ABSTRACT

This project introduces an intelligent traffic light optimization system

that combines computer vision with fuzzy logic to dynamically allocate
signal timings at intersections. Traditional traffic signal systems
operate on fixed timing patterns, often leading to inefficient traffic
management and unnecessary delays. Our approach addresses this
limitation by developing an adaptive system that analyzes real-time
traffic conditions using YOLOv8 for vehicle detection and
classification, and employs fuzzy logic to determine optimal signal
timing based on multiple parameters including vehicle density, vehicle
types, and waiting times. The system processes images from four
different lanes (north, east, south, west) at an intersection, calculates
weighted vehicle counts and traffic density, and applies fuzzy rules to
determine lane priorities. Green signal time is then proportionally
allocated based on these priorities while respecting minimum and
maximum time constraints. Performance visualization includes
annotated images showing detected vehicles, graphical
representations of traffic conditions, and detailed timing decisions. The
proposed system aims to reduce congestion, decrease average
waiting times, and improve overall traffic flow efficiency through
intelligent, context-aware signal timing.
 CHAPTER 1: Introduction

1.1 Background and Motivation

The rapid urbanization and increasing vehicle ownership around the
world have led to unprecedented traffic congestion in metropolitan
areas. According to the 2023 Global Traffic Congestion Index, drivers
in major cities spend an average of 102 hours per year stuck in traffic
congestion, resulting in significant economic losses estimated at $305
billion annually in the United States alone. This congestion not only
causes frustration and wasted time for commuters but also contributes
to increased fuel consumption, higher carbon emissions, and
degraded air quality in urban environments.
At the heart of urban traffic management systems are traffic signals,
which control the flow of vehicles at intersections. Traditional traffic
signal systems operate on fixed time cycles that are programmed
based on historical traffic data collected during traffic surveys. While
these systems may include different timing plans for peak hours, off-
peak hours, and weekends, they fundamentally lack the ability to
adapt in real-time to changing traffic conditions. This inflexibility leads
to suboptimal traffic flow, with some lanes experiencing excessive
congestion while others remain underutilized.
Furthermore, conventional traffic management approaches often treat
all vehicles equally, failing to account for the heterogeneous nature of
traffic. In reality, different vehicle types—such as cars, motorcycles,
buses, and trucks—have varying impacts on traffic flow due to
differences in size, acceleration capabilities, and passenger capacity.
Public transportation vehicles like buses, which may carry dozens of
passengers, deserve higher priority than private vehicles carrying
single occupants when considering efficient people movement rather
than merely vehicle movement.
The emergence of smart city initiatives worldwide has placed renewed
emphasis on intelligent transportation systems (ITS) as a key
component of sustainable urban development. These initiatives aim to
leverage modern technologies to improve the quality of urban life, with
efficient traffic management being a critical aspect. The potential
benefits of intelligent traffic control systems are substantial: reduced
travel times, decreased fuel consumption, lower emissions, and
improved road safety.

1.2 Current Technological Landscape

Recent advancements in computational technologies, computer
vision, and artificial intelligence have created new opportunities for
developing more sophisticated traffic management systems. These
technologies enable real-time monitoring and analysis of traffic
conditions, allowing for adaptive control strategies that respond
dynamically to changing traffic patterns.
Computer vision techniques, particularly deep learning-based object
detection models such as YOLO (You Only Look Once), have
revolutionized the field of traffic monitoring. These models can
accurately detect and classify vehicles from video feeds in real-time,
eliminating the need for expensive in-road sensors like inductive loops
or radar detectors. Camera-based systems offer several advantages,
including ease of installation, lower maintenance costs, and the ability
to capture rich visual information beyond mere vehicle presence.
Concurrently, artificial intelligence methods like fuzzy logic have
emerged as powerful tools for decision-making in complex systems
characterized by uncertainty and imprecision. Unlike classical binary
logic, fuzzy logic can handle vague or approximate information
through degrees of membership in sets, making it particularly well-
suited for traffic control applications where decisions must balance
multiple competing factors and objectives.

