120 164 101 –

Dynamic Model ( 120 164 119 17.82%

Table 8. Final Result Showing Overall Improvement in Traffic Flow

Fig. 1 Comparison between the conventional traffic

signal vs dynamic traffic signal

3.8.4 Key Observations

The dynamic model successfully identified imbalanced traffic loads among
lanes and redistributed green time accordingly.
A noticeable increase in vehicle throughput was achieved without exceeding
the 120-second cycle time.
The system demonstrated adaptive behavior and responsiveness to real-time
traffic conditions, unlike the rigid structure of the fixed-time approach.
These observations confirm that the dynamic signal allocation logic, when
based on PCU-weighted distribution, leads to more efficient intersection
performance. The validation supports the feasibility of deploying such
adaptive systems in real-world urban intersections to improve traffic flow and
reduce congestion.

33
 Chapter - 4
Result and Discussion
4.1. Introduction to Results

This chapter presents the results obtained from the implementation of our project
“Smart Signal Timing for Urban Intersections Using Real-Time Vehicle Detection
and PCU Calculations”, which utilizes real-time video-based vehicle detection,
PCU calculation, and dynamic traffic signal timing logic. The outcomes have been
analyzed using data collected through actual traffic video footage at a four-way
intersection. The primary goal is to evaluate how effectively the proposed model
improves lane-wise traffic flow compared to traditional fixed-time signal systems.
The results focus on two core metrics: first the total PCU throughput during a
standard 120-second signal cycle, and second the efficiency of green signal time
distribution among lanes based on traffic density. These parameters are crucial in
understanding the real-world performance of the model, especially under varying
traffic volumes recorded on different days and time slots (Sunday morning,
Monday afternoon, and Monday evening).
By comparing the number of vehicles (in terms of PCU) passing through each
lane under fixed versus dynamic signal allocation, this chapter provides clear, data-
driven evidence of the improvements achieved. Additionally, the analysis includes
how the dynamic model responded to real-time traffic variations and how it
allocated green time proportionally, ensuring fairness and increased throughput.
Overall, the findings in this chapter confirm that the PCU-based dynamic timing
strategy designed in this project leads to more efficient and responsive intersection
management, validating its potential for real-world application.

4.2. PCU Throughput Comparison

The effectiveness of the developed model—Smart Signal Timing for Urban

Intersections Using Real-Time Vehicle Detection and PCU Calculations—is
evaluated through a direct comparison of vehicle throughput under two traffic
control strategies: the conventional fixed-time signal method and the proposed
dynamic, PCU-based method.
During a standard 120-second signal cycle, the fixed-time approach allowed only
101 PCUs to pass through the intersection. In contrast, the dynamic model, which
allocates green time proportionally based on real-time PCU values per lane,
enabled 119 PCUs to pass in the same duration. This marks a clear improvement
of 17.82% in vehicle throughput.

34
This increase in PCU throughput demonstrates the efficiency of allocating green
time based on actual traffic volume rather than assigning equal or arbitrary
durations. By identifying which lanes have higher traffic loads and adjusting signal
durations accordingly, the model effectively reduces idle time at low-traffic lanes
and prioritizes high-demand lanes, resulting in a smoother and more responsive
traffic flow.
Such results validate the core logic of the project and indicate that the proposed
system can significantly enhance the performance of urban intersections,
especially during peak and variable traffic conditions.

4.3. Green Time Efficiency

The efficiency of green signal time allocation plays a critical role in managing
urban traffic intersections. In conventional traffic systems, green time is either
distributed equally among all directions or assigned using fixed durations without
considering the actual traffic volume at each lane. This often leads to inefficient
signal usage—vehicles in less congested lanes wait unnecessarily, while high-
density lanes remain congested due to insufficient green time.
In the developed PCU-based model, green time is dynamically allocated based on
the real-time traffic load captured through vehicle detection and converted into
PCU values. A base green time is assigned to all lanes to ensure minimum passage,
and the remaining time from the 120-second cycle is proportionally distributed
among the lanes according to their PCU share. This results in a more logical and
need-based allocation of green durations.
For instance, during the analysis, it was observed that a lane with a significantly
higher PCU count received more green time compared to others, enabling it to
clear more vehicles without causing prolonged queues. On the other hand, lanes
with lower PCU values were given just enough green time to manage their flow
efficiently, thus preventing unnecessary wastage of signal duration.
This intelligent allocation ensures that every second of green time is utilized
effectively, maximizing throughput and minimizing delays. The green time is no
longer a fixed input but a calculated output driven by real-time traffic demands,
making the system more adaptable and fair compared to traditional method

4.4. Survey-Based Validation

To validate the real-world applicability of the proposed model, traffic surveys were
conducted at a four-way intersection using actual video recordings. The aim was to
analyze how the dynamic signal system responds to varying traffic patterns across
different days and time slots. For this purpose, data was collected during three
distinct periods: Sunday morning (low traffic), Monday noon (moderate traffic),
and Monday evening (peak traffic).

