ABSTRACT

In rural areas globally, the coexistence of human populations and wildlife poses persistent
challenges, often leading to conflicts endangering community safety and wildlife conserva-
tion efforts. To address this, the Advanced Animal Detection and Alert System is proposed.
Grounded in the YOLOV5 algorithm, this system enables real-time detection and classification
of wildlife near human settlements. Integration with OpenCV facilitates seamless data prepro-
cessing and analysis, ensuring prompt threat identification and timely alerts sent via email. By
empowering communities to take proactive measures, it mitigates conflict risks and minimizes
damages caused by wildlife intrusions. Moreover, it fosters collaboration among stakeholders,
promoting effective wildlife management and community safety. By reducing conflicts and pro-
moting peaceful interactions, it contributes to biodiversity conservation and ecosystem preser-
vation. The object detection model shows promising performance, achieving a mean Average
Precision (mAP) at 0.5 for all classes of 0.965, accurately identifying and localizing animals
within the dataset. Through awareness-raising and sustainable wildlife management practices,
it aims for lasting positive impacts on rural communities and their environments.

Keywords: Real-time detection, YOLOV5, OpenCV

iii
 Contents

DECLARATION i

ACKNOWLEDGEMENT ii

ABSTRACT iii

LIST OF FIGURES vi

LIST OF TABLES vii

LIST OF ABBREVIATIONS viii

1 INTRODUCTION 1

1.1 Need for the Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 Scope of the Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.4 Organization of Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 LITERATURE SURVEY 3

2.1 Existing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.2 Study on Existing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.3 Gap Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 PROPOSED SYSTEM 5

3.1 Features of Proposed System . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

iv
 3.2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2.1 Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2.2 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.2.3 Architectures Employed for Real-Time Wildlife Detection . . . . . . . 9

4 RESULTS AND ANALYSIS 12

4.1 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.2 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.2.1 PR Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.2.2 Matrics for Different Epochs and Batch Sizes . . . . . . . . . . . . . . 14

5 CONCLUSION 16

6 FUTURE SCOPE 17

BIBLIOGRAPHY 18

ANNEXURE 19

v
 LIST OF FIGURES

3.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.2 YOLOv5 Architecture [7] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.1 PR Curve at Batch Size 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.2 PR Curve at Batch Size 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.3 PR Curve at Batch Size 32 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6.1 Webpage Upload File Option . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6.2 Webpage Webcam Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6.3 Alert Message . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6.4 GitHub History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

vi
 LIST OF TABLES

2.1 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

4.1 Object Detection Performance Metrics . . . . . . . . . . . . . . . . . . . . . . 12

4.2 Results for Different Epochs and Batch Sizes YOLOv5 . . . . . . . . . . . . . 15

vii
 LIST OF ABBREVIATIONS

Abbreviations Definition
CCTV Closed-Circuit Television
mAP Mean Average Precision
OPENCV Open Source Computer Vision Library
YOLOV5 You Only Look Once Version 5

viii
 Chapter 1

INTRODUCTION

This chapter introduces the project’s primary objectives, emphasizing its goal of tackling
pressing issues in rural communities through the implementation of an Advanced Animal De-
tection and Alert System. By harnessing advanced computer vision technologies, the system
aims to mitigate human-wildlife conflicts by enabling real-time detection and proactive alerting
measures.

1.1 Need for the Project

In rural areas worldwide, the coexistence of human populations and wildlife presents a per-
sistent challenge, often leading to conflicts that endanger both community safety and wildlife
conservation efforts. These conflicts arise due to the encroachment of human settlements into
wildlife habitats, as well as the depletion of natural resources that drive wildlife closer to human-
inhabited areas. Additionally, the lack of effective mechanisms for early detection and alerting
exacerbates the risks associated with human-wildlife interactions. Thus, there is an urgent need
for innovative solutions that can mitigate these conflicts, promote community safety, and pre-
serve biodiversity.

1.2 Objective
The primary objective of this project is to develop and deploy an Advanced Animal Detec-
tion and Alert System capable of mitigating human-wildlife conflicts and enhancing community
safety in rural areas. Specific objectives include:

• Developing a robust real-time detection system for accurately identifying and categoriz-
ing wildlife near human settlements.

• Implementing an alert mechanism to promptly notify residents and authorities upon de-
tecting wildlife presence, enabling timely preventive actions.

• Improving community safety by reducing the frequency and severity of human-wildlife

conflicts through early detection and proactive alerting measures.

1
 • Contributing to wildlife conservation efforts by promoting coexistence between humans
and wildlife, thereby minimizing negative interactions and preserving natural ecosystems.

• Fostering collaboration among stakeholders to ensure the effective implementation and

sustainability of the detection and alert system, thereby enhancing the well-being of rural
communities and safeguarding biodiversity.

1.3 Scope of the Project

The scope of this project encompasses the development and implementation of an Ad-
vanced Animal Detection and Alert System designed to address the challenges posed by human-
wildlife conflicts in rural areas. The system will utilize cutting-edge computer vision technique
YOLO (You Only Look Once) algorithm, for real-time detection and classification of wildlife
approaching human settlements. Integrated with OpenCV for seamless data preprocessing and
analysis, the system will facilitate prompt identification of potential threats and timely alerts to
residents and authorities. Furthermore, the project will involve collaboration with stakehold-
ers, including local communities, wildlife experts, and governmental authorities, to ensure the
effective implementation and sustainability of the detection and alert system.