1.3 Project Overview and Approach

Our project introduces an intelligent traffic light optimization system
that combines computer vision with fuzzy logic to dynamically allocate
signal timings at intersections. This hybrid approach leverages the
strengths of both technologies: computer vision provides accurate
real-time traffic state assessment, while fuzzy logic enables nuanced
decision-making that balances multiple criteria.
The system architecture consists of four main components:
1. Image Acquisition Module: High-definition cameras mounted at
each approach to the intersection capture real-time video feeds
of traffic conditions. These cameras are positioned to provide
clear views of vehicles approaching the intersection from all
directions.
2. Vehicle Detection and Classification Module: The system
employs YOLOv8, a state-of-the-art deep learning object
detection algorithm, to process the video feeds and identify
vehicles. YOLOv8 not only detects vehicles but also classifies
them into categories (cars, motorcycles, buses, trucks) and
provides their spatial coordinates within the image.
3. Fuzzy Logic Controller: This component forms the decision-
making core of the system. It processes multiple input variables,
including weighted vehicle counts, traffic density (percentage of
road area occupied by vehicles), and waiting times for each lane.
These inputs are fuzzified using membership functions,
processed through a rule base that encapsulates traffic
management expertise, and then defuzzified to produce priority
scores for each lane.
 4. Signal Timing Allocation Module: Based on the priority scores
determined by the fuzzy controller, this module calculates
optimal green times for each lane. The allocation algorithm
ensures that lanes with higher priorities receive proportionally
more green time while still guaranteeing minimum service levels
for all lanes. It also manages the transition between phases,
including yellow signal timing.
Our approach differs from existing systems in several key aspects:
First, it considers the heterogeneous nature of traffic by assigning
different weights to various vehicle types based on their impact on
traffic flow. For example, buses and trucks receive higher weights than
cars due to their larger size and different dynamics.
Second, it employs traffic density as a more accurate measure of
congestion instead of simple vehicle counting. By calculating the
percentage of road area occupied by vehicles, the system can better
assess the actual congestion level, especially in mixed traffic
conditions where vehicle sizes vary significantly.
Third, it incorporates waiting time as an important factor to ensure
fairness. Lanes that have been waiting longer receive increased
priority, preventing situations where minor approaches might
experience excessive delays during peak hours on major approaches.
Fourth, the fuzzy logic framework enables the system to balance
these multiple, sometimes competing factors in a way that mimics
human decision-making but operates continuously and consistently.
The fuzzy rules encapsulate traffic engineering expertise in a form that
can be applied automatically and adjusted as needed.
1.4 Anticipated Benefits and Significance
The implementation of our intelligent traffic light control system is
expected to yield several significant benefits:
 Reduced Average Waiting Times: By allocating green time
proportionally to actual traffic demand, the system minimizes
unnecessary waiting times across all approaches to the
intersection.
 Improved Traffic Flow Efficiency: The dynamic response to
changing traffic conditions ensures that signal timing is always
optimized for current rather than historical patterns.
 Decreased Fuel Consumption and Emissions: Less idle time
at intersections translates directly to fuel savings and reduced
greenhouse gas emissions, contributing to environmental
sustainability goals.
 Enhanced Public Transportation Efficiency: By giving
appropriate weight to buses, the system indirectly prioritizes
mass transit, potentially encouraging greater use of public
transportation.
 Scalable and Cost-Effective Implementation: Camera-based
detection eliminates the need for expensive in-road sensors,
making the system more affordable to deploy across multiple
intersections.
The significance of this project extends beyond immediate traffic
management improvements. It represents a step toward fully
intelligent transportation systems that can eventually coordinate
multiple intersections in a network, integrate with navigation systems
to provide routing recommendations, and even communicate with
connected vehicles. As cities continue to grow and vehicle numbers
increase, such intelligent systems will become increasingly essential
for maintaining mobility and quality of life in urban environments.
In the following chapters, we will delve deeper into the theoretical
foundations of our approach, analyze existing solutions and their
limitations, define the specific problems we aim to solve, describe our
proposed methodology in detail, and present a comprehensive
framework for implementation and evaluation.
 CHAPTER 2: Fundamentals and Literature Survey