35
Each vehicle recorded in the videos was classified using the trained detection
model and converted to its equivalent PCU value. This data was then used to
simulate green time allocation using the dynamic model. By comparing the green
time assigned under dynamic conditions to the actual PCU distribution across
lanes, the adaptability of the system became evident.
During low-traffic hours on Sunday morning, the model provided a more balanced
green time across lanes since PCU values were nearly equal. However, in the
Monday evening slot, where traffic volume was heavily skewed toward one or two
lanes, the model successfully reallocated green time in favor of those lanes without
any manual intervention. This behavior showcases the system's strength in
adapting to real-time conditions instead of relying on static timing plans.
These survey-based validations clearly demonstrate that the system dynamically
reacts to changing traffic patterns throughout the day. Rather than treating all
lanes equally, it adjusts in response to the current demand, thereby improving
fairness, throughput, and responsiveness. The same intersection, with different
traffic conditions, resulted in different signal behavior—all automatically handled
by the model without pre-programmed schedules.
This flexibility and real-time adaptability form the core strength of the proposed
system and make it a viable solution for smart urban traffic control.

4.5. Real-Time Responsiveness

A key feature of the proposed traffic signal model is its ability to respond to live
traffic conditions in near real-time. Unlike conventional systems that rely on pre-
set schedules or manual adjustments, the developed system processes live video
input to detect and classify vehicles, convert them into PCU values, and calculate
optimal green signal durations accordingly—all in a short, efficient cycle.
In the current offline implementation phase, videos from actual traffic
intersections were used as input. The system, powered by YOLOv5 for detection
and SORT for tracking, was able to detect multiple vehicle classes per lane and
generate PCU counts within seconds. Once the lane-wise PCU data was available,
the algorithm computed the base green time and distributed the remaining cycle
time based on the PCU ratio of each lane.
In a test run using pre-recorded traffic footage, the entire process from video input
to signal time calculation was completed within 10–12 seconds. This indicates that,
when integrated with real-time traffic camera feeds and automated signal
controllers, the system has the potential to function in a live environment with
minimal delay.

36
This responsiveness ensures that the system reacts promptly to dynamic traffic
scenarios—such as sudden vehicle surges in one direction or a drop in volume in
another. Instead of waiting for the next scheduled update, the system recalculates
and redistributes green time based on current demand, helping prevent congestion
and idle green signals.
Although full real-time deployment (with continuous live feed and automated
signal switching) is yet to be implemented, the current setup confirms that the
system is technically capable of achieving real-time responsiveness once integrated
with the required infrastructure.

4.6. Limitations
While the proposed model has demonstrated significant improvements in signal
timing efficiency and throughput, several limitations must be acknowledged to
provide a balanced perspective and guide future enhancements.
1. Lighting and Weather Sensitivity:
The accuracy of vehicle detection using computer vision is highly dependent on
lighting conditions. Poor visibility during nighttime or extreme weather conditions
(like fog or heavy rain) may reduce detection accuracy, leading to incorrect PCU
calculations and suboptimal signal time distribution.
2. False Positives and Missed Detections:
Although YOLOv5 is a robust detection model, occasional false positives (e.g.,
mistaking shadows or parked objects for vehicles) and missed detections
(especially for small or partially occluded vehicles) can affect the reliability of PCU
estimates, and consequently, the fairness of green time allocation.
3. Limited Vehicle Class Categorization:
The system currently categorizes a limited number of vehicle types (e.g., cars,
bikes, buses, trucks, autos). More complex traffic environments with non-standard
vehicles (e.g., handcarts, cycle rickshaws) may not be adequately represented in the
PCU calculations, slightly reducing accuracy in such regions.
4. Offline Processing Only:
As of now, the entire analysis is performed using recorded video data processed
offline. Real-time implementation—where live camera feeds continuously update
signal timings—is not yet deployed due to hardware and integration constraints.
5. Intersection-Specific Calibration:
The model’s green time logic (base time, PCU thresholds, etc.) was tuned for a
specific intersection. To scale the system city-wide, individual calibration may be
necessary for different intersection geometries, vehicle flow patterns, and traffic
rules.
Despite these limitations, the project successfully demonstrates the core feasibility
and effectiveness of using real-time PCU data for dynamic signal control. With
further development—such as real-time integration, broader vehicle classification,
and lighting compensation—these limitations can be addressed for more robust
deployment.
37
4.7. Discussion Summary

The results and observations from the implementation and testing of this project
—Smart Signal Timing for Urban Intersections Using Real-Time Vehicle Detection
and PCU Calculations—highlight its potential to significantly enhance traffic
signal efficiency in urban environments.
Through direct comparison, the dynamic PCU-based model outperformed the
conventional fixed-time method, achieving a 17.82% increase in vehicle
throughput during a standard 120-second signal cycle. This improvement was
made possible by intelligently allocating green signal time based on real-time
traffic demand rather than relying on rigid, pre-assigned durations.
The model proved capable of adapting to different traffic conditions across
multiple time slots and days, as validated by actual survey data. It responded
effectively to variations in lane-wise vehicle density, distributing green time fairly
and efficiently in accordance with the PCU load. Even in its offline phase, the
system demonstrated near real-time responsiveness—completing detection,
classification, PCU calculation, and green time distribution within seconds.
At the same time, limitations such as offline processing, lighting sensitivity, and
restricted vehicle classification were identified. These areas provide a clear
direction for future enhancement and integration with live traffic systems.
In summary, this project achieved its primary goals of improving green time
utilization, enhancing intersection throughput, and ensuring fairness through
PCU-based logic. The findings indicate strong potential for real-world application,
especially if supported by appropriate infrastructure like live traffic feeds, smart
cameras, and automated signal controllers.

