Similarity Report ID: oid:3618:60680463

PAPER NAME AUTHOR

Final Report (2).pdf New file A

WORD COUNT CHARACTER COUNT

13546 Words 79671 Characters

PAGE COUNT FILE SIZE

56 Pages 1.4MB

SUBMISSION DATE REPORT DATE

Jun 4, 2024 11:02 AM GMT+5:30 Jun 4, 2024 11:03 AM GMT+5:30

12% Overall Similarity

The combined total of all matches, including overlapping sources, for each database.
10% Internet database 7% Publications database
Crossref database Crossref Posted Content database

Excluded from Similarity Report

Submitted Works database

Summary
 A PROJECT REPORT ON

NEURAL-THE SMART CCTV SURVEILLANCE

1
SUBMITTED IN PARTIAL FULFILLMENT FOR DEGREE OF

BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE AND ENGINEERING (AI&ML)

BY

ANAS SIDDIQUI (2001321530010)

SUNNY PATEL (2001321550052)

SHUBH LODHI (2101321539010)

AAKASH SRIVASTAVA (2001321530001)

UNDER THE GUIDANCE OF

DR. JAYSHANKAR PRASAD

4
DEPARTMENT OF CSE-(AI&ML)

1
GREATER NOIDA INSTITUTE OF TECHNOLOGY
GREATER NOIDA

DR.A.P.J. ABDUL KALAM UNIVERSITY, LUCKNOW

MAY 2024

1
 72
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING (AI&ML)
Session: 2023-2024
Project Completion Certificate

Date: 3th JUNE, 2024

This is to certify that Mr. ANAS SIDDIQUI, Mr. SUNNY PATEL, Mr. SHUBH
LODI, Mr. AAKASH SRIVASTAVA bearing Roll no.:
th
2001321530010,2001321550052,2101321539010,2001321530001 student of 4 year
CSE(AI&ML) has completed project (KCS -851) with the Department of CSE(AI&ML)
He worked on the project titled “NEURAL-THE SMART CCTV SURVEILLANCE”
under the guidance of DR. J.S. PRASAD.
This project work has not been submitted anywhere for any diploma/degree.

DR. J.S. PRASAD

HOD, Dept. of CSE AI&ML Project Coordinator

2
 ACKNOWLEDGEMENT

7
We would like to express my sincere thanks to Prof. (DR) JAYSHANKAR PRASAD
1
valuable guidance and Support in completing my project.

I would also
1
like to express my gratitude towards our Prof (DR).JAYSHANKAR
PRASAD for giving me this great opportunity to do a project on NEURAL -THE
1
SMART CCVT SURVILLIANCE . Without their support and suggestions, this project
would not have been completed.

Place: Greater Noida

Date – 03/06/2024

Student Name : ANAS SIDDIQUI

SUNNY PATEL
SHUBH LODHI
AAKASH SRIVASTAVA

3
 ABSTRACT

The rapid expansion of security cameras in public places has necessitated the
development of smart CCTV surveillance systems. These advanced systems can process
live video feeds and detect security threats in real time. They utilize sophisticated image
processing techniques and artificial intelligence (AI) algorithms to analyze video data,
identify suspicious behaviors, and alert security personnel when necessary. Python has
emerged as a popular programming language for developing these smart CCTV
surveillance systems, owing to its simplicity and powerful libraries tailored for computer
vision and machine learning. Python's versatility allows developers to efficiently
construct complex video analysis pipelines that can perform many tasks, including object
detection, tracking, and classification. One of the primary benefits of employing Python
in smart CCTV surveillance systems is its compatibility with open-source computer
vision libraries like OpenCV. OpenCV offers a comprehensive suite of image and video
processing tools, enabling developers to implement sophisticated
7
algorithms for object
detection and tracking. Techniques such as YOLO V4 (You Only Look Once) and SSD
(Single Shot Detector) can be rapidly deployed to detect objects in real time with
remarkable accuracy.

Additionally, Python's ability to integrate seamlessly with deep learning frameworks like
TensorFlow and PyTorch provides a significant advantage. These frameworks empower
developers to train and deploy deep neural networks capable of performing various tasks,
including facial recognition, crowd detection, and anomaly detection. By leveraging these
frameworks, developers can enhance the functionality of smart CCTV systems, making
them more responsive and efficient in identifying and addressing security threats. The
implementation of Python in smart CCTV surveillance systems is becoming increasingly
prevalent across diverse applications, ranging from public safety and security to traffic
monitoring and retail analytics. The synergy between Python and its associated libraries
and frameworks enables developers to swiftly construct robust and efficient video
analysis pipelines. These pipelines are crucial for identifying and responding to security
threats in real time, thereby significantly enhancing public safety and security in a world
that is becoming increasingly complex.