2.1 Fundamentals of Traffic Signal Control

2.1.1 Traditional Traffic Signal Control Methods
Traffic signal control has evolved significantly since the installation of
the first electric traffic signal in Cleveland, Ohio in 1914. Traditional
traffic signal control methods can be broadly categorized into three
generations:
First Generation: Fixed-Time Control
Fixed-time control systems operate on predetermined timing plans
without considering real-time traffic conditions. These systems
typically have different timing plans for different periods (morning
peak, off-peak, evening peak) and days of the week. The signal timing
parameters are calculated based on historical traffic data collected
through manual surveys or automatic counters. Webster's method
(1958) is commonly used to determine optimal cycle length and green
time allocation based on saturation flow and traffic volume. The
primary advantage of fixed-time control is its simplicity and low cost,
but its inability to adapt to unexpected traffic variations limits its
effectiveness in dynamic urban environments.
Second Generation: Vehicle-Actuated Control
Vehicle-actuated control systems use detectors (typically inductive
loops embedded in the pavement) to register vehicle presence and
adjust signal timing accordingly. These systems can be fully actuated,
where all approaches have detectors, or semi-actuated, where only
minor roads have detectors. When a vehicle is detected on a minor
approach with a red signal, the controller terminates the green phase
on the major approach (after a minimum green time) and transfers the
right-of-way to the minor approach. While more responsive than fixed-
time systems, these actuated controllers still rely on predetermined
maximum and minimum green times and operate based on simple
vehicle presence rather than comprehensive traffic analysis.
Third Generation: Responsive Control
Responsive control systems use real-time traffic data from multiple
detectors to optimize signal timing continuously. Examples include
SCOOT (Split Cycle Offset Optimization Technique) developed in the
UK and SCATS (Sydney Coordinated Adaptive Traffic System) from
Australia. These systems collect traffic data at key locations, transmit
it to a central computer that runs optimization algorithms, and send
updated timing parameters back to local controllers. While more
effective than previous generations, these systems still face
challenges with detector maintenance, communication reliability, and
computational complexity when handling large networks.
2.1.2 Computer Vision for Traffic Analysis
Computer vision technology has revolutionized traffic monitoring by
enabling non-intrusive, comprehensive traffic data collection. The
evolution of computer vision techniques for traffic analysis includes:
Traditional Image Processing Approaches
Early computer vision systems for traffic analysis relied on background
subtraction, frame differencing, and feature tracking. Background
subtraction identifies moving objects by comparing each video frame
with a reference background model. These techniques work
reasonably well under controlled conditions but struggle with
environmental variations (lighting changes, shadows, weather
conditions) and complex traffic scenarios.
Feature-Based Vehicle Detection
More robust approaches employ feature extraction methods such as
Histogram of Oriented Gradients (HOG), Haar-like features, and
Scale-Invariant Feature Transform (SIFT). These features, combined
with classifiers like Support Vector Machines (SVM) or AdaBoost, can
detect vehicles with greater reliability across various conditions.
However, they typically require careful parameter tuning and feature
engineering.
Deep Learning Revolution
The introduction of deep learning, particularly Convolutional Neural
Networks (CNNs), marked a paradigm shift in computer vision.
Region-based CNNs (R-CNN and its variants Fast R-CNN and Faster
R-CNN) significantly improved object detection performance but
required substantial computational resources. The YOLO (You Only
Look Once) family of algorithms addressed the speed limitations by
reformulating object detection as a single regression problem,
predicting bounding boxes and class probabilities directly from full
images in one evaluation.
YOLOv8 Architecture
YOLOv8, the latest in the YOLO series, incorporates several
architectural improvements:
 A modified CSPDarknet53 backbone network with Cross-Stage
Partial connections for efficient feature extraction
 Path Aggregation Network (PANet) for feature pyramid
construction that enhances information flow between different
resolution levels
 Spatial Pyramid Pooling (SPP) to increase the receptive field
without affecting computational complexity
 Mish activation function for improved gradient flow and model
performance
 Mosaic data augmentation for training with multiple images
simultaneously
 Figure: YOLO Object Detection Process

