ZONEWATCH: A DEEP LEARNING APPROACH TO VEHICLE

ZONE RECOGNITION AND SPEED MANAGEMENT

A PROJECT REPORT
Submitted by

J. ABINAYA 912420104002
B. MEENAKSHI 912420104019
S. PRAVINA 912420104025

in partial fulfilment for the award of the degree

of
BACHELOR OF ENGINEERING
IN

COMPUTER SCIENCE AND ENGINEERING

SHANMUGANATHAN ENGINEERING COLLEGE

ARASAMPATTI

ANNA UNIVERSITY :: CHENNAI 600 025

MAY 2024

1
 ANNA UNIVERSITY :: CHENNAI 600 025

BONAFIDE CERTIFICATE

Certified that this project report “ZONEWATCH: A DEEP LEARNING

APPROACH TO VEHICLE ZONE RECOGNITION AND SPEED
MANAGEMENT” is the bonafide work of “J. ABINAYA (912420104002),
B. MEENAKSHI (912420104019), S. PRAVINA (912420104025)” who carried
out the project work under my supervision.

SUPERVISOR HEAD OF THE DEPARTMENT

Mr. S. Saravanakumar, M.E., Dr. S. Saravanakumar.,
Assistant Professor, Department of Computer
Department of Computer Science and Engineering,
Science and Engineering, Shanmuganathan Engineering
Shanmuganathan Engineering College,
College, Arasampatti-622507
Arasampatti-622507.

Submitted for the viva voce to be held on _____________

INTERNAL EXAMINER EXTERNAL EXAMINER

2
 ACKNOWLEDGEMENT

"HARD WORK NEVER FAILS" So we thank god for having gracefully

blessed us to come up till now and thereby giving strength and courage to
complete the project successfully. We sincerely submit this project to the
almighty lotus feet.
We wish to acknowledge with thanks to the significant contribution given
by the management of our college chairman "KALLVI VALLAL
Mrs.Pichappa valliammai”, Correspondent Dr.P.Manikandan, and Secretary
Mr.Vishvanathan, Shanmuganathan Engineering College, Arasampatti, for their
extensive support.
We convey our indebted humble thanks to our energetic Principal
Dr.KL.Muthuramu M.E., (W.R), M.E.(S.E),M.I.S.T.E., F.I.E., Ph.D., for his
moral support to complete this project.
We are grateful to our Head of the Department Dr.S.Saravanakumar .,
for is valuable guidelines during the courses of the project.
Gratitude will never fail towards our dynamic and effective internal guide
Mr.S.Saravanakumar M.E., for is unending helps that have been provided to
us throughout the project and during code debugging.
We wish to acknowledge the department project coordinator
Mr.R.Rajapandian M.E., for is valuable, innovative suggestion, constructive
instruction, constant encouragement, during project proceedings.
We also convey our humble thanks to all the staff members of the CSE
Department who had provide their contribution to steer our project towards the
successful completion.
Finally, we thank our family members who have spent their wealth for our
studies and have motivated us to complete this project successfully.

iii
 ABSTRACT
In contemporary society, where time is a precious commodity, the
imperative to complete tasks efficiently is a pervasive human tendency. One
notable arena where this urgency manifests is in vehicular transportation.
However, the escalating density of vehicles coupled with lax adherence to traffic
regulations pose significant challenges to road safety. The prevalence of drivers
exceeding speed limits in restricted zones exacerbates the problem, leading to a
surge in road accidents. To address these concerns, we propose an innovative
solution: an Automatic Speed Control System (ASCS) for vehicles employing
video processing technology. This system aims to mitigate the risks associated
with over-speeding by detecting restricted areas and gradually reducing vehicle
speed in response. The methodology involves capturing real-time video footage
of road scenes using small cameras installed on vehicles. Through the
implementation of Convolutional Neural Network (CNN) algorithms, the system
identifies and interprets road signs denoting speed limits, school zones, and
hospital zones. Upon detection of a restricted area, the microcontroller integrated
into the vehicle initiates a process to gradually reduce vehicle speed. Additionally,
the system alerts the driver, ensuring awareness of the impending speed
adjustment. By integrating video processing technology with existing vehicle
systems, our proposed ASCS offers a proactive approach to enhancing road
safety. Through its ability to detect and respond to restricted areas in real-time,
this system has the potential to significantly reduce the incidence of over-
speeding-related accidents.

Keywords: Automatic Speed Control System, Image Processing, Convolutional

Neural Network, Road Safety, Real-time Detection.

iv
 TABLE OF CONTENT

CHAPTER NO TITLE PAGE NO

ABSTRACT iv
LIST OF FIGURES viii
LIST OF ABBERVATION ix
1. INTRODUCTION 1
1.1 Vehicle Density and Accidents 2
1.2 Contributing Factors to Road 2
Accidents
1.3 Role of Human Behaviour in 3
Road Safety
1.4 Utilization of Deep Learning 3
Techniques for Zone Detection
1.5 Understanding Artificial 4
Intelligence in Road Safety
1.6 The Significance of Speed Limits 4
in Road Safety
2 LITERATURE SURVEY 5
2.1 Double FCOS: A Two-Stage 5
Model Utilizing FCOS for Vehicle
Detection in Various Remote Sensing
Scenes
2.1.1 Merits 5
2.1.2 Demerits 5
2.2 A DVB-S-Based multichannel 6
passive radar system for vehicle
detection
2.2.1 Merits 6
2.2.2 Demerits 6

v
 2.3 VAID: An Aerial Image Dataset 7
for Vehicle Detection and
Classification
2.3.1 Merits 7
2.3.2 Demerits 7
2.4 CMNet: A Connect-and-Merge 8
Convolutional Neural Network for
Fast Vehicle Detection in Urban
Traffic Surveillance
2.4.1 Merits 8
2.4.2 Demerits 8
2.5 Part-Aware Region Proposal for 9
Vehicle Detection in High Occlusion
Environment
2.5.1 Merits 9
2.5.2 Demerits 9
3 SYSTEM ANALYSIS 10
3.1 Existing System 10
3.1.1 Limitation of Existing 11
System
3.2 Proposed System 11
3.2.1 Advantage of Proposed 12
System
3.3 System Architecture 13
3.3.1 Architecture Diagram 13
3.3.2 Block Diagram 14
3.3.3 Flowchart Diagram 15
3.4 Module Description 16
3.4.1 Image capture and 16
Recognition Module

vi
 3.4.2 Decision-Making Logic 17
Module
3.4.3 Voice Prompt Generation 18
Module
3.4.4 Integration With Pic 19
Microcontroller Module
3.4.5 Relay Drive For Speed 19
4 SYSTEMS DESIGN 20
4.1 UML Diagrams 20
4.2 Types Of UML Diagram 21
4.2.1 Usecase diagram 22
4.2.2 Class diagram 23
4.2.3 Activity diagram 24
4.2.4 State chart diagram 25
4.2.5 Deployment diagram 26
5 IMPLEMENTATION 27
5.1 Optimization and Classification 27
5.1.1 CNN Algorithm 27
5.2 System Specification 33
5.2.1 Software Requirements 33
5.2.2 Hardware Requirements 34
5.2.3 Software Description 34
5.2.4 Hardware Description 47
6 CONCLUSION AND FUTURE WORK 60
6.1 Results and Discussion 60
6.2 Conclusion 61
6.3 Future Work 61
APPENDIX I 62
APPENDIX II 70
REFERENCES 72

vii
 LIST OF FIGURES

FIGURE NO FIGURE NAME PAGE NO

3.1 Architecture Diagram 13

3.2 Block diagram 14

3.3 Flowchart Diagram 15

4.1 Usecase diagram 22

4.2 Class diagram 23

4.3 Activity diagram 24

4.4 Statechart diagram 25

4.5 Deployment diagram 26

5.1 Feature 56

5.2 Pin description 57

viii
 LIST OF ABBREVATIONS

CNN - Convolutional Neural Network

AI - Artificial Intelligence
FCOS - Fully Convolution One-Stage
CC - Cross-Correlation
MCF - Multichannel Fusion
DVB-S - Digital Video Broadcast-Satellite
VAID - Vehicle Aerial Imaging from Drone
CMRN - Connect-And-Merge Residual Network
MSPN - Multi-Scale Prediction Network
R-CNN - Region-based Convolutional Neural Network
CMDP - Constrained Markov Decision Process
PIC - Python Imaging Library
FC - Fully Connected Layer
RGB - Red, Green, and Blue
PIL - Peripheral Interface Controller
MPLAB IDE - Microchip's Integrated Development Environment

ix
 CHAPTER 1
1. INTRODUCTION

The surge in vehicle numbers has become an undeniable facet of modern

life, particularly pronounced in nations like India, where the trajectory of vehicle
density has witnessed exponential growth in recent decades. This proliferation of
vehicles, emblematic of progress and mobility, has, however, precipitated a
myriad of challenges, chief among them being the alarming escalation in road
accidents. Over the period spanning from 2001 to 2015, India experienced a
marked uptick in vehicle density, a trend that has significantly heightened
concerns surrounding road safety on a national scale.

This report undertakes the ambitious task of dissecting the multifaceted issues
underpinning the escalating rate of accidents, with a primary focus on the
pervasive problem of over-speed driving. Concurrently, the report endeavors to
present innovative solutions harnessing advanced technologies such as deep
learning and artificial intelligence (AI). By delving into the intricate interplay
between societal dynamics, infrastructural deficiencies, and human behaviour,
this report aspires to provide comprehensive insights into the root causes of the
prevailing road safety crisis. Moreover, it seeks to articulate a roadmap for
effective intervention, leveraging cutting-edge technological tools to mitigate
risks and cultivate a safer driving environment for all road users.