Smart CCTV surveillance systems, fortified by the 3

capabilities of Python, represent a
critical tool in modern security infrastructure. The ability to quickly analyze huge
amounts of video data and identify potential threats allows for proactive measures to be
taken, ensuring a higher level of safety and security for the public. As these systems
continue to evolve, their applications are likely to expand, further solidifying their role in
maintaining order and safety in various environments.

In summary, the utilization of Python in the development of smart CCTV surveillance

systems offers numerous benefits. Its ease of use, combined with powerful libraries for
computer vision and machine learning, facilitates the rapid construction of advanced
4
video analysis pipelines. OpenCV, TensorFlow, and PyTorch are just a few of the tools
that enable developers to implement high-accuracy object detection and tracking
algorithms, as well as deep neural networks for complex tasks. As a result, smart CCTV
systems are becoming indispensable in ensuring public safety and security, addressing
the challenges of an increasingly complex world with efficiency and precision.

Accuracy (in %) after performing test - 92.33%.

5
 35
TABLE OF CONTENTS

Certificate ii
Acknowledgment iii
Abstract iv-v
List of Tables vi
List of Figures vii

1. CHAPTER 1
50
INTRODUCTION
1.1 GENERAL
1.2 NEED OF MACHINE LEARNING
1.3 CNN
1.4 COMPUTER VISION BASED SURVEILLANCE
1.5 OBJECTIVES
1.6 ORGANISATION OF THESIS

2. CHAPTER 2 LITEREATURE STUDY

2.1 FACE RECOGNITION
2.2 OBJECT DETECTION
2.3 INTRUDER ALERT
2.4 INFERENCE OF L.S

3. CHAPTER 3 METHODOLOGY
3.1 DEVELOPMENT E.S
3.2 IMAGE PROCESS HANDLING
3.3 MODULES
1
4. CHAPTER 4 RESULT

5. CHAPTER 5 CONCLUSION
6. REFERENCES

6
 LIST OF FIGURES

1. BLOCK DIAGRAM OF CNN 10

2. SHOWING WORKING OF FACE DETECTION 16

3. EXPLANING ABOUT CASCADE CLASSIFIER 18

4. EXPLAINING ABOUT LBPH FACE RECOGNITION WORKS 19

5. ABNORMAL DETECTION PROCESS 29

6
ABNORMAL DETECTION TRAINING PROCESS 31

7. OUTPUT OF FACE RECOGNITION 51

7
 CHAPTER 1

NEURAL- THE SMART SURVEILLANCE

SYSTEM

1. Introduction
1.1 General
Surveillance systems have become an essential component in maintaining public safety and security.
With the increasing incidence of criminal activities, both businesses and governments are continuously
seeking to enhance the effectiveness of their surveillance systems. Recent technological advancements
have paved the way for the development of intelligent surveillance systems capable of analyzing vast
amounts of data and detecting unusual activities in real-time. This project report delves into the
creation of a smart CCTV surveillance system utilizing Python.

62
The main objective of this project is to develop a smart CCTV surveillance system that can analyze
3
video footage in real-time and identify unusual activities. The system will leverage advanced machine
learning algorithms to learn from past historical data, thereby identifying those patterns that may
indicate potential threats. Additionally, the system will have the capability to alert security personnel
immediately, enabling swift and effective responses to any security breaches.
The development of this system will be undertaken using Python, a high-level programming language
renowned for its extensive use in machine learning and artificial intelligence. Python’s rich ecosystem
of libraries and frameworks simplifies the development of intelligent systems. For this project, the
OpenCV library will be employed for image and video processing, the TensorFlow library for machine
learning, and the Flask framework for web development.
25
The architecture of the system will be designed to be both scalable and modular, ensuring it can be
easily adapted to several environments and configurations. The system will consist of several key
components, including cameras, servers, and client applications. The cameras will capture video
footage, which will be processed by the servers. These servers will utilize machine learning algorithms
to analyze the footage and detect any unusual activities. Client applications will provide a user
interface for security personnel to monitor the system and respond to potential threats. Privacy
considerations will be integral to the system’s design, ensuring compliance with all relevant privacy
regulations. Video footage will only be stored for a limited duration, and access to the system will be
restricted to authorized personnel. The smart CCTV surveillance system offers several advantages
8
 2
over traditional surveillance systems. Firstly, it can detect unusual activities in real time, allowing
security personnel to respond promptly and efficiently to potential threats. Secondly, by learning from
past events, the system can identify patterns that may signal potential threats. Finally, its scalable and
modular design allows for easy adaptation to various environments and configurations.