YOLOv8 achieves state-of-the-art performance on the COCO dataset

with 55.8% AP (Average Precision) while maintaining real-time
inference speeds of 45 frames per second on standard GPUs, making
it ideal for traffic monitoring applications.
Vehicle Classification and Tracking
Beyond detection, computer vision systems can classify vehicles into
categories (cars, motorcycles, trucks, buses) and track their
movement across video frames. Classification typically uses the visual
features extracted by the detection network, passing them through
additional classification layers. Tracking algorithms like SORT (Simple
Online and Realtime Tracking) or DeepSORT combine detection with
motion prediction to maintain vehicle identity across frames, enabling
measurement of travel times and trajectories.
2.1.3 Fuzzy Logic Control Systems
Fuzzy logic, introduced by Lotfi Zadeh in 1965, provides a
mathematical framework for dealing with uncertain or imprecise
information. Unlike classical set theory where an element either
belongs to a set (membership value of 1) or does not (membership
value of 0), fuzzy logic allows partial membership, with values ranging
from 0 to 1.
Fuzzy Set Theory
A fuzzy set A in a universe of discourse X is characterized by a
membership function μA(x) that associates each element x in X with a
real number in the interval [0,1]. The value μA(x) represents the
degree of membership of x in the fuzzy set A. For example, in traffic
control, a vehicle count of 15 might have a membership degree of 0.7
in the "Medium" fuzzy set and 0.3 in the "High" fuzzy set.
Fuzzy Logic Controller (FLC) Architecture
A fuzzy logic controller consists of four main components:
1. Fuzzification Interface: Converts crisp input values into fuzzy
values using membership functions. The shape of these
functions (triangular, trapezoidal, Gaussian) affects the
controller's behavior and must be carefully designed.
2. Knowledge Base: Contains the rule base and database. The
rule base consists of fuzzy IF-THEN rules that capture expert
knowledge about the system behavior. The database defines the
membership functions used by the rules.
3. Inference Engine: Evaluates which control rules are relevant at
the current time and decides what the control input should be.
Common inference methods include Mamdani and Sugeno
types.
4. Defuzzification Interface: Converts the fuzzy output into a crisp
value that can be used as a control signal. Popular
 defuzzification methods include centroid, weighted average, and
maximum membership.
Advantages of Fuzzy Logic for Traffic Control
Fuzzy logic offers several advantages for traffic signal control:
 It handles imprecision and uncertainty inherent in traffic systems
 It can incorporate human expertise through linguistic rules
 It manages multiple, sometimes conflicting criteria in decision-
making
 It does not require precise mathematical models of the controlled
system
 It provides smooth control actions through gradual transitions
between operating conditions
Mathematical Foundations
While the linguistic interpretation makes fuzzy logic intuitive, its
mathematical foundation ensures rigorous implementation. Key
operations include:
 Fuzzy Set Operations: Union (OR), intersection (AND), and
complement (NOT) operations are extended to fuzzy sets using
operators like maximum, minimum, and complement:
o μA∪B(x) = max[μA(x), μB(x)]
o μA∩B(x) = min[μA(x), μB(x)]
o μĀ(x) = 1 - μA(x)
 Fuzzy Relations: Represent mappings between fuzzy sets in
different universes of discourse, crucial for encoding IF-THEN
rules.
 Fuzzy Inference: The process of mapping from fuzzy inputs to
fuzzy outputs based on fuzzy rules. For example, using Mamdani
inference:
 o The firing strength of each rule is determined by applying
fuzzy operators to antecedent parts
o The consequent of each rule is a fuzzy set scaled by the
firing strength
o The overall output is the aggregation of all rule
consequents
 Defuzzification Methods: Mathematical techniques to convert
fuzzy output to crisp values. The centroid method calculates the
center of gravity of the aggregated fuzzy set:
z* = ∫μA(z)·z dz / ∫μA(z) dz

2.2 Literature Survey

2.2.1 Early Developments in Intelligent Traffic Control
The application of artificial intelligence to traffic signal control began in
the late 1970s with the pioneering work of Pappis and Mamdani [1],
who developed the first fuzzy logic controller for an isolated
intersection. Their system used queue length and arrival rate as inputs
and demonstrated superior performance compared to conventional
controllers, particularly under varying traffic conditions.
The 1990s saw increased research interest in intelligent traffic control
methods. Niittymäki [2] extended the fuzzy control approach by
incorporating additional variables such as pedestrian waiting time and
vehicle speeds. Field tests in Helsinki showed reductions in average
delay of 12-15% compared to optimized fixed-time control. During the
same period, Choi et al. [3] explored reinforcement learning
techniques where the controller learned optimal signal timing patterns
through trial and error interaction with a simulated traffic environment.
While promising, these early reinforcement learning approaches
required extensive training in simulation before deployment.
2.2.2 Integration of Computer Vision in Traffic Management
The early 2000s marked the beginning of camera-based traffic
monitoring systems. Chen and Yang [4] combined cellular automata
traffic models with fuzzy decision systems but relied on simplified
visual processing for vehicle detection. Their simulation results
indicated potential delay reductions of 17-22% compared to fixed-time
systems but lacked real-world validation.
A significant advancement came with Srinivasan et al.'s work [5] on
multi-agent reinforcement learning for traffic signal control. Their
system used multiple coordinated controllers that learned cooperative
strategies to manage traffic across adjacent intersections. The agents
shared information about traffic states and planned actions,
demonstrating improvements in network-wide performance metrics.
Zhou et al. [6] developed one of the first integrated systems using
computer vision for real-time traffic monitoring. Their approach
employed background subtraction techniques to detect and count
vehicles at intersections. While innovative, the system's performance
degraded under adverse weather and lighting conditions, highlighting
the challenges of robust visual processing in uncontrolled
environments.