Furthermore, the report aims to explore the socio-economic implications of road

accidents, including the toll on public health systems, loss of human lives, and
economic burdens on families and society at large. Through rigorous analysis and
empirical evidence, it endeavours to underscore the urgency of implementing
proactive measures to address the underlying factors contributing to road
accidents. By fostering collaboration between stakeholders across government,
industry, and civil society, the report advocates for a holistic approach to road

1
safety that integrates technological innovations with policy reforms and public
awareness campaigns. Ultimately, the goal is to pave the way for a future where
road accidents are minimized, and every individual can travel safely and
confidently on the nation's roads.

1.1 Vehicle Density and Accidents

The period between 2001 and 2015 marked a significant juncture in the
trajectory of vehicular traffic on Indian roadways, emblematic of both economic
expansion and societal progress. However, this epoch of exponential growth in
vehicle density has not unfolded without its accompanying challenges. Statistical
analyses conducted during this period underscore a clear and alarming correlation
between the burgeoning number of vehicles and the escalating frequency of road
accidents, thereby casting a profound shadow over the nation's road safety
landscape. Indeed, the proliferation of vehicles has served to spotlight a myriad
of intricate issues, necessitating a concerted and multifaceted approach to
addressing the underlying causes and effectively mitigating the associated risks
posed to public safety.

1.2 Contributing Factors to Road Accidents

A confluence of factors contributes to the alarming frequency of road

accidents in India. While the exponential proliferation of vehicles undeniably
constitutes a significant factor, it represents just one facet of a complex and
multifaceted issue. Despite sporadic advancements, Indian road infrastructure,
with the notable exception of national highways, has largely remained inadequate
in keeping pace with the burgeoning vehicular population. This deficiency in
infrastructural development exacerbates the inherent risks associated with
driving, particularly within densely populated urban locales characterized by
congested thoroughfares and unpredictable traffic dynamics.The failure to
address these infrastructural shortcomings amplifies the vulnerability of road

2
users to accidents and underscores the imperative for comprehensive
interventions aimed at enhancing road safety across the nation.

1.3 Role of Human Behaviour in Road Safety

In addition to infrastructure deficiencies, human behaviour plays a pivotal

role in shaping the state of road safety. The pervasive issue of over-speed driving,
exacerbated by a plethora of distractions including mobile phone usage and
alcohol consumption, characterizes the driving culture in India. Furthermore, a
flagrant disregard for traffic laws amplifies the risks inherent on the roads, as
drivers routinely flout speed limits and engage in reckless manoeuvres. This
confluence of factors not only heightens the likelihood of accidents but also
underscores the urgent need for comprehensive interventions targeting
behavioural change and enforcement of regulations to foster a safer driving
environment.

1.4 Utilization of Deep Learning Techniques for Zone Detection

In recent times, there has been a notable surge in the utilization of deep
learning techniques, particularly Convolutional Neural Networks (CNNs), within
the domain of road safety. This trend reflects the growing recognition of AI's
potential to revolutionize traditional approaches to image analysis, classification,
and pattern recognition. By harnessing the power of AI, specifically in the realms
of image processing and feature extraction, these advanced techniques offer a
promising avenue for enhancing road safety measures. Through the analysis of
real-time images captured by strategically positioned cameras, these systems can
discern critical information regarding accident-prone zones, hospital precincts,
and school areas. This analytical prowess enables proactive identification of
potential hazards, empowering authorities to implement timely interventions and
alert drivers, thereby mitigating risks and fostering a safer driving environment
for all road users.

3
1.5 Understanding Artificial Intelligence in Road Safety

Artificial Intelligence (AI) occupies a pivotal position in the ongoing

endeavours to bolster road safety, owing to its inherent capability to emulate
human cognitive functions. Situated within the realm of computer science, AI
encompasses a diverse array of technologies meticulously designed to endow
machines with human-like perceptual, reasoning, and decision-making
capabilities. Within the specific context of road safety, AI emerges as a potent tool
with unparalleled potential for the detection and mitigation of risks. From
discerning hazardous driving behaviours to forecasting accident-prone regions,
AI facilitates proactive interventions aimed at averting potential disasters and
safeguarding lives on the road. Through its adaptive learning mechanisms and
predictive analytics, AI promises to usher in a paradigm shift in road safety
management, heralding a future characterized by enhanced efficiency, efficacy,
and ultimately, a reduction in road accidents.

1.6 The Significance of Speed Limits in Road Safety

Among the myriad factors influencing road safety, adherence to speed

limits emerges as a cornerstone principle. Traffic control systems disseminate
speed limit information through signage, delineating safe speeds for specific areas
such as residential neighbourhoods, marketplaces, school zones, and hospital
precincts. However, despite the clear delineation of speed restrictions,
compliance among drivers remains a persistent challenge, precipitating a cascade
of preventable accidents. In the subsequent sections, this report will delve deeper
into the methodologies and technologies employed in addressing these pressing
road safety concerns, elucidating how advancements in deep learning and AI can
revolutionize the landscape of vehicular transportation in India. esponders,
helping them coordinate their efforts more effectively.

4
 CHAPTER 2

2. LITERATURE SURVEY

2.1 TITLE : Double FCOS: A Two-Stage Model Utilizing FCOS for Vehicle
Detection in Various Remote Sensing Scenes
AUTHOR : Peng Gao , Tian Tian
YEAR : 2022

It proposes a vehicle detection model called double fully convolution

one-stage object detection (FCOS) and a vehicle dataset called multiscene,
multiquality, multiscale, and multiclass vehicle dataset (4MVD) for vehicle
detection in various remote sensing scenes. Double FCOS is a two-stage detection
model based on FCOS. FCOS is exploited in the RPN stage to generate candidate
boxes in various scenes. The two-stage positive and negative sample model is
carefully designed to enhance the positive proposal sampling effects, particularly
the tiny or weak vehicles, which are ignored in FCOS. A two-step classification
model has been designed in the RCNN stage with a proposal classification branch
and point classification branch to enhance the classification performance between
the various types of vehicles. 4MVD is collected from various remote sensing
scenes to evaluate the performance of double FCOS.

2.1.1 MERITS
• Improved detection accuracy, especially for tiny or weak vehicles, through
the carefully designed two-stage detection process.
2.1.2 DEMERITS
• Potentially increased computational complexity compared to single-stage
detection models, due to the additional stages and classification branches.

5
2.2 TITLE : a DVB-S-Based multichannel passive radar system for vehicle
detection
AUTHOR : Junjie Li Junkang Wei
YEAR : 2021

In the developed system, cross-correlation (CC) operation is applied to

obtain suboptimal detection gain and computational complexity. To decrease the
impact of interference that is independently and identically distributed (i.i.d.)
across isolated frequency channels and that of the Doppler effect produced by a
mobile vehicle, a multichan- nel joint detection scheme based on multichannel
fusion (MCF) is proposed and implemented. Furthermore, to achieve high
resolution, an illuminator of opportunity (IO) with wide bandwidth is selected,
and the system is then designed to receive a digital video broadcast-satellite
(DVB-S) signal with a bandwidth of 37.5 MHz over multiple channels
simultaneously. Simulation results show that the proposed MCF achieves better
performance than the single-channel-based CC, and the advantage can be
enhanced by increasing the number of channels for joint detection. Field test
results further verify the novelty of the MCF and the developed passive radar
prototype.

2.2.1 MERITS
• Simple implementation and low computational complexity compared to
multichannel techniques.
2.2.2 DEMERITS
• Limited detection gain and susceptibility to interference, particularly in
environments with high levels of independently and identically distributed
(i.i.d.) interference and Doppler effects.

6
2.3 TITLE : VAID: An Aerial Image Dataset for Vehicle Detection and
Classification
AUTHOR : Huei-Yung Lin
YEAR : 2020

In this study, we introduce a novel aerial image dataset, VAID (Vehicle

Aerial Imaging from Drone), tailored specifically for the advancement and
assessment of vehicle detection algorithms. Comprising approximately 6000

meticulously annotated images captured under diverse traffic conditions, VAID

serves as a comprehensive resource for network training and testing. These
images are meticulously annotated with seven common vehicle categories,
ensuring a robust dataset for algorithm evaluation and comparison.Furthermore,
we conduct an in-depth analysis by benchmarking the performance of state-of-
the-art network architectures against various existing aerial image datasets. By
juxtaposing the detection results obtained using VAID with those derived from
other datasets, we aim to provide valuable insights into the efficacy and
generalizability of different detection algorithms across diverse aerial imaging
scenarios. Through this comparative study, we endeavor to facilitate informed
decision-making and foster advancements in the field of aerial vehicle detection.

2.3.1 MERITS
• VAID dataset offers a rich collection of aerial images with diverse traffic
conditions and annotated vehicle categories, facilitating robust algorithm
development and evaluation.
2.3.2 DEMERITS
• Limited to aerial imagery, VAID may not fully capture the complexities of
vehicle detection in ground-level scenarios, potentially impacting
algorithm generalization.

7
2.4 TITLE : CMNet: A Connect-and-Merge Convolutional Neural Network
for Fast Vehicle Detection in Urban Traffic Surveillance
AUTHOR : Fukai Zhang
YEAR : 2019

It propose a connect-and-merge residual network (CMRN) for performin

feature extraction. Speci_cally, the CMRN assembles residual branches in
parallel through a connectand- merge mapping: Connect the input to the outputs
of two residual branches separately (Connect), and merge the outputs of the
connection as the input of the subsequent residual block (Merge), respectively.
Second, we present a multi-scale prediction network (MSPN) to accurately
regress the vehicle shape and classify vehicle _ne-grained categories. In addition,
the feature maps from the CMRN are merged with their corresponding upsampled
features from the MSPN using concatenation. Thus, we can improve the
information _ow of the framework and make it easy to train. The experimental
results on the KITTI dataset and the UA-DETRAC dataset demonstrate that the
CMNet can achieve ef_cient detection performance in terms of accuracy and
speed for the real-world traf_c surveillance data.