38
 Chapter - 5
CONCLUSION
Urban traffic management has long been a challenge, with conventional fixed-time
signal systems often leading to inefficiencies, especially during periods of
fluctuating traffic flow. This project aimed to address these inefficiencies by
developing a smart traffic signal control system that dynamically allocates green
time based on real-time vehicle detection and PCU (Passenger Car Unit)
calculations. The system utilizes advanced computer vision techniques, such as
YOLOv5 for vehicle detection and SORT for tracking, alongside an innovative
green time allocation algorithm based on real-time traffic data.
The results of the project demonstrate significant improvements in traffic signal
efficiency. The dynamic model, which adapts to actual traffic conditions, achieved
a 17.82% improvement in vehicle throughput compared to conventional fixed-time
control methods. This improvement was achieved by allocating green time based
on the real-time density of vehicles, ensuring that high-traffic lanes received more
green time, thereby reducing delays and increasing overall intersection throughput.
One of the key strengths of this system is its adaptability to varying traffic
patterns. Validation through real-world traffic survey data confirmed that the
system efficiently adjusts its green time allocation across different time slots, from
low-demand periods to peak traffic hours. This adaptability ensures that the
system can handle a wide range of traffic conditions, making it a versatile solution
for urban intersections.
While the current implementation was tested offline, the model demonstrated
strong real-time responsiveness, processing vehicle detection, tracking, and green
time calculation within a short time frame. This indicates that the system is well-
suited for live deployment with minimal delay, provided it is integrated with
appropriate hardware and live video feeds.
Despite the promising results, the project also highlighted several limitations that
need to be addressed in future work. These include challenges such as lighting
conditions that may affect vehicle detection accuracy, occasional false positives or
missed detections, and limited vehicle class coverage. Additionally, the system’s
offline processing capability should be transitioned to a live setup, and further
enhancements, such as emergency vehicle prioritization, could be incorporated to
make the system more robust.
In conclusion, this project presents a novel and effective approach to improving
urban traffic management through intelligent, data-driven signal control. The
dynamic allocation of green time based on real-time vehicle detection and PCU
calculations not only enhances signal efficiency but also promises to contribute to
smoother traffic flow and reduced congestion in urban intersections. With the
potential for further refinement and integration into smart city infrastructure, this
system offers a significant step forward in the evolution of intelligent traffic signal
systems.

39
 Chapter - 6
REFERENCES
Goenawan, C. R. (2024). ASTM :Autonomous Smart Traffic Management System
Using Artificial Intelligence CNN and LSTM. http://arxiv.org/abs/2410.10929

Sangeetha, R. G., Hemanth, C., Dipesh, R., Samriddhi, K., Venetha, S., Abbas
Alif, M., Arjun, S., & S, V. K. (2024). Density Based Real-time Smart Traffic
Management System along with Emergency Vehicle Detection for Smart Cities.
International Journal of Intelligent Transportation Systems Research, 22(2), 328–
338. https://doi.org/10.1007/s13177-024-00400-9

Sakhare, N., Hedau, M., Malpure, O., Shah, T., & Ingle, A. (n.d.). International
Journal of INTELLIGENT SYSTEMS AND APPLICATIONS IN
ENGINEERING Smart Traffic: Integrating Machine Learning, and YOLO for
Adaptive Traffic Management System. In Original Research Paper International
Journal of Intelligent Systems and Applications in Engineering IJISAE (Vol. 2024,
Issue 12s). www.ijisae.org

T, B. S., Kumar, R., & Rao, S. G. (2022). Smart Traffic Management System: A
Literature Review. International Journal of Innovative Research in Electrical,
Electronics, Instrumentation and Control Engineering Impact, 2.
https://doi.org/10.17148/IJIREEICE.2022.10201

Sarrab, M., Pulparambil, S., & Awadalla, M. (2020). Development of an IoT

based real-time traffic monitoring system for city governance. Global Transitions,
2, 230–245. https://doi.org/10.1016/j.glt.2020.09.004

Sarrab, M., Pulparambil, S., & Awadalla, M. (2020). Development of an IoT

based real-time traffic monitoring system for city governance. Global Transitions,
2, 230–245. https://doi.org/10.1016/j.glt.2020.09.004

Mishra, R., Kumar, P., Arkatkar, S. S., Sarkar, A. K., & Joshi, G. J. (2017). Novel
area occupancy–based method for passenger car unit estimation on multilane
urban roads under heterogeneous traffic scenario. Transportation Research
Record, 2615(1), 82–94. https://doi.org/10.3141/2615-10

40