2.2.3 Hybrid and Advanced Intelligent Systems

The integration of multiple AI techniques characterizes more recent
developments in traffic control. Askerzade and Mahmood [7] proposed
a hybrid system combining fuzzy logic with neural networks. The
neural network component adaptively adjusted the membership
functions based on traffic patterns learned from historical data, while
the fuzzy controller handled real-time decision-making. Their
simulation results showed a 20% reduction in average waiting time
compared to conventional actuated controllers.
Garg et al. [8] developed an image processing-based traffic density
estimation system for signal control. Their approach used edge
detection and blob analysis to count vehicles and estimate congestion
levels. While simpler than deep learning methods, their system
demonstrated the feasibility of camera-based traffic management with
limited computational resources.
The advent of deep learning revolutionized computer vision
capabilities for traffic analysis. Gao et al. [9] combined YOLO-based
vehicle detection with deep reinforcement learning for signal control.
Their system processed traffic camera feeds to extract detailed state
representations that served as inputs to a deep Q-network.
Simulations using real traffic data from Beijing showed improvements
of up to 26% in average travel time during peak hours. However, the
high computational requirements of their approach limited its practical
deployment on standard traffic controller hardware.

2.2.4 Recent Advances and Current State of the Art

Recent research has focused on addressing the practical challenges
of deploying intelligent traffic systems in real-world conditions. Sharma
et al. [10] developed a multi-criteria fuzzy decision system that
balanced vehicle density, waiting time, and queue length. Their
approach used a simplified vision system based on background
subtraction and contour analysis, making it feasible for implementation
on embedded platforms. Field tests at a single intersection in Delhi
showed average delay reductions of 19% compared to fixed-time
control.
The most recent advances leverage transformer-based architectures
for more robust visual processing. Khalid et al. [11] combined vision
transformers with adaptive fuzzy control for traffic management. Their
system used attention mechanisms to focus on relevant areas in traffic
scenes, improving detection accuracy under challenging conditions.
 The fuzzy controller featured adaptive membership functions that
automatically adjusted based on historical performance data.
Simulations using the SUMO (Simulation of Urban Mobility) platform
demonstrated improvements of 23-30% in average waiting time
across various traffic scenarios.

2.2.5 Analysis of Existing Approaches

The following table provides a detailed comparison of key research in
traffic signal control, highlighting their approaches, technologies,
performance metrics, and limitations:

Reference Year Methodology Key Performanc Limitations Real-

Technologies e Metrics World
Testing
Pappis and 1977 Fuzzy logic Simple fuzzy 10–15% Limited to 2 Simulation
Mamdani control rules reduction in input variables, only
average single
delay intersection
Niittymäki 2001 Multi-variable Extended rule 12–15% Limited Field
fuzzy control base, reduction in coordination tested in
inductive loop delays, 8– with adjacent Helsinki
detectors 17% intersections
reduction in
stops
Choi et al. 2002 Reinforcemen Q-learning 15–20% Slow Simulation
t learning algorithm, reduction in convergence, only
state-action queue length limited state
mapping representation
Chen and 2003 Cellular Traffic flow 17–22% Simplified Simulation
Yang automata with modeling, delay traffic model, only
fuzzy rules simple reduction in limited visual
detection simulation processing
Srinivasan et 2006 Multi-agent Distributed 20% High Limited
al. reinforcement control, agent reduction in computational field
learning communicatio network-wide complexity, testing
n delay communication
overhead
Zhou et al. 2010 Background Real-time 10–15% Poor Tested at
subtraction image reduction in performance in 2
for vehicle processing, average adverse intersectio
 detection adaptive delay weather, ns
control occlusion
issues
Askerzade 2013 Neuro-fuzzy Neural 20% Extensive Simulation
and hybrid system network reduction in training only
Mahmood adaptation, waiting time requirements,
fuzzy decision calibration
issues
Garg et al. 2016 Density- Edge 15–18% Limited Tested at
based control detection, blob reduction in accuracy in 1
with image analysis waiting time congested intersectio
processing conditions, n
binary
classification
Gao et al. 2019 Deep CNN-based 26% High Simulation
reinforcement detection, reduction in computational with real
learning with deep Q- travel time cost, complex data
YOLO network implementation
Sharma et al. 2021 Multi-criteria Background 19% delay Limited vehicle Field
fuzzy system subtraction, reduction classification, tested in
contour single Delhi
analysis intersection
Khalid et al. 2023 Vision Attention 23–30% Specialized Simulation
transformer mechanisms, reduction in hardware with
with adaptive adaptive waiting time requirements, SUMO
fuzzy control membership complex
functions calibration