2.4.1 MERITS
• The CMNet architecture offers efficient feature extraction through
connect-and-merge residual branches, enhancing detection accuracy and
speed for real-world traffic surveillance data.
2.4.2 DEMERITS
• The complexity of the CMNet architecture may require significant
computational resources, potentially limiting its applicability on resource-
constrained platforms.

8
2.5 TITLE : Part-Aware Region Proposal for Vehicle Detection in High
Occlusion Environment
AUTHOR : Weiwei Zhang
YEAR : 2019

This paper presents a two-stage detector based on Faster R-CNN for high
occluded vehicle detection, in which we integrate a part-aware region proposal
network to sense global and local visual knowledge among different vehicle
attributes. That entails the model simultaneously generating partial-level
proposals and instance-level proposals at the _rst stage. Then, different parts
belong to the same vehicle are encoded and recon_gured into a compositional
entire proposal through a part af_nity _elds, allowing the model to generate
integral candidates and mitigate the impact of occlusion challenge to the utmost
extent. Extensive experiments conducted on KITTI benchmark exhibit that our
method outperforms most machine-learning-based vehicle detection methods and
achieves high recall in the severely occluded application scenario.

2.5.1 MERITS
• The integration of a part-aware region proposal network enables the model
to generate both partial-level and instance-level proposals simultaneously,
enhancing detection accuracy and robustness, especially in highly
occluded scenarios.
2.5.2 DEMERITS
• The complexity introduced by the part-aware region proposal network may
increase computational overhead, potentially impacting real-time
performance on resource-constrained platforms.

9
 CHAPTER 3
3. SYSTEM ANALYSIS

3.1 EXISTING SYSTEM

Agent-based approaches have garnered considerable attention in

engineering applications, given their potential to model complex systems with
decentralized decision-making. However, their application to advanced traffic
control systems remains relatively nascent. This paper addresses this gap by
presenting a sophisticated multi-agent framework that effectively models various
traffic control instruments and their interactions with road traffic dynamics.
Central to the proposed framework is the utilization of a constrained Markov
decision process (CMDP) model, which enables agents to make decisions
considering multiple conflicting objectives. By transforming these objectives into
a single optimization objective and treating others as constraints, the CMDP
framework provides a systematic approach to addressing multi-objective
optimization problems in traffic control.

Moreover, the paper introduces a reinforcement learning-based

computational framework tailored for control applications. This framework
leverages the CMDP model to learn optimal control policies in a dynamic traffic
environment, thereby facilitating adaptive and responsive traffic management
strategies.To operationalize the multi-objective decision model, a threshold
lexicographic ordering method is introduced. This method allows for the
prioritization of policy goals, ensuring that higher-priority objectives are met
while satisfying constraints imposed by lower-priority goals.

Overall, the integration of agent-based modelling , CMDP, reinforcement

learning, and threshold lexicographic ordering presents a comprehensive and
innovative approach to advanced traffic control, with the potential to significantly
improve traffic flow efficiency, safety, and environmental sustainability.
10
3.1.1 LIMITATION OF EXISTING SYSTEM

• Limited real-world validation: The evaluation of the intelligent control

approach was primarily conducted through simulation studies,
potentially limiting the generalizability of the findings to real-world
traffic scenarios.
• Lack of comprehensive comparison: While the performance of the
intelligent control approach was compared with several other signal
control methods, the scope of comparison may not have been
exhaustive, potentially overlooking alternative approaches with
superior efficacy.
• Ineffectiveness in accident prevention: The existing system's failure to
prevent vehicle accidents in restricted zones highlights a critical
limitation, as ensuring safety in such areas is paramount for mitigating
potential harm to both motorists and pedestrians.

3.2 PROPOSED SYSTEM

The proposed system for accident, school, and hospital zone detection,
utilizing Convolutional Neural Networks (CNN), is designed to enhance road
safety by accurately pinpointing areas prone to accidents. Through the integration
of cameras and sensors, the system captures images or video footage of targeted
areas, which are then analysed by CNN to identify and categorize accident,
school, and hospital zones. By training the CNN on extensive datasets comprising
images of these specific zones, the system becomes adept at detecting accidents
swiftly.

The primary objective of the system is to furnish real-time information to drivers,

enabling them to exercise caution and reduce speed when traversing accident,
hospital, or school zones.
11
Furthermore, the system's integration with existing traffic management systems
facilitates the dissemination of real-time data on accident density, thereby
facilitating proactive measures to optimize traffic flow and diminish the
likelihood of accidents.

Incorporating machine learning techniques, the system automatically detects

school and college zones through camera inputs and adjusts vehicle speed
accordingly. Data pertaining to the identified zones is transmitted to the
microcontroller, which enforces the appropriate speed limit for each designated
area. Continuously monitoring the surroundings, the system ensures ongoing
vigilance and responsiveness to evolving road conditions.

3.2.1 ADVANTAGE OF PROPOSED SYSTEM

• Accurate identification of accident, school, and hospital zones

enhances road safety.
• Real-time information provision enables drivers to make informed
decisions.
• Integration with traffic management systems optimizes traffic flow
and reduces congestion.
• Automated speed control in school and college zones ensures
compliance with speed limits.
• Continuous monitoring ensures ongoing vigilance and responsiveness
to changing road conditions.
• Enhanced safety measures mitigate the risk of accidents and promote
safer driving practices.
• Swift detection of accidents contributes to prompt response and
prevention strategies.

12
3.3 SYSTEM ARCHITECTURE

The system architecture employs Convolutional Neural Networks

(CNNs) to identify accident-prone areas, school zones, and hospital zones
from captured images. Cameras and sensors capture targeted areas,
processed by CNNs to classify zones. Real-time information is relayed to
drivers, prompting caution and reduced speed in designated zones.
Integration with traffic management systems optimizes traffic flow and
accident prevention. Machine learning enables automatic speed reduction in
school and college zones. Data from detections are sent to a microcontroller
to determine zone-specific speed limits. Continuous monitoring ensures
ongoing effectiveness in maintaining road safety standards.

3.3.1 ARCHITECTURE DIAGRAM

Fig 3.1. Architecture diagram

13
3.3.2 BLOCK DIAGRAM

The system employs cameras and sensors to capture information about the
surrounding environment, which is then processed by Convolutional Neural
Networks (CNNs). These CNNs analyse the data to identify accident-prone areas,
school zones, and hospital zones. The system then relays real-time information to
drivers, prompting them to exercise caution and reduce speed in designated zones.
Additionally, the system integrates with traffic management systems to optimize
traffic flow and prevent accidents. By automatically adjusting vehicle speed in
school and college zones based on the detected data, the system contributes to
maintaining road safety standards. Continuous monitoring ensures that the system
remains effective in its operation.

Fig 3.2. Block diagram

14
3.3.3 FLOWCHART DIAGRAM

The flowchart illustrates the step-by-step process of problem-solving. It

starts with problem identification and requirement gathering, followed by
solution design and implementation. Testing is then conducted to ensure the
solution meets the requirements. If issues arise, debugging and retesting are
performed until the solution is ready for deployment. Finally, the solution is
deployed, marking the completion of the process.

Fig 3.3.Flowchart diagram

15
3.4 MODULES USED
1. Image Capture And Recognition Module
2. Decision-Making Logic Module
3. Voice Prompt Generation Module
4. Integration With Pic Microcontroller Module
5. Relay Drive For Speed Control Module

MODULE DESCRIPTION

3.4.1 IMAGE CAPTURE AND RECOGNITION MODULE

The Image Capture and Recognition Module serves as the foundational

component for gathering visual data and identifying critical features on the road.
It utilizes cameras to capture images of the surrounding environment, which are
then processed using image processing techniques, notably Convolutional Neural
Networks (CNNs). These CNNs analyse the captured images to recognize
specific features such as accident-prone areas, school zones, and hospital zones.
By leveraging advanced algorithms, this module enables the system to extract
relevant information from the captured images, providing crucial insights for
subsequent decision-making processes regarding speed control and safety alerts.
DATASET
Building a high-quality image dataset is integral for various computer vision
tasks, demanding meticulous preparation and execution. Collection methods
range from manual capture to accessing publicly available datasets, yet creating
one from scratch entails significant time and resources. Careful consideration of
dataset architecture, ethical implications, breadth, size, and labeling requirements
is essential. Ensuring data accuracy through validation and quality control
techniques is paramount. In summary, the process of gathering image datasets for
computer vision necessitates thorough planning, execution, and rigorous quality
assuranceprotocal.

16
3.4.2 DECISION-MAKING LOGIC MODULE

The Decision-Making Logic Module serves as the core intelligence behind

the system, responsible for interpreting the data collected by the Image Capture
and Recognition Module and making informed decisions to ensure road safety.
17
Upon receiving information about detected zones such as accident-prone areas,
school zones, and hospital zones, this module employs predefined rules and logic
to assess the current situation. It analyses factors such as the type and severity of
detected zones, traffic conditions, and vehicle speed to determine appropriate
actions. For instance, if an accident-prone area is identified, the module may
trigger a safety alert to prompt drivers to reduce speed and exercise caution.
Similarly, in school zones, it may enforce lower speed limits to ensure the safety
of pedestrians and students. The decision-making process is dynamic and
adaptable, allowing the system to respond effectively to changing road conditions
in real-time. By integrating advanced algorithms and logic, this module plays a
crucial role in optimizing traffic flow, preventing accidents, and enhancing
overall road safety.

3.4 .3 VOICE PROMPT GENERATION MODULE

The Voice Prompt Generation Module is responsible for providing real-

time communication with drivers by generating audible prompts or alerts based
on the decisions made by the system. Upon receiving instructions from the
Decision-Making Logic Module, this module converts the decision outputs into
spoken messages that can be easily understood by drivers. These messages
typically include information about speed limits, cautionary warnings, or
upcoming zones such as school zones or accident-prone areas.
By utilizing voice prompts, the module ensures that drivers receive timely and
clear instructions while maintaining their focus on the road. This proactive
approach enhances driver awareness and responsiveness, contributing to overall
road safety.