In summary, the smart CCTV surveillance system developed using Python represents an innovative
approach to enhancing public safety and security. The system will provide real-time monitoring and
detection of potential threats while being scalable and modular. This project report will offer a
comprehensive overview of the system’s architecture, design, implementation, and evaluation. It will
also discuss the challenges encountered during development and suggest future research and
development directions in this field.

1.2 Need for machine learning

The integration of machine learning into smart CCTV surveillance systems is essential to overcome
the limitations of traditional surveillance methods, which struggle with the overwhelming volume of
33
data, human error, and the need for real-time analysis. Machine learning algorithms can process large
7
amounts of video footage in real time, consistently identify patterns and anomalies that may indicate
32
potential threats, and provide immediate alerts to security personnel. This not only enhances the
accuracy and efficiency of threat detection but also enables proactive security measures by learning
from historical data. Moreover, machine learning systems are highly scalable and adaptable, capable
of evolving with emerging threats and expanding security needs. By automating detection and
reducing the reliance on constant human supervision, these intelligent systems optimize resource use,
allowing security personnel to focus on responding to verified threats and critical tasks. Thus, the
incorporation of machine learning is a necessity for creating advanced, efficient, and effective CCTV
surveillance systems that can significantly enhance public safety and security.

1.3 CNN
Some widely used CNN architectures are mentioned-ResNet, GoogLeNet, AlexNet,VGGNet, LeNet.
Convolutional Neural Networks can be thought of as an artificial brain at work, solving a number of
problems that are happening around us every day. It uses a huge variety of multilayer perceptron that
do not require much pre-processing and collectively try to mimic a biological neural network. The
building blocks of a CNN are five different layers, namely,
9
 38
1. Convolutional layer
2. Rectified Linear Unit layer
3. Fully connected layer
4. Pooling layer
5. Loss layer

Figure 1.1: Block Diagram of CNN

The terms are commonly used layers in convolutional neural networks (CNNs), which are widely
employed for tasks such as image recognition and classification. Here's a brief explanation of each layer:

16
1. Convolutional Layer: The convolutional layer is the core building block of a CNN. It applies a set of
learnable filters (also known as kernels) to the input data, performing convolution operations. This layer
extracts spatial hierarchies of features by sliding the filters across the input, capturing different patterns
and details. Each filter learns to detect specific local patterns, such as edges or textures, which are then
combined to form higher-level features.

43
2. Rectified Linear Unit (ReLU) Layer: The ReLU layer is an activation function applied element-wise
11
to the output of a convolutional layer or a fully connected layer. It introduces non-linearity into the
39
network, allowing it to model complex relationships. The ReLU function sets all negative values to zero
and leaves positive values unchanged. It helps in improving the network's ability to learn and converge
faster during training.

10
 15
3. Fully Connected Layer: The fully connected layer, also known as the dense layer, connects each
11
neuron to every neuron in the previous layer. In CNNs, fully connected layers are typically placed at the
end of the network, after the convolutional and pooling layers. These layers perform the classification
or regression task by learning the relationships between the extracted features and the target labels. The
fully connected layer outputs the final predictions or feature representations.

5
4. Pooling Layer: The pooling layer reduces the spatial dimensions (width and height) of the input
37
volume while retaining important features. Commonly used pooling operations include max pooling and
average pooling. Pooling helps in reducing the number of parameters and computations in the network,
70
as well as providing a degree of translation invariance, allowing the network to recognize patterns
regardless of their precise location in the input.

15 5
5. Loss Layer: The loss layer, also known as the output layer, calculates the loss or error between the
predicted output and the ground truth labels. The choice of the loss function depends on the specific
task, such as classification (e.g., Softmax cross-entropy loss) or regression (e.g., mean squared error
26
loss). The loss value is used to measure the performance of the network during training and to update
the network's parameters through backpropagation.

5
These layers work together to form a CNN architecture, allowing the network to automatically learn and
71
extract relevant features from input data, such as images, and make predictions or classifications based
on those features.

About the components of CNN:-

1. It comprises Convolution layers, pooling layers, completely associated layers, and initiation
capacities. All these join tasks to adequately give a very exact yield forecast.
2. Convolution layers and pooling layers essentially act like monster sifts that assist in channeling
through highlights that might be superfluous for the preparation procedure.
3. A convolution layer essentially includes the spot result of the ’channel’ and the information
volume to give an element map.
4. Walk is the rate at which the channel crosses. Cushioning is utilized to add zeros to the info lattice
with the goal that we can modify its size according to our needs.
61
5. A pooling layer works as a reducer of the spatial size of portrayal, to decrease the number of
parameters and calculation. It follows up on each component map autonomously.
6. The last layers in a CNN are the completely associated layers(FC) whose activity is togroup the