Table: Analysis of Intelligent Traffic Signal Control Approaches

2.2.6 Research Gaps and Opportunities

The literature review reveals several important research gaps and
opportunities in intelligent traffic signal control:
1. Integration Challenges: While both computer vision and
intelligent control techniques have advanced separately, their
effective integration remains challenging. Most systems rely on
simplifications in either the detection component or the control
algorithm to manage computational complexity.
2. Vehicle Classification Impact: Few studies comprehensively
investigate how different vehicle types affect traffic flow and
 signal optimization. Most approaches treat all vehicles equally or
use simple binary classifications (light/heavy vehicles).
3. Real-Time Performance: Many advanced approaches
demonstrate excellent results in simulation but face
implementation challenges in real-time systems due to
computational requirements or hardware limitations.
4. Reliability Under Varying Conditions: Computer vision
systems still struggle with environmental variations such as
lighting changes, weather conditions, and occlusions. More
robust approaches are needed for consistent performance.
5. Comprehensive Optimization Criteria: Most existing systems
optimize for a limited set of parameters (typically average delay
or throughput) without considering broader factors like fairness,
fuel consumption, or emissions.
6. Scalability to Network Level: While many approaches work
well for isolated intersections, extending them to coordinate
multiple intersections remains challenging due to the exponential
increase in state space and the complexity of coordination.
Our proposed system aims to address several of these gaps by
developing a comprehensive approach that balances detection
accuracy, control sophistication, and computational efficiency. By
leveraging the latest advances in YOLOv8 for efficient object detection
and designing a specialized fuzzy logic controller that considers
multiple traffic parameters, we aim to create a practical system that
can be deployed in real-world traffic management scenarios.
 CHAPTER 3: Problem Statement

The current traffic management systems face several significant

challenges that impact urban mobility and quality of life:
3.1 Limitations of Fixed-Time Traffic Signal Systems
Traditional traffic control systems operate on predefined timing
patterns regardless of actual traffic conditions. These fixed-time
systems are designed based on historical traffic data and often fail to
adapt to:
 Rush hour congestion
 Special events causing unexpected traffic surges
 Accidents or road closures that alter typical traffic patterns
 Seasonal variations in traffic volume
 Day-to-day fluctuations in vehicle flow
3.2 Inefficient Utilization of Intersection Capacity
Current systems frequently result in:
 Extended wait times for vehicles in less congested lanes while
the green signal is allocated to empty lanes
 Inadequate green time for lanes with heavy traffic
 Uniform treatment of all vehicle types despite their different
impacts on traffic flow
 Failure to account for cumulative waiting times, leading to
fairness issues
3.3 Technical Challenges in Existing Adaptive Systems
Attempts to create adaptive traffic control systems have encountered
numerous difficulties:
 Reliance on expensive and maintenance-intensive in-road
sensors like inductive loops
 Limitations of single-parameter optimization (considering only
vehicle count)
 Difficulty in balancing multiple competing objectives (throughput,
fairness, emergency vehicle priority)
 High computational requirements limiting real-time performance
 Lack of integration between detection systems and control
algorithms