18
3.4 .4 INTEGRATION WITH PIC MICROCONTROLLER MODULE
The Integration with PIC Microcontroller Module acts as the intermediary
between the Decision-Making Logic Module and the PIC microcontroller
responsible for vehicle speed control. This module enables seamless
communication by transmitting instructions and relevant data from the decision-
making logic to the PIC microcontroller. These instructions include commands
for implementing speed control measures based on the detected zones, such as
adjusting vehicle speed in accident-prone areas or school zones.By facilitating
this communication, the module ensures that the PIC microcontroller receives
accurate and timely information to effectively regulate vehicle speed in response
to changing road conditions. This integration enhances the system's ability to
enforce safety measures and optimize traffic flow, ultimately contributing to
improved road safety outcomes.

3.4 .4 RELAY DRIVE FOR SPEED CONTROL MODULE

The Relay Drive for Speed Control Module plays a crucial role in translating
electronic signals from the PIC microcontroller into mechanical actions that
control vehicle speed. Upon receiving signals from the PIC microcontroller,
indicating the need for speed adjustment based on detected zones and safety
requirements, this module activates the appropriate relays. These relays, in turn,
regulate the flow of power to the vehicle's engine or braking system, thereby
adjusting its speed accordingly. By effectively interfacing with the PIC
microcontroller and activating relays in response to real-time data from the
system, this module ensures precise and responsive speed control. This capability
enables the system to enforce safety measures, such as reducing speed in
accident-prone areas or school zones, contributing to enhanced road safety and
traffic management.

19
 CHAPTER 4

4 SYSTEMS DESIGN

The system design phase is a pivotal step in any project, laying the
groundwork for its architecture and functionality. It begins with a comprehensive
analysis of requirements to ensure alignment with user needs. Architectural
blueprints are then crafted, outlining the structure and relationships of system
components. Detailed designs for each element are developed, clarifying
functionality and inter-component interfaces. Data flow pathways are
meticulously mapped to ensure efficient handling throughout the system.
Technology selection is critical, with tools chosen to best implement each
component's requirements. Integration planning is paramount to ensure seamless
operation and data exchange among components. Rigorous testing strategies are
devised to validate functionality and performance. Deployment, maintenance,
and support plans are established to ensure successful implementation and
ongoing operation. Ultimately, the system design process ensures that the
proposed solution is robust, scalable, and capable of meeting user expectations.

4.1 UML DIAGRAM

Unified Modeling Language (UML) diagrams serve as visual

representations of system structures, behaviors, and interactions, aiding in the
understanding and communication of complex systems. These diagrams are
instrumental throughout the software development lifecycle, from initial design
to implementation and maintenance.In the context of system design, UML
diagrams offer a graphical depiction of various aspects of the system, such as its
architecture, components, interactions, and behaviors. For instance, a class
diagram illustrates the structure of the system by showing classes, their attributes,
methods, and relationships. This helps stakeholders understand the organization
of the system's elements and how they interact with each other.
20
Additionally, sequence diagrams showcase the interactions between objects or
components over time, portraying the flow of messages or actions between them.
This aids in visualizing the dynamic behavior of the system and identifying
potential bottlenecks or inefficiencies in its operation.

Furthermore, use case diagrams highlight the system's functionalities from the
perspective of its users, illustrating the different use cases and the actors involved
in each scenario. This provides stakeholders with a clear understanding of the
system's capabilities and how users interact with it.

Overall, UML diagrams play a vital role in system design by providing a visual
representation of its structure, behaviour, and functionalities, facilitating
communication, collaboration, and decision-making among stakeholders
throughout the development process.

4.2 TYPES OF UML DIAGRAM

• CLASS DIAGRAM
• USECASE DIAGRAM
• ACTIVITY DIAGRAM
• STATECHART DIAGRAM
• DEPLOYMENT DIAGRAM

21
4.2.1 USECASE DIAGRAM

Use case diagrams offer a simplified overview of a system's functionalities

and user interactions. Actors represent the roles users or external systems play in
the system, while use cases depict the specific tasks or goals users can
accomplish. Relationships between actors and use cases illustrate how users
interact with the system to achieve these objectives. Overall, use case diagrams
serve as a useful communication tool for stakeholders to understand the system's
behaviour and requirements.

Fig 4.1. Usecase diagram

22
4.2.2 CLASS DIAGRAM

A class diagram represents the structure of a system by showing the classes

of objects, their attributes, and relationships. Each class is depicted as a box with
three compartments: the top contains the class name, the middle lists its attributes,
and the bottom displays its methods. Arrows indicate associations between
classes, such as one-to-one or one-to-many relationships. Inheritance is
represented by an arrow with a solid line pointing from the subclass to the
superclass.

Fig 4.2. Class diagram

23
4.2.3 ACTIVITY DIAGRAM

Activity diagrams visually represent the sequential flow of activities within

a system. They use symbols such as actions, decisions, and transitions to illustrate
different elements of the workflow. Each activity represents a specific task, while
transitions indicate the flow of control between activities. Decision points allow
for branching based on conditions or choices, aiding in the analysis and
understanding of system processes.

Fig 4.3. Activity diagram

24
4.2.4 STATECHART DIAGRAM

Statechart diagrams depict the various states of an object or system and the
transitions between these states in response to events. States represent different
conditions or modes in which the system can exist. Transitions depict the
movement from one state to another triggered by specific events or conditions.
Initial and final states mark the start and end points of the diagram, respectively.
Actions associated with transitions denote the behavior or activities performed
when transitioning between states. Statechart diagrams aid in modeling complex
systems, capturing their dynamic behavior and interaction over time. They are
valuable tools in software engineering for designing and understanding stateful
systems and their behavior under different scenarios.

Fig 4.4. Statechart diagram

25
4.2.4 DEPLOYMENT DIAGRAM

A deployment diagram visually represents the physical deployment of

software components in a system. It shows how software artifacts are deployed
on hardware nodes such as servers, PCs, or other devices. Each component is
mapped to a specific node, indicating where it runs. Relationships between nodes
and components are depicted, clarifying the system's architecture. Deployment
diagrams aid in understanding system deployment configurations and optimizing
resource allocation.

Fig 4.5. Deployment diagram

26
 CHAPTER 5

5 IMPLEMENTATION

5.1 OPTIMIZATION AND CLASSIFICATION

Optimization and classification are essential processes in various fields,

including machine learning, engineering, and operations research. Optimization
involves finding the best solution from a set of possible alternatives, typically
with the goal of maximizing efficiency or minimizing costs. Classification, on the
other hand, involves categorizing data into distinct classes or groups based on
specific criteria or features. In optimization, algorithms such as linear
programming, genetic algorithms, or gradient descent are used to iteratively
improve solutions until an optimal solution is reached. In classification, machine
learning algorithms like decision trees, support vector machines, or neural
networks are employed to categorize data points based on patterns or features.
Both processes play crucial roles in problem-solving, decision-making, and
pattern recognition across various domains, contributing to improved efficiency,
accuracy, and understanding of complex systems and datasets.

5.1.1 CNN ALGORITHM

One of the most well-known ANN is the convolutional neural network. The
domains of image and video recognition make extensive use of it. It is founded
on the mathematical idea of convolution. It closely resembles a multi-layer
perceptron, however before the completely linked hidden neuron layer, it has a
succession of convolution layers and pooling layers.

It has three important layers:

• Convolution layer: It is the primary building block and perform computational

tasks based on convolution function.
27
• Pooling layer: It is arranged next to convolution layer and is used to reduce the
size of inputs by removing unnecessary information so computation can be
performed faster.

• Fully connected layer: It is arranged to next to series of convolution and

pooling layer and classify input into various categories.

A simple CNN can be represented as below:

Here,

• 2 series of Convolution and pooling layer is used and it receives and process
the input (e.g. image).

• A single fully connected layer is used and it is used to output the data (e.g.
classification of image)

CNN, unlike other neural networks, processes input as a two-dimensional array

and addresses recognition challenges. Leading companies like Google and
Facebook invest in research and development to enhance recognition capabilities
for faster task completion. CNN utilizes geographical correlations in input data,

28
where each layer's neurons analyze specific areas called "local receptive fields"
without considering changes beyond these borders. Convolutions occur as
connections pass weights from one layer to another, enabling feature extraction.
Pooling layers, following CNN declarations, condense feature maps generated by
convolutional networks, facilitating the creation of new layers with neurons from
earlier ones.

A convolutional neural network uses three basic ideas:

• Local respective fields

• Convolution

• Pooling

A particular kind of artificial neural network called a convolutional neural

network (CNN) analyses data using supervised learning with perceptron's, a
machine learning unit technique. CNNs can be used for processing images,
natural language, and other cognitive tasks. An input layer, an output layer, and
many hidden layers make up a convolutional neural network.

layers. Convolutional layers use a mathematical model to transmit results

to next layers.

• The input will contain the image's raw pixel values and its three colour
channels, R, G, and B.

• CONV layer will compute the output of neurons that are connected to
local regions in the input, each computing a dot product between their weights
and a small region they are connected to in the input volume.

• RELU layer will apply an element wise activation function. This leaves
the size of the volume unchanged.

29
 • The POOL layer conducts downsampling across spatial dimensions,
yielding a volume like [16x16x12]. The FC (fully-connected) layer calculates
class scores, resulting in a volume of a specific size. Similar to regular Neural
Networks, each neuron in this layer connects to all numbers in the preceding
volume.