1.4 Organization of Report

Chapter 2 delves into a comprehensive exploration of the existing systems and literature,
highlighting gaps identified in current project. Following this, Chapter 3 presents the proposed
system, detailing its design, methodology, and tools utilized. Chapter 4 then presents the results
obtained and initiates discussions around them, analyzing findings in the context of the pro-
posed system’s objectives. Chapter 5 concludes the document, summarizing key insights and
implications. Finally, in Chapter 6, the future scope of the project will be explored, outlining
potential areas for further advancement within the field.

2
 Chapter 2

LITERATURE SURVEY

This chapter provides a comprehensive overview of existing research and studies relevant to
the wild animal detection and alert system, providing insights into current methodologies.

2.1 Existing System

The current system primarily relies on manual checking of CCTV videos to identify wildlife
presence. However, this approach is labor-intensive and time-consuming, as it necessitates
personnel to review footage continuously. Moreover, current systems lack the capability to
detect animals in real-time or provide immediate alerts. Instead, they depend on reviewing
CCTV footage after an animal is spotted, which can result in delayed responses to potential
wildlife threats. The absence of real-time detection makes it challenging to act swiftly when
wildlife comes close, leading to increased risks, potential damage, and safety concerns.

2.2 Study on Existing System

L. Ting et al.[1] presented an enhanced Ship Detection Algorithm based on an Improved
YOLO V5 model for ship detection, acknowledging its high computational cost. This enhance-
ment aims to improve the accuracy of ship detection while addressing the challenges posed by
its computational demands.

R. Thangaraj et al.[2] introduced YOLOV5 and YOLOV3 models for automatic detection
and classification of wild animal species. However, these models suffer from limited accuracy,
particularly in complex environments. Their work highlights the potential of YOLO models in
wildlife monitoring but underscores the need for further refinement to enhance accuracy.

B. K and D. V[3] , in their review, highlighted the reliance of computer vision techniques
on extensive datasets for successful animal detection and classification. They emphasized the
necessity of mitigating dataset dependency to propel the field of animal detection and classifi-
cation forward, presenting opportunities for future enhancements in application.

Glenn et al.[4] introduced YOLO-Z in their work ”YOLO-Z: Improving small object detec-
tion in YOLOv5 for autonomous vehicles” to tackle the challenge of small object detection in

3
autonomous vehicles. This modification is designed to enhance YOLOv5’s performance in this
specific aspect. Their research aims to enhance the overall efficiency and safety of autonomous
vehicle systems by accurately detecting small objects.

Table 2.1: Literature Survey

Sl No. Title Author Model Disadvantage

[1]
1 Ship Detection Algorithm based on L. Ting et al. Improved YOLO V5 High computational cost
Improved YOLO V5(2021)
2 Automatic Detection and Classifi- R. Thangaraj et al.[2] YOLOV5, YOLOV3 Limited accuracy in complex
cation of Wild Animal Species Us- environments
ing YOLO Models(2023)
3 A Review on Animal Detection and B. K and D. V[3] Computer Vision, Neu- Dependency on large datasets
Classification using Computer Vi- ral Network
sion Techniques: Scope for Future
Enhancement to Application(2023)
4 YOLO-Z: Improving small ob- Glenn et al.[4] YOLOv5 YOLOv5’s small object de-
ject detection in YOLOv5 for au- tection struggle for AVs.
tonomous vehicles(2024)

2.3 Gap Identification

Through extensive review and analysis of existing literature, a notable gap emerges in
the domain of real-time detection and alert systems for wildlife monitoring. Currently, many
wildlife monitoring systems rely on manual observation and retrospective analysis of CCTV
footage to identify wildlife presence. However, this approach is labor-intensive, time-consuming,
and lacks the ability to provide immediate alerts. The absence of real-time detection capabilities
hinders the ability to respond promptly to wildlife threats, potentially leading to increased risks,
damage, and safety concerns.

Moreover, the lack of real-time alert systems limits the scalability of wildlife management
practices and inhibits the adoption of modern technologies in wildlife monitoring. There is a
clear need for an enhanced system that can detect wild animals in real-time and send immediate
alerts to relevant stakeholders. Such a system would enable timely response and mitigation of
potential risks, thereby improving wildlife management efficiency and enhancing human safety
in areas prone to wildlife encounters. Closing this gap is imperative for advancing the field
of wildlife monitoring and management, facilitating the integration of innovative technologies
such as deep learning-based object detection models

4
 Chapter 3

PROPOSED SYSTEM

The proposed system aims to develop a revolutionary approach to wildlife monitoring through
real-time detection and alert mechanisms powered by the YOLOv5 deep learning model. Tradi-
tional methods often lack immediate alert capabilities, leading to delayed responses to wildlife
threats. By leveraging YOLOv5, the system enables instant identification of wild animals from
CCTV videos, triggering timely alerts to stakeholders. This innovative approach enhances
wildlife management efficiency, mitigates risks, and improves human safety in wildlife-prone
areas, ushering in a new era of proactive wildlife monitoring.