3.4 Specific Problems Addressed by Our Project

Our project aims to overcome these limitations by addressing the
following specific problems:
1. Dynamic Traffic Condition Assessment: How to accurately detect
and classify vehicles in real-time across multiple lanes to create
a comprehensive traffic state representation.
2. Multi-criteria Decision Making: How to balance multiple factors
(vehicle density, vehicle types, waiting times) to determine
optimal signal timing for each lane.
3. Adaptive Signal Timing: How to dynamically allocate green time
that is proportional to actual traffic needs while ensuring
minimum service levels for all lanes.
4. Heterogeneous Traffic Consideration: How to account for the
varied impact of different vehicle types (cars, motorcycles,
buses, trucks) on overall traffic flow.
 5. System Integration: How to create a unified system that
combines computer vision capabilities with intelligent decision-
making in a computationally efficient manner.
By addressing these specific problems, our proposed system aims to
significantly improve traffic flow efficiency, reduce congestion,
decrease average waiting times, and enhance overall urban mobility.
 CHAPTER 4: Proposed Work

Our proposed intelligent traffic light control system integrates

computer vision with fuzzy logic to optimize signal timing at
intersections. The methodology is designed to address the limitations
of traditional systems by creating a comprehensive solution that
adapts to real-time traffic conditions.

4.1 System Architecture

The proposed system consists of five main components:
1. Image Acquisition Module: Captures traffic images from cameras
positioned at each lane of the intersection
2. Vehicle Detection and Classification Module: Processes
images to identify and categorize vehicles
3. Fuzzy Logic Controller: Determines lane priorities based on
multiple traffic parameters
4. Signal Timing Allocation Module: Calculates optimal green
times based on fuzzy priorities
5. Visualization Module: Generates annotated images showing
detected vehicles and visual timing layout for the intersection.
 System Architecture Diagram

4.2 Vehicle Detection and Classification

We employ YOLOv8, a state-of-the-art deep learning-based object
detection model, to detect and classify vehicles in real-time. The
model identifies four primary vehicle types:
 Cars
 Motorcycles
 Buses
 Trucks
Each vehicle type is assigned a weight based on its impact on traffic
flow:
 Cars: 1.0
 Motorcycles: 0.5
 Buses: 2.5
 Trucks: 2.0
These weights reflect the space occupied by each vehicle type and
their influence on traffic dynamics. For example, buses and trucks
occupy more space and typically move slower than cars, thus having
a higher impact on congestion.
The system calculates traffic density for each lane as the percentage
of road area occupied by vehicles. This provides a more accurate
representation of congestion compared to simple vehicle counting.

4.3 Fuzzy Logic Control System

The fuzzy logic controller processes three input variables to determine
lane priorities:
1. Vehicle Count: The weighted sum of vehicles present in the lane
o Fuzzy sets: Low (0-10), Medium (5-25), High (20-30)
2. Density: The percentage of road area occupied by vehicles
o Fuzzy sets: Low (0-30%), Medium (20-80%), High (70-
100%)
3. Waiting Time: The duration a lane has been waiting for a green
signal
o Fuzzy sets: Short (0-20s), Medium (15-45s), Long (40-60s)
The output variable is Priority, with fuzzy sets:
 Very Low (0-20)
 Low (10-50)
 Medium (40-80)
 High (70-100)
 Very High (90-100)
Fuzzy rules map combinations of input conditions to appropriate
priority levels. For example:
 IF vehicle_count is HIGH AND density is HIGH AND
waiting_time is LONG THEN priority is VERY_HIGH
 IF vehicle_count is LOW AND density is LOW AND waiting_time
is SHORT THEN priority is VERY_LOW
The system employs 11 rules to cover various traffic scenarios,
ensuring a comprehensive decision-making framework.

4.4 Dynamic Signal Timing Allocation

Green time allocation is performed using the following algorithm:
Base Algorithm (Previous Approaches):
1. Count vehicles in each lane
2. Allocate fixed green time proportional to vehicle count
3. Apply fixed minimum and maximum time constraints
Proposed Algorithm (GSA - Fuzzy-Based Green Signal Allocation):
1. For each lane:
a. Detect and classify vehicles using YOLOv8
b. Calculate traffic density and weighted vehicle count
c. Determine lane priority using fuzzy logic controller based on:
- Weighted vehicle count
- Traffic density
- Waiting time
2. Calculate total priority across all lanes
3. Distribute available cycle time proportionally based on priorities:
green_time[lane] = max(min_green_time, min(max_green_time,
(priority[lane]/total_priority) * available_time))
4. Adjust timings to fit within cycle constraints
5. Update waiting times for non-active lanes
The system ensures that:
 Each lane receives at least the minimum green time (10
seconds)
 No lane exceeds the maximum green time (60 seconds)
 The total cycle time (including yellow phases) remains consistent
(120 seconds)
 Lanes with higher priority receive proportionally more green time