Layers in CNN

There are five different layers in CNN

➢ Input layer
➢ Convo layer (Convo + ReLU)
➢ Pooling layer
➢ Fully connected(FC) layer
➢ Softmax/logistic layer
➢ Output layer

Different layers of CNN

I. Input Layer

Image data in CNN requires transformation into a single column from the
original three-dimensional matrix representation, such as converting a 28x28
image into a 784x1 array before input.
30
Convo Layer

Due to the fact that characteristics of the picture are extracted within this layer,
this layer is also known as feature extractor layer. In order to perform the
convolution operation we observed before and calculate the dot product between
the receptive field—a local area of the input image that has the same size as the
filter—and the filter, a portion of the image is first connected to the Convo layer.
One integer representing the output volume is the operation's output. Next, via a
Stride, we move the filter over the following receptive area of the identical input
picture and repeat the process. Until we have completed the entire image, the
same procedure will be repeated. Input will become the output. for the next layer.
Convo layer also contains ReLU activation to make all negative value to zero.

Pooling Layer

Pooling layers provide several benefits to CNNs. Firstly, they help in achieving
translation invariance, meaning that the network can recognize patterns
regardless of their position in the input image. Additionally, pooling layers
contribute to feature learning by extracting the most relevant information from
the input feature maps. Finally, pooling layers aid in reducing overfitting by
discarding unnecessary information and promoting generalization.

31
There is no parameter in pooling layer but it has two hyper parameters — Filter(F)
and Stride(S).

In general, if we have input dimension W1 x H1 x D1, then

W2 = (W1−F)/S+1

H2 = (H1−F)/S+1

D2 = D1

Where W2, H2 and D2 are the width, height and depth of output.

Fully Connected Layer (FC)

Fully connected layer involves weights, biases, and neurons. It connects neurons
in one layer to neurons in another layer. It is used to classify images between
different categories by training.

Softmax / Logistic Layer

Softmax or Logistic layer is the last layer of CNN. It resides at the end of FC
layer. Logistic is used for binary classification and softmax is for multi-
classification.
32
Output Layer

Output layer contains the label which is in the form of one-hot encoded.

5.2 SYSTEM SPECIFICATION

The system specifications outline the technical requirements and capabilities

of the proposed solution. It includes details such as hardware and software
requirements, performance metrics, scalability, and compatibility with existing
systems. Additionally, system specifications define user interface elements, data
processing capabilities, security features, and any regulatory compliance
considerations. These specifications serve as a blueprint for the development and
implementation of the system, ensuring that it meets the needs of stakeholders
and operates effectively within its intended environment.

5.2.1 SOFTWARE REQUIREMENTS

• Open CV – Image Processing

• Python Programming Language

• CNN

• Embedded C language

33
5.2.2 HARDWARE REQUIREMENTS

• Power supply

• PIC(16F877A) Microcontroller

• Relay driver

• DC motor

• CAMERA

5.2.3 SOFTWARE DESCRIPTION

➢ PYTHON
➢ TENSORFLOW
➢ KERAS
➢ NUMPY
➢ PILLOW
➢ SCIPY
➢ OPENCV
➢ CONVOLUTIONAL NEURAL NETWORK (CNN)
➢ MPLAB IDE

PYTHON

Python is a popular programming language. It was created by Guido van

Rossum, and released in 1991.

It is used for:

• Web development (server-side),

• Software development,

• Mathematics,

• System scripting.
34
PYTHON NUMPY

Our Python NumPy Tutorial explains both the fundamental and more
complex NumPy principles. Both professionals and beginners can benefit from
our NumPy tutorial. Numeric Python, sometimes known as NumPy, is a Python
library for computing and processing multidimensional and linear array elements.

The use of NumPy for data analysis has the following benefits.

1. NumPy performs array-oriented computing.

2. It efficiently implements the multidimensional arrays.
3. It performs scientific computations.
4. It is capable of performing Fourier Transform and reshaping the data stored
in multidimensional arrays.
5. NumPy provides the in-built functions for linear algebra and random
number generation.

OPEN CV

OpenCV tutorial explains both fundamental and advanced concepts. Both

experts and amateurs can benefit from our Open CV course. An open-source
computer vision library is called OpenCV. It gives the system the ability to
identify faces or other objects. Using the Python programming language, we will
master the fundamentals of OpenCV in this lesson. The entire range of topics
covered in our OpenCV tutorial is covered, including Read and Save Image,

35
Canny Edge Detection, Template Matching, Blob Detection, Contour, Mouse
Event, Gaussian Blur, and others.

There are a two main task which are defined below:

o Object Classification - In the object classification, we train a model on a

dataset of particular objects, and the model classifies new objects as
belonging to one or more of your training categories.
o Object Identification - In the object identification, our model will identify
a particular instance of an object - for example, parsing two faces in an
image and tagging one as Virat Kohli and other one as Rohit Sharma.

How does computer recognize the image?

A lot of information is provided by human eyes based on what they observe.

Machines are given the ability to observe everything, translate that vision into
numbers, and store those numbers in memory. Here, the issue of how computers
translate visuals into numbers arises. The pixel value is utilised to translate
images into numbers, which is the answer. The smallest component of a digital
picture or graphic that can be shown on a digital display device is a pixel.

36
The picture intensity at the particular location is represented by the numbers. In
the above image, we have shown the pixel values for a grayscale image consist
of only one value, the intensity of the black color at that location.

There are two common ways to identify the images:

1. Grayscale

Images that solely have the two hues black and white are known as
grayscale images. Black is considered to have the lowest intensity whereas white
has the highest, according to the contrast assessment of intensity. The computer
gives each pixel in the grayscale image a value dependent on how dark it is when
we use it.

2. RGB

An RGB is a mixture of the colors red, green, and blue that results in a new
color. Each pixel's value is extracted by the computer, which then organizes the
information into an array for interpretation.

37
Installation of the OpenCV

first step is to download the latest Anaconda graphic installer for Windows
from it official site. Choose your bit graphical installer. You are suggested to
install 3.7 working with Python 3.

Install OpenCV in the Windows via pip

OpenCV is a Python library so it is necessary to install Python in the system

and install OpenCV using pip command:

pip install opencv-python

38
Open the command prompt and type the following code to check if the OpenCV
is installed or not.

Scikit-learn :

Scikit-learn, a Python library for machine learning, offers a comprehensive set of

tools for statistical modeling and various machine learning tasks, including
classification, regression, clustering, and dimensionality reduction, making it
accessible to programmers of all levels.

PANDAS

For the purpose of manipulating and analysing data, the Python

programming language has a software package called pandas. It includes specific
data structures and procedures for working with time series and mathematical
tables. It is free software distributed under the BSD license's three clauses. [2]
The word is derived from "panel data," a phrase used in econometrics to refer to
data sets that contain observations for the same persons throughout a range of
time periods.

39
Library feature :

Data manipulation tools offer a range of functionalities, including reading

and writing data between different file formats, handling missing data, reshaping
and pivoting datasets, and subset operations using indexing and slicing.
Additionally, these tools enable inserting and deleting columns, grouping and
combining datasets, and joining and merging data. They also support working
with high-dimensional data via hierarchical indexing and provide time-series
capabilities such as date shifting, moving window statistics, and frequency
conversion.

CAMERA :

• A webcam is a hardware camera and input device that connects to a

computer and the Internet and captures either still pictures or motion
video of a driver drowsiness.
• A webcam is a video camera that feeds or streams an image or video in
real time to or through a computer to a computer network, such as the
Internet. Webcams are typically small cameras that sit on a desk, attach to
a user's monitor, or are built into the hardware
• Webcams are usually cheaper than a standard video camera and allow for
face-to-face communication online, making it easy to illustrate things
visually to the person you are talking to. This makes the webcam a very
versatile device for home or office use.

40
TENSORFLOW- INTRODUCTION

The Google team created the software library or framework TensorFlow to

make machine learning and deep learning principles easy to deploy. For the quick
calculation of numerous mathematical equations, it combines computational
algebra and optimization techniques. The official website of TensorFlow is
mentioned below.

TensorFlow — Installation

To install TensorFlow, it is important to have “Python” installed in your

system. Python version 3.4+ is considered the best to start with TensorFlow
installation. Consider the following steps to install TensorFlow in Windows
operating system.

pip install tensorflow

41
Following are the two TensorFlow — Convolutional Neural NetworksAfter
understanding machine-learning concepts, we can now shift our focus to deep
learning concepts. Deep learning is a division of machine learning and is
considered as a crucial step taken by researchers in recent decades. The examples
of deep learning implementation include applications like image recognition and
speech recognition.

important types of deep neural networks:

• Convolutional Neural Networks

• Recurrent Neural Networks In this chapter, we will focus on the CNN.

METHODOLOGY:

CNNs consist of convolutional layers, pooling layers, ReLU layers, and fully
connected layers, forming two main subnetworks: convolutional layers for
feature extraction and a deep neural network. Pooling layers, such as max
pooling, downscale the image by determining the maximum value in a region.
Fully connected layers, akin to traditional neural network hidden layers, require
consideration of the number of outputs during training and can utilize transfer
learning techniques.

Pre trained models for computer vision

The pre-trained model is essential for the usage of transfer learning. Several
neural network architectures have been built by the deep learning community and
are available in the Python and Keras libraries as well as the MATLAB deep
42
learning toolbox. For image classification and and object detection the most
popular and state of the are are the following:

• AlexNet

• VGG-16

• VGG-19

• Inception V3

• XCeption

• ResNet-50

The differences among these networks are mainly the network architecture (the
number and distribution and parameters of the convolution, pooling, and dropout,
and fully connected layers ). And so, the overall number of trainable parameters,
the computatonal complexity of the training process and attained performance.

PYTHON PILLOW ― OVERVIEW

In the realm of Python image processing, popular libraries like Scikit-

image, OpenCV, PIL, and Pillow are widely used. Among them, Pillow stands
out for its extensive features and compatibility with all major operating systems,
43
including Python 3. Originally built on top of PIL, which is no longer supported,
Pillow offers support for a wide range of image formats such as JPEG, PNG,
BMP, GIF, PPM, and TIFF. With Pillow, users can perform various image
processing tasks, including point operations, color space conversions, and
convolution kernel-based filtering.

IMAGE ARCHIVES

The Python Imaging Library is best suited for image archival and batch
processing applications. Python pillow package can be used for creating
thumbnails, converting from one format to another and print images, etc.