3.1 Features of Proposed System

The proposed system for real-time detection of wild animals from CCTV videos and
alert messaging is designed to provide a comprehensive solution to enhance both human safety
and wildlife conservation efforts. Leveraging the advanced object detection capabilities of the
YOLOv5 deep learning model, the system swiftly identifies wild animals within video streams,
enabling an immediate response to potential wildlife threats. Once a wild animal is detected,
the system triggers an alert mechanism to notify relevant stakeholders, such as wildlife author-
ities or local residents, via email. This ensures timely action to mitigate risks associated with
wildlife presence, minimizing potential human-wildlife conflicts. Moreover, the system priori-
tizes reliability and accuracy in animal detection, minimizing false positives and false negatives
even in challenging environmental conditions. Its adaptability to various scenarios and light-
ing conditions enhances its effectiveness in real-world applications. Additionally, the system is
scalable and seamlessly integrates with existing CCTV camera infrastructure, facilitating easy
deployment and compatibility with current monitoring systems.

Beyond immediate threat detection, the proposed system contributes to long-term wildlife
conservation efforts. By accurately identifying and tracking wild animals in real-time, the sys-
tem generates valuable data insights that can inform habitat preservation strategies and wildlife
management practices. This data-driven approach aids in understanding wildlife behavior pat-
terns, migration routes, and habitat utilization, allowing for informed decision-making by con-
servation authorities. Furthermore, the system’s alert messaging feature serves as a proactive

5
measure to prevent human-wildlife conflicts. By promptly notifying relevant stakeholders about
the presence of wild animals in specific areas, preemptive actions can be taken to minimize po-
tential risks to both humans and animals. This proactive stance not only enhances human safety
but also reduces the likelihood of wildlife disturbance and stress, thereby promoting harmonious
coexistence between humans and wildlife.

3.2 Materials and Methods

This section provides a concise overview of materials, tools and experimental approach. By
outlining my methodology, I offer readers insights into the systematic procedures followed in
my project. This section serves as a guide for understanding the methods employed in my study,
facilitating reproducibility and highlighting the rigor of my scientific inquiry.

3.2.1 Tools
1. Visual Studio: The study was conducted using Visual Studio, offering a comprehensive
integrated development environment (IDE) for software development tasks. Visual Studio
provided a robust platform for writing, debugging, and executing code, particularly in the
context of developing and testing deep learning models. The IDE’s features facilitated
seamless experimentation with deep learning algorithms and the implementation of the
proposed classification system.

2. Jupyter Notebook: The study was conducted using the Jupyter Notebook, providing an
interactive and collaborative platform for developing and executing Python code. This
environment facilitated seamless experimentation with deep learning models and the im-
plementation of the proposed classification system.

3. LabelImg (For Annotating Images): LabelImg was employed as a crucial tool for an-
notating images, a fundamental step in preparing data for training deep learning mod-
els. Annotating images involves labeling objects of interest within the images, providing
ground truth data for model training. LabelImg’s intuitive interface and annotation capa-
bilities streamlined the annotation process, ensuring accurate and consistent labeling of
objects in the dataset.

4. CPU: AMD Ryzen 7 5800H: The computational power for this study was provided by
an AMD Ryzen 7 5800H CPU. The CPU played a pivotal role in various aspects of
the study, including model training, feature extraction, and overall execution of the deep

6
 learning algorithms. The Ryzen 7 5800H’s high performance and efficiency significantly
contributed to the timely processing of image data and the training of complex neural
network architectures, ensuring smooth progress in the experimentation and development
phases of the project.

3.2.2 System Architecture

The system architecture is tailored to leverage the YOLOv5 deep learning model for real-
time detection of wild animals from CCTV videos and alert messaging. YOLOv5 is renowned
for its single-stage object detection abilities, making it suitable for rapid and accurate identifi-
cation of animals within video streams.

To facilitate the training process, a frame slicing algorithm is implemented to segment the
video into individual frames. Each frame is then labeled using LabelImg, a tool specifically de-
signed for annotating images. This meticulous labeling process ensures that each frame contains
accurate annotations of wild animals, providing ground truth data for training and evaluation of
the YOLOv5 model.

The YOLOv5 model undergoes fine-tuning to specialize in recognizing the distinct features
and characteristics of various animal species. Upon detection of a wild animal, the system
triggers an alert mechanism to notify relevant stakeholders, such as wildlife authorities or lo-
cal residents, via email. This ensures a timely response and action to mitigate potential risks
associated with wildlife presence.

The system architecture emphasizes simplicity and efficiency, focusing on the integration of
YOLOv5 for real-time animal detection and alert messaging systems for timely notifications.
This streamlined approach allows for effective detection of wild animals and enhances human
safety in areas monitored by CCTV cameras. By combining advanced deep learning techniques
with practical alert mechanisms, the system contributes to wildlife conservation efforts and
minimizes human-animal conflicts.