4.5 Performance Visualization and Analysis

The system provides comprehensive visualization tools to analyze its
performance:
1. Annotated Images: Processed images showing detected vehicles
with bounding boxes and class labels
2. Traffic Statistics: Visual representation of vehicle counts, density,
waiting times, and priorities for each lane
3. Signal Timing Charts: Graphical display of allocated green times
based on traffic conditions
4. Fuzzy Membership Visualizations: Illustrations of fuzzy
membership functions and control surfaces
These visualizations enable traffic engineers to understand the
system's decision-making process and evaluate its effectiveness.
 CHAPTER 6: Conclusion

The proposed fuzzy logic-based traffic light optimization system

represents a significant advancement in intelligent transportation
systems. By integrating state-of-the-art computer vision with fuzzy
logic control, our approach addresses the fundamental limitations of
traditional fixed-time traffic signal systems.
The key contributions of our work include:
1. A comprehensive methodology that considers multiple traffic
parameters simultaneously, including vehicle types, traffic
density, and waiting times, providing a more nuanced approach
to traffic management.
2. The implementation of YOLOv8 for accurate and efficient vehicle
detection and classification, enabling real-time processing of
traffic conditions.
3. A fuzzy logic controller that effectively handles the uncertainty
and complexity inherent in traffic systems, making balanced
decisions that optimize overall traffic flow.
4. A dynamic signal timing allocation algorithm that ensures
proportional distribution of green time based on actual traffic
needs while maintaining fairness across all lanes.
5. Visualization tools that provide insights into the system's
operation and facilitate performance analysis and fine-tuning.
The proposed system has several practical implications:
 Reduced congestion and waiting times at intersections
 Improved traffic flow efficiency and reduced travel times
  Decreased fuel consumption and emissions due to less idling
 Enhanced urban mobility and quality of life
 A flexible framework that can be adapted to various intersection
configurations
While our current implementation focuses on a single intersection, the
methodology can be extended to coordinate multiple intersections in a
traffic network. Future work could incorporate pedestrian detection,
emergency vehicle prioritization, and machine learning techniques to
further enhance the system's adaptability and effectiveness.
In conclusion, this project demonstrates the feasibility and potential
benefits of intelligent traffic light control systems that leverage
computer vision and fuzzy logic to create more efficient and
responsive urban transportation infrastructure.
References
[1] C. P. Pappis and E. H. Mamdani, "A Fuzzy Logic Controller for a
Traffic Junction," IEEE Transactions on Systems, Man, and
Cybernetics, vol. 7, no. 10, pp. 707-717, 1977.
[2] J. Niittymäki, "Installation and experiences of field testing a fuzzy
signal controller," European Journal of Operational Research, vol.
131, no. 2, pp. 273-281, 2001.
[3] Y. K. Choi, Y. S. Lim, and J. H. Kim, "Implementation of Traffic
Signal Control System using Reinforcement Learning," in Proceedings
of the 10th International Conference on Machine Learning
Applications, pp. 368-375, 2002.
[4] B. Chen and H. H. Yang, "Automated Traffic Light Control System
Using Fuzzy Logic and Cellular Automata," in Proceedings of the
IEEE International Conference on Networking, Sensing and Control,
vol. 2, pp. 1154-1159, 2003.
[5] B. Zhou, J. Cao, X. Zeng, and H. Wu, "Adaptive Traffic Light
Control in Wireless Sensor Network-Based Intelligent Transportation
System," in IEEE Vehicular Technology Conference, pp. 1-5, 2010.
[6] I. N. Askerzade and M. Mahmood, "Control the Extension Time of
Traffic Light in Single Junction by Using Fuzzy Logic," International
Journal of Electrical & Computer Sciences, vol. 10, no. 2, 2013.
[7] D. Garg, Mahesh Ramamoorthy, R. S. Chandel, G. M. Deshpande,
and R. N. Sharma, "Density-based Traffic Signal System Using Image
Processing," International Journal of Innovation and Scientific
Research, vol. 23, no. 2, pp. 329-337, 2016.
[8] Jocher, G., et al., "Ultralytics YOLOv8," 2023. [Online]. Available:
https://github.com/ultralytics/ultralytics.