IMAGE DISPLAY

The Tk PhotoImage, BitmapImage, and Windows DIB interfaces, which

are compatible with PythonWin and other Windows-based toolkits as well as
many other Graphical User Interface (GUI) toolkits, can be used to show
images.There is a show () function to save the image to disc for debugging
reasons, which invokes the external display tool.

Image Processing

All of the fundamental image processing capabilities are available in the

Pillow library. You can modify, rotate, and resize images.You can extract some
statistics from an image using the Pillow module's histogram function, and then
use those statistics for automatic contrast amplification and statistical analysis.

PYTHON PILLOW — ENVIRONMENT SETUP

This chapter discusses how to install pillow package in your computer.

Installing pillow package is very easy, especially if you’re installing it using pip.

44
3.Installing Pillow using pip

To install pillow using pip, just run the below command in your command
prompt:

python -m pip install pip

python -m pip install pillow

In case, if pip and pillow are already installed in your computer, above commands
will simply mention the ‘requirement already satisfied’ as shown below:

PYTHON PILLOW — USING IMAGE MODULE

To display the image, pillow library is using an image class within it. The
image module inside pillow package contains some important inbuilt functions
like, load images or create new images, etc.

Example

Following example demonstrates the rotation of an image using python pillow:

from PIL import Image

#Open image using Image module

im = Image.open("images/cuba.jpg")

#Show actual Image im.show()

#Show rotated Image

im = im.rotate(45) im.show()

45
MPLAB IDE SOFTWARE
MPLAB is aproprietary freeware integrated development environment for
the development of embedded applications on PIC and dsPIC microcontrollers,
and is developed by Microchip Technology.

MPLAB X is the latest edition of MPLAB, and is developed on

the NetBeans platform. MPLAB and MPLAB X support project management,
code editing, debugging and programming of Microchip 8-bit, 16-bit and 32-bit
PIC microcontrollers.

MPLAB is designed to work with MPLAB-certified devices such as the MPLAB

ICD 3 and MPLAB REAL ICE, for programming and debugging PIC
microcontrollers using a personal computer. PICKit programmers are also
supported by MPLAB.

MPLAB 8.X is the last version of the legacy MPLAB IDE technology,
custom built by Microchip Technology in Microsoft Visual C++. MPLAB
supports project management, editing, debugging and programming of Microchip
8-bit, 16-bit and 32-bit PICmicrocontrollers. MPLAB only works on Microsoft
Windows. MPLAB is still available from Microchip's archives, but is not
recommended for new projects.

MPLAB supports the following compilers:

• MPLAB MPASM Assembler

• MPLAB ASM30 Assembler
• MPLAB C Compiler for PIC18
• MPLAB C Compiler for PIC24 and dsPIC DSCs
• MPLAB C Compiler for PIC32
• HI-TECH C

MPLAB X is the latest version of the MPLAB IDE built by Microchip

Technology, and is based on the open-source NetBeans platform.
46
MPLAB X is the first version of the IDE to include cross-platform support
for Mac OS X and Linux operating systems, in addition to Microsoft Windows.

MPLAB X supports the following compilers:

• MPLAB XC8 — C compiler for 8-bit PIC devices

• MPLAB XC16 — C compiler for 16-bit PIC devices
• MPLAB XC32 — C/C++ compiler for 32-bit PIC devices
• HI-TECH C — C compiler for 8-bit PIC devices
• SDCC — open-source C compiler
• HI-TECH C compiler for PIC10/12/16 MCUs (PRO)
This compiler has been discontinued and is no longer supported. This
compiler has been replaced by the MPLAB® XC8 PRO (SW006021-2).
HI-TECH C Compiler for PIC10/12/16 MCUs - PRO fully implements the
optimizations of Omniscient Code Generation™ - a whole-program compilation
technology - to provide denser code and better performance on PIC MCUs.

5.2.4 HARDWARE DESCRIPTION

➢ POWER SUPPLY
➢ PIC MICRO CONROLLER
➢ DRIVER RELAY
➢ DC MOTOR

POWER SUPPLY

Power supply is a reference to a source of electrical power. A device or

system that supplies electrical or other types of energy to an output load or group
of loads is called a power supply unit or PSU. The term is most commonly applied
to electrical energy supplies, less often to mechanical ones, and rarely to others.
Power supplies for electronic devices can be broadly divided into linear and
switching power supplies. The linear supply is a relatively simple design that
47
becomes increasingly bulky and heavy for high current devices; voltage
regulation in a linear supply can result in low efficiency. A switched-mode supply
of the same rating as a linear supply will be smaller, is usually more efficient, but
will be more complex.

Linear Power supply:

An AC powered linear power supply typically uses a transformer to convert

mains voltage, with a rectifier and capacitor to smooth the resulting DC. Ripple,
small periodic deviations in the output, can occur due to AC frequency.
Unregulated supplies vary with load and AC variations, but linear regulators
stabilize voltage and reduce ripple. They often provide current limiting for
protection. Adjustable linear power supplies are common lab equipment,
allowing voltage to be set over a wide range, useful for circuit designers and
testing.

Transformer:

Transformers efficiently convert AC electricity from one voltage to

another, with step-up transformers increasing voltage and step-down transformers
reducing it. They work solely with AC, hence the prevalence of AC in mains
electricity. The primary coil receives input voltage, while the secondary coil
outputs the transformed voltage, linked by an alternating magnetic field in the
core. Power loss in transformers is minimal, making output power nearly equal
48
to input power. The turn's ratio, determined by the number of coil turns, governs
voltage transformation, with step-down transformers having more turns on the
primary coil and fewer on the secondary coil.

Turns ratio=Vp/Vs=Nn/Ns and Power out=Power in

Vs*Is=Vp * Ip

Vp = primary (input) voltage Vs = secondary (output) voltage

Np = number of turns on primary Ns = number of turns on secondary
coil coil
Ip = primary (input) current Is = secondary (output) current

The low voltage AC output is suitable for lamps, heaters and special AC motors.
It is not suitable for electronic circuits unless they include a rectifier and a
smoothing capacitor.

Rectifier:

There are several ways of connecting diodes to make a rectifier to convert

AC to DC. The bridge rectifier is the most important and it produces full-wave
varying DC. A full-wave rectifier can also be made from just two diodes if a
centre-tap transformer is used, but this method is rarely used now that diodes are
cheaper. A single diode can be used as a rectifier but it only uses the positive (+)
parts of the AC wave to produce half-wave varying DC.

49
The varying DC output is suitable for lamps, heaters and standard motors. It is
not suitable for electronic circuits unless they include a smoothing capacitor.

Bridge rectifier:

A bridge rectifier, comprising four diodes, is available in special packages

for ease of use, facilitating full-wave rectification by utilizing both positive and
negative sections of the AC wave. Each diode consumes 0.7V when conducting,
resulting in a total voltage drop of 1.4V in the bridge rectifier. These rectifiers are
rated by their maximum current and reverse voltage capacity, with the latter
typically three times the supply RMS voltage to handle peak voltages effectively.
Refer to the Diodes page for additional information and illustrations of bridge
rectifiers.

Alternate pairs of diodes conduct, changing over the connections so the

alternating directions of AC are converted to the one direction of DC.
Output: full-wave varying DC: (using the entire AC wave):

50
Single diode rectifier:
A single diode can be used as a rectifier but this produces half-wave
varying DC which has gaps when the AC is negative. It is hard to smooth this
sufficiently well to supply electronic circuits unless they require a very small
current so the smoothing capacitor does not significantly discharge during the
gaps. Please see the Diodes page for some examples of rectifier diodes.

Output: half-wave varying DC (using only half the AC wave):

Smoothing:

51
 Smoothing involves using a large electrolytic capacitor across the DC
supply to act as a reservoir, supplying current to the output during voltage drops.
When the rectified DC voltage fluctuates, the capacitor quickly charges near its
peak and then discharges as it provides current to the output. This process results
in a smoothed output voltage, reducing fluctuations and creating a more stable
power supply.

Smoothing Capacitor for 10% ripple, C=5*10/vs.*f

C = smoothing capacitance in farads (F)

Io = output current from the supply in amps (A)
Vs = supply voltage in volts (V), this is the peak value of the unsmootheDC
f = frequency of the AC supply in hertz (Hz), 50Hz in the UK.

The smooth DC output has a small ripple. It is suitable for most electronic circuits.

Regulator:
52
 Voltage regulator ICs come in fixed (like 5V, 12V, and 15V) or variable
output voltages and are rated by their maximum current capacity. Negative
voltage regulators are also available, often for dual supply setups. These ICs
typically feature automatic protection against excessive current and overheating.
The LM78XX series offers various fixed output voltages, suitable for diverse
applications such as local on-card regulation, logic systems, instrumentation, and
HiFi equipment. While primarily fixed voltage regulators, they can also be
configured for adjustable voltages and currents using external components. Many
fixed voltage regulator ICs have 3 leads and resemble power transistors, such as
the 7805 +5V 1A regulator, often equipped with a heat sink attachment hole if
needed.

1.Positive regulator

1. input pin
2. ground pin
3. output pin
It regulates the positive voltage

2. Negative regulator
1. ground pin
2. input pin
3. output pin
It regulate the negative voltage

53
The regulated DC output is very smooth with no ripple. It is suitable for all
electronic circuits.

PIC MICROCONTROLLER

PIC microcontrollers, derived from the PIC1640, are a family of

architecture microcontrollers produced by Microchip Technology. Initially
developed by General Instrument's Microelectronics Division, "PIC" originally
stood for "Programmable Interface Controller." These microcontrollers are
favored by industrial developers and hobbyists alike due to their affordability,
wide availability, large user base, extensive application notes, and support for
low-cost or free development tools. With the capability for serial programming
and re-programming using flash memory, Microchip announced the shipment of
its six billionth PIC processor in February 2008. Among the most renowned
microcontrollers in the industry is the PIC16f877a, known for its ease of use and
programming convenience, thanks to its FLASH memory technology allowing
for repeated write-erase cycles. Featuring a total of 40 pins, with 33 pins
dedicated to input and output, it remains a popular choice in various applications.

54
pic IC pin
The PIC16f877a microcontroller is widely applied in devices such as remote
sensors, security systems, home automation, and various industrial instruments.
Its EEPROM feature enables permanent storage of vital information like
transmitter codes and receiver frequencies, enhancing its utility. With low cost
and user-friendly handling, it finds versatility in coprocessor applications and
timer functions, expanding its usage beyond conventional microcontroller
applications. The PIC series, known for its RISC architecture and CMOS
fabrication, offers advantages like low power consumption, compact chip size,
and immunity to noise, making it a preferred choice for diverse projects.

PIC 16877A:

Various microcontrollers offer different kinds of memories. EEPROM,

EPROM, FLASH etc. are some of the memories of which FLASH is the most
recently developed. Technology that is used in PIC 16877 is flash technology, so
that data is retained even when the power is switched off. Easy programming and
erasing are other features of PIC 16F877. PIC16F877A microcontroller is used in
the project.

CORE FEATURES

CPU 8-bit PIC

Number of Pins 40

Operating Voltage (V) 2 to 5.5 V

55
Number of I/O pins 33

ADC Module 8ch, 10-bit

Timer Module 8-bit(2), 16-bit(1)

Comparators 2

DAC Module Nil

Communication UART(1), SPI(1), I2C(1), MSSP(SPI/I2C)

External Oscillator Up to 20Mhz

Internal Oscillator Nil

Program Memory Type Flash

Program Memory (KB) 14KB

CPU Speed (MIPS) 5 MIPS

RAM Bytes 368

Data EEPROM 256 bytes

Tab 5.1. Features

56
PIN DESCRIPTION

PIC16F877A consists of 40 pins enclosed in 5 ports. Each port holds 8 pins

which are bidirectional input/output pins. Pin diagram of PIC 16F877 is.

Fig 5.2. Pin diagram

RELAY DRIVER
A relay is an electro-magnetic switch which is useful if you want to use a
low voltage circuit to switch on and off a light bulb (or anything else) connected
to the 220v mains supply.

Relays are electrically operated switches that use a magnetic field generated by
the coil current to change the switch contacts.
57
Typically, relays have two switch positions and
double throw (changeover) switch contacts. They
allow one circuit to switch a second circuit,
completely separate from the first, using a
Circuit symbol for a relay
magnetic and mechanical link rather than an
electrical connection. The coil of a relay requires
a relatively large current, typically around 30mA
for a 12V relay, which most ICs cannot provide.
Therefore, transistors are commonly used to
amplify the small IC current to the larger value
required for the relay coil. Relays are electrically
operated switches with two switch positions and
double throw switch contacts. They allow one
circuit to control another separate circuit using a
Relays
magnetic and mechanical connection.

Protection diodes for relays

Transistors and ICs must be protected

from the brief high voltage produced when
a relay coil is switched off. The diagram
shows how a signal diode (eg 1N4148) is
connected 'backwards' across the relay coil
to provide this protection.
DC MOTOR
DC motors are part of the electric motors using DC power as energy
source. These devices transform electrical energy into mechanical energy. The
basic principle of DC motors is same as electric motors in general, the magnetic
interaction between the rotor and the stator that will generate spin.

58
Simple motor has six parts:
1. Armature or rotor
2. Commutator
3. Brushes
4. Axle
5. Field magnet
6. DC power supplyof some sort
Working Principle of DC motor

The major classes of DC motors are

• Shunt wound.
• Series wound.
• Compound wound.
• Separately excited.

59
 CHAPTER 6

6 CONCLUSION AND FUTURE WORK

6.1 RESULTS AND DISCUSSION

The implementation of the proposed system for accident zone, school zone,
and hospital zone detection using Convolutional Neural Networks (CNN) has
yielded promising results in enhancing road safety. Through rigorous testing and
analysis, the system's effectiveness in accurately identifying these critical areas
in real-time has been demonstrated. The system showcased high accuracy in
detecting accident-prone areas, school zones, and hospital zones using image
processing techniques and CNN algorithms. Integration with a PIC
microcontroller facilitated efficient speed control measures based on the detected
zones, successfully adjusting vehicle speeds and mitigating the risk of accidents.
Discussions surrounding the system's performance emphasized its scalability and
potential for integration with existing traffic management infrastructure.
Challenges include refining zone detection algorithms, optimizing speed control
mechanisms, and addressing scalability issues in large-scale deployments. Future
work may explore sensor integration for enhanced data collection and advanced
AI algorithms for predictive analysis. Moreover, its user-friendly interface and
seamless integration with existing traffic control systems ensure ease of adoption
and operation. Further refinements in the system's hardware and software
components can enhance its robustness and reliability in diverse environmental
conditions. Continuous evaluation and optimization are essential to ensure the
system's long-term effectiveness in mitigating road accidents and ensuring safer
roadways for all users. Overall, leveraging advanced technologies such as deep
learning and artificial intelligence can significantly improve road safety and
reduce road accidents.

60
6.2 CONCLUSION
Implemented within the current infrastructure, this project not only ensures
seamless integration but also boasts a cost-effective and durable design,
guaranteeing maximum safety for pedestrians and the community. By providing
drivers with real-time road information without distracting them from their
primary task of driving, the system enhances overall road safety. Additionally,
drivers receive timely notifications regarding their vehicle's status, further
improving their awareness and control. With its low power consumption, the
project prioritizes energy efficiency, contributing to sustainability efforts.
Furthermore, the automatic speed control feature enhances safety measures by
promptly responding to hazard signals from the external environment. Overall,
this comprehensive approach underscores the project's commitment to enhancing
road safety and user experience.

6.3 FUTURE WORK

In future iterations, this project could explore integrating more sensors to
enrich data collection and enhance situational awareness. Real-time analytics
could be implemented for quicker decision-making, while advanced
communication protocols could improve interaction with external systems.
Further development of AI algorithms could enable predictive analysis, while the
integration of autonomous driving features could enhance overall safety.
Scalability considerations should be prioritized for broader deployment, and
collaboration with regulatory bodies could ensure alignment with traffic
management systems and regulations, fostering widespread adoption and
effectiveness in enhancing road safety.

61
 APPENDICES

APPENDIX I
SOURCE CODE
SOFTWARE
import numpy as np
import os
import sys
import tensorflow as tf
from distutils.version import StrictVersion
from collections import defaultdict
from PIL import Image
from object_detection.utils import ops as utils_ops
sys.path.append("..")
if StrictVersion(tf.__version__) < StrictVersion('1.9.0'):
raise ImportError('Please upgrade your TensorFlow installation to v1.9.*
or later!')
from utils import label_map_util
from utils import visualization_utils as vis_util
MODEL_NAME = 'inference_graph'
PATH_TO_FROZEN_GRAPH = MODEL_NAME +
'/frozen_inference_graph.pb'
PATH_TO_LABELS = 'training/labelmap.pbtxt'
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(PATH_TO_FROZEN_GRAPH, 'rb') as fid:
62
 serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
category_index =
label_map_util.create_category_index_from_labelmap(PATH_TO_LABE
LS, use_display_name=True)
def run_inference_for_single_image(image, graph):
if 'detection_masks' in tensor_dict:
# The following processing is only for single image
detection_boxes = tf.squeeze(tensor_dict['detection_boxes'], [0])
detection_masks = tf.squeeze(tensor_dict['detection_masks'], [0])
# Reframe is required to translate mask from box coordinates to
image coordinates and fit the image size.
real_num_detection = tf.cast(tensor_dict['num_detections'][0],
tf.int32)
detection_boxes = tf.slice(detection_boxes, [0, 0],
[real_num_detection, -1])
detection_masks = tf.slice(detection_masks, [0, 0, 0],
[real_num_detection, -1, -1])
detection_masks_reframed =
utils_ops.reframe_box_masks_to_image_masks(
detection_masks, detection_boxes, image.shape[0],
image.shape[1])
detection_masks_reframed = tf.cast(
tf.greater(detection_masks_reframed, 0.5), tf.uint8)
# Follow the convention by adding back the batch dimension
tensor_dict['detection_masks'] = tf.expand_dims(
detection_masks_reframed, 0)

63
 image_tensor =
tf.get_default_graph().get_tensor_by_name('image_tensor:0')
# Run inference
output_dict = sess.run(tensor_dict,
feed_dict={image_tensor: np.expand_dims(image, 0)})
# all outputs are float32 numpy arrays, so convert types as appropriate
output_dict['num_detections'] = int(output_dict['num_detections'][0])
output_dict['detection_classes'] = output_dict[
'detection_classes'][0].astype(np.uint8)
output_dict['detection_boxes'] = output_dict['detection_boxes'][0]
output_dict['detection_scores'] = output_dict['detection_scores'][0]
global a2
if 'detection_masks' in output_dict:
output_dict['detection_masks'] = output_dict['detection_masks'][0]
if output_dict['detection_classes'][0] == 1 and
output_dict['detection_scores'][0] > 0.70:
print('Accedent_zone')
#a2=1
if output_dict['detection_classes'][0] == 2 and
output_dict['detection_scores'][0] > 0.70 :
print('Hospital_zone')
#a2=1
if output_dict['detection_classes'][0] == 3 and
output_dict['detection_scores'][0] > 0.70:
print('School_zone')
if a2==1:
a2=0

64
## sleep(1)
## email()
## sleep(1)
return output_dict
a1=0
a2=0
import cv2
cap = cv2.VideoCapture(0)
try:
with detection_graph.as_default():
with tf.Session() as sess:
# Get handles to input and output tensors
ops = tf.get_default_graph().get_operations()
all_tensor_names = {output.name for op in ops for output in
op.outputs}
tensor_dict = {}
for key in [
'num_detections', 'detection_boxes', 'detection_scores',
'detection_classes', 'detection_masks'
]:
tensor_name = key + ':0'
if tensor_name in all_tensor_names:
tensor_dict[key] =
tf.get_default_graph().get_tensor_by_name(
tensor_name)
while True:

65
 (__, image_np) = cap.read()
# Expand dimensions since the model expects images to have
shape: [1, None, None, 3]
image_np_expanded = np.expand_dims(image_np, axis=0)
cv2.imwrite('capture.jpg',image_np)
# Actual detection.
output_dict = run_inference_for_single_image(image_np,
detection_graph)
# Visualization of the results of a detection.
vis_util.visualize_boxes_and_labels_on_image_array(
image_np,
output_dict['detection_boxes'],
output_dict['detection_classes'],
output_dict['detection_scores'],
category_index,
instance_masks=output_dict.get('detection_masks'),
use_normalized_coordinates=True,
line_thickness=8)
cv2.imshow('object_detection',
cv2.resize(image_np,(800,600)))
if cv2.waitKey(1)& 0xFF == ord('q'):
cap.release()
cv2.destroyAllWindows()
break
except Exception as e:
print(e)
#cap.release()
66
HARDWARE
#include<pic.h>
__CONFIG(0X3F72);
#define R1 RC0int x;
void delay(unsigned int X)
{
while(X--);
}
//**************************
// UART INIT
//**************************
void TX_init()
{
TXSTA=0X24;
RCSTA=0X90;
SPBRG=25;
}
void RX_init()
{
TXSTA=0x24;
RCSTA=0x90;
SPBRG=25;
BRGH=1;
GIE=1;
PEIE=1;
RCIE=1;
}

67
void interrupt rcx(void)
{
if(RCIF==1)
{
RCIF=0;
x=RCREG;
}
}
void serial(const unsigned char *a)
{
unsigned char o;for(o=0;a[o]!=0;o++)
{
TXREG=a[o];delay(20000);
//TXREG=0X0D;
}
}
int main()
{
TRISC=0X80;
PORTC=0X00;
RX_init();
TX_init();
delay(100);
while(1)
{
if(x=='1’)
{

68
 R1=1;
delay(2000);
}
}
if(x=='2’)
{
R1=0;
delay(2000);
}
}
}

69
APPENDIX II

➢ ACCIDENT_ZONE

➢ HOSPITAL_ZONE

➢ SCHOOL_ZONE

70
71
REFERENCES
1. S. Arabi, A. Haghighat and A. Sharma, "A deep-learning-based computer
vision solution for construction vehicle detection", Comput.-Aided Civil
Infrastruct. Eng., vol. 35, no. 7, pp. 753-767, 2020.

2. H. Luo, J. Liu, W. Fang, P. E. D. Love, Q. Yu and Z. Lu, "Real-time smart

video surveillance to manage safety: A case study of a transport mega-project",
Adv. Eng. Informat., vol. 45, Aug. 2020.

3. H.-H. Jebamikyous and R. Kashef, "Autonomous vehicles perception (AVP)

using deep learning: Modeling assessment and challenges", IEEE Access, vol.
10, pp. 10523-10535, 2022.

4. X. Gao, T. Chen, R. Niu and A. Plaza, "Recognition and mapping of landslide

using a fully convolutional DenseNet and influencing factors", IEEE J. Sel.
Topics Appl. Earth Observ. Remote Sens., vol. 14, pp. 7881-7894, Aug. 2021.

5. X. Zhao, P. Sun, Z. Xu, H. Min and H. Yu, "Fusion of 3D LiDAR and camera
data for object detection in autonomous vehicle applications", IEEE Sensors
J., vol. 20, no. 9, pp. 4901-4913, May 2020.

6. G. A. Kumar, J. H. Lee, J. Hwang, J. Park, S. H. Youn and S. Kwon, "LiDAR

and camera fusion approach for object distance estimation in self-driving
vehicles", Symmetry, vol. 12, no. 2, pp. 324, Feb. 2020.

7. D.-Y. Ge, X.-F. Yao, W.-J. Xiang and Y.-P. Chen, "Vehicle detection and
tracking based on video image processing in intelligent transportation
system", Neural Comput. Appl., vol. 35, no. 3, pp. 1-13, 2022.

8. Z. Zakria, J. Deng, R. Kumar, M. S. Khokhar, J. Cai and J. Kumar, "Multiscale

and direction target detecting in remote sensing images via modified YOLO-
v4", IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 15, pp. 1039-
1048, 2022.

72
9. D. N.-N. Tran, L. H. Pham, H.-H. Nguyen and J. W. Jeon, "City-scale multi-
camera vehicle tracking of vehicles based on YOLOv7", Proc. IEEE Int. Conf.
Consum. Electron.-Asia (ICCE-Asia), pp. 1-4, Oct. 2022.

10.A. Holla, U. Verma and R. M. Pai, "Enhanced vehicle re-identification for ITS:
A feature fusion approach using deep learning", Proc. IEEE Int. Conf.
Electron. Comput. Commun. Technol. (CONECCT), pp. 1-6, Jul. 2022.

11.K.-S. Yang, Y.-K. Chen, T.-S. Chen, C.-T. Liu and S.-Y. Chien, "Tracklet-
refined multi-camera tracking based on balanced cross-domain re-
identification for vehicles", Proc. IEEE/CVF Conf. Comput. Vis. Pattern
Recognit.

12.B. A. Holla, M. M. M. Pai, U. Verma and R. M. Pai, "Vehicle re-identification

in smart city transportation using hybrid surveillance systems", Proc. IEEE
Region 10 Conf. (TENCON), pp. 335-340, Dec. 2021.

13.Q. Zhang, C. Esterwood, A. K. Pradhan, D. Tilbury, X. J. Yang and L. P.

Robert, "The impact of modality technology suspicion and NDRT engagement
on the effectiveness of AV explanations", IEEE Access, vol. 11, pp. 81981-
81994, 2023.

14.S. Han, P. Huang, H. Wang, E. Yu, D. Liu and X. Pan, "MAT: Motion-aware
multi-object tracking", Neurocomputing, vol. 476, pp. 75-86, Mar. 2022.

15.M. Chen, S. Banitaan, M. Maleki and Y. Li, "Pedestrian group detection with
K-means and DBSCAN clustering methods", Proc. IEEE Int. Conf. Electro
Inf. Technol. (eIT), pp. 1-6, May 2022.

16.E. Johnson and E. Nica, "Connected vehicle technologies autonomous driving

perception algorithms and smart sustainable urban mobility behaviors in
networked transport systems", Contemp. Readings Law Social Justice, vol.
13, no. 2, pp. 37-50, 2021.

73
17.Y. Liu, X. Zhang, B. Zhang, X. Zhang, S. Wang and J. Xu, "Multi-camera
vehicle tracking based on occlusion-aware and inter-vehicle information",
Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. Workshops (CVPRW),
pp. 3257-3264, Jun. 2022.

18.N. M. Chung, H. D.-A. Le, V. A. Nguyen, Q. Q.-V. Nguyen, T. D.-M. Nguyen,

T.-T. Thai, et al., "Multi-camera multi-vehicle tracking with domain
generalization and contextual constraints", Proc. IEEE/CVF Conf. Comput.
Vis. Pattern Recognit. Workshops (CVPRW), pp. 3327-3337, Jun. 2022.

19.Y. Zhang, C. Wang, X. Wang, W. Zeng and W. Liu, "FairMOT: On the fairness
of detection and re-identification in multiple object tracking", Int. J. Comput.
Vis., vol. 129, no. 11, pp. 3069-3087, Nov. 2021.

20.R. Wang, Y. Cui, X. Song, K. Chen and H. Fang, "Multi-information-based

convolutional neural network with attention mechanism for pedestrian
trajectory prediction", Image Vis. Comput., vol. 107, Mar. 2021.

21.X. Xie, G. Cheng, J. Wang, X. Yao and J. Han, "Oriented R-CNN for object
detection", Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), pp. 3500-3509,
Oct. 2021.

22.P. Sun, R. Zhang, Y. Jiang, T. Kong, C. Xu, W. Zhan, et al., "Sparse R-CNN:
End-to-end object detection with learnable proposals", Proc. IEEE/CVF Conf.
Comput. Vis. Pattern Recognit. (CVPR), pp. 14449-14458, Jun. 2021.

23.Z. Sun, J. Chen, L. Chao, W. Ruan and M. Mukherjee, "A survey of multiple
pedestrian tracking based on tracking-by-detection framework", IEEE Trans.
Circuits Syst. Video Technol., vol. 31, no. 5, pp. 1819-1833, May 2021.

24.M. Z. Shanti, C.-S. Cho, Y.-J. Byon, C. Y. Yeun, T.-Y. Kim, S.-K. Kim, et al.,
"A novel implementation of an AI-based smart construction safety inspection
protocol in the UAE", IEEE Access, vol. 9, pp. 166603-166616, 2021.

74
25.K. Han and X. Zeng, "Deep learning-based workers safety helmet wearing
detection on construction sites using multi-scale features", IEEE Access, vol.
10, pp. 718-729, 2022.

26.H. D. Najeeb and R. F. Ghani, "A survey on object detection and tracking in
soccer videos", Muthanna J. Pure Sci., vol. 8, no. 1, pp. 1-13, Jan. 2021.

27.P. Garnier and T. Gregoir, "Evaluating soccer player: From live camera to deep
reinforcement learning" in arXiv:2101.05388, 2021.

28.P. Kothari, S. Kreiss and A. Alahi, "Human trajectory forecasting in crowds:

A deep learning perspective", IEEE Trans. Intell. Transp. Syst., Apr. 2021.

29.N. D. Nath, A. H. Behzadan and S. G. Paal, "Deep learning for site safety:
Real-time detection of personal protective equipment", Autom. Construct.,
vol. 112, Apr. 2020.

75