11
 highlights created by the past layers.
7. Essentially every layer in a ConvNet completes a change on a 3D input volume. CNN is designed
for minimal pre-processing in comparison with other algorithms forimage classification. Therefore,
the network in CNN learns the filters whereas the otheralgorithms had to get them engineered by
developers. This gives independence fromhuman effort and prior knowledge, which stands out as an
advantage for CNN over other algorithms. The Convolution Layer applies an operation called the
Convolution Operationto the input. The result from the previous layer is then passed to the succeeding
layer. Theresponse on an individual neuron to a visual stimuli is emulated by a neuron. Pooling Layer
then combines output of the neuron clusters present in a single layer and puts it into a single layer
in the succeeding layer. The dimension of the data gets reduced. Pooling layersare either global or local
in nature. Fully Connected Layer connects each neuron belonging to a particular layer to all the neurons in
another layer of the same network. The area for the input of a neuron in a neural network is the receptive field.
64
That is, the input neurons from the previous layer form the receptive field of a neuron. Biases and weights are
used to form a vector which specifies the input values that are derived from the receptive field.Then a function
is applied to the input values to calculate an output from each neuron. Theconvolution layer and the pooling layer
aid in reducing few of the features which may notbe required for training.

1.4 COMPUTER VISION BASED SURVILLIANCE SYSTEM

The evolution of smart CCTV surveillance systems has been significantly influenced by advancements
68
in machine learning, particularly the application of Convolutional Neural Networks (CNNs). Various
research efforts and practical implementations have demonstrated the potential of these technologies
in enhancing the effectiveness and efficiency of surveillance systems.

Object Detection and Tracking:

60
One of the foundational works in this area is the development of the YOLO (You Only Look Once)
67
algorithm by Redmon et al., which has become a standard for real-time object detection in surveillance
footage. YOLO's capability to detect multiple objects within a single frame in real-time is particularly
advantageous for surveillance in busy environments such as airports and public squares. This method
has been refined in subsequent versions to improve accuracy and speed, making it a crucial tool for
modern surveillance systems.

Anomaly Detection:
Research by Sultani et al. has focused on anomaly detection in video surveillance using deep learning
models. Their framework employs CNNs to learn normal activity patterns from extensive datasets and

12
then identify deviations that could signify unusual or suspicious behavior. This approach minimizes
the need for manual monitoring and enhances the system's ability to detect potential threats in real-
time, even in complex scenarios with numerous variables.

Facial Recognition:
The integration of CNNs in facial recognition systems has also been a major area of development. The
FaceNet model by Schroff et al. introduced an innovative approach to facial recognition by mapping
faces into a Euclidean space, where distances between points represent facial similarities. This model
has been incorporated into surveillance systems to improve the accuracy of identifying individuals,
which is critical in maintaining security in high-risk areas such as government buildings and financial
institutions.

Video Summarization:
Ji et al. have contributed to the field with their work on video summarization using CNNs. Their
approach involves extracting key frames from lengthy surveillance videos, thus reducing the volume
of data that needs to be reviewed by human operators. This method ensures that essential events are
captured and highlighted, facilitating more efficient monitoring and quicker response times.

Hybrid Models:
5
The combination of CNNs with other machine learning models, such as Recurrent Neural Networks
(RNNs), has shown promising results in enhancing surveillance systems. This hybrid approach
leverages the strengths of CNNs in spatial data processing and RNNs in temporal sequence analysis.
For example, an integrated system can analyze the movement patterns of individuals over time,
improving the prediction and identification of suspicious activities.

Real-World Applications:
Several real-world implementations have demonstrated the practical benefits of smart CCTV systems
enhanced by machine learning. For instance, smart surveillance systems deployed in smart cities use
CNNs for real-time monitoring and analytics, providing city administrators with valuable insights for
managing public safety. Similarly, retail businesses have adopted these technologies to monitor
customer behavior and prevent theft, showcasing the versatility and impact of smart CCTV systems
across various domains.

In summary, the related work in smart CCTV surveillance underscores the transformative impact of

13
 CNNs and other machine learning techniques on enhancing public safety and security. These
advancements have led to the development of more intelligent, efficient, and responsive surveillance
systems capable of real-time threat detection and analysis. Ongoing research continues to explore new
methodologies and applications, further advancing the capabilities of smart CCTV surveillance.

1.5 OBJECTIVES

We need to design a solution that can detect any illegal activity or unusual activity from camera
surveillance in real-time and send the information to the Police(an interface for government
54
organizations) with the nature of the crime, i.e., low-risk, medium-risk, or high-risk.
1) Camera Placement and Setup:
Strategically place CCTV cameras in key locations prone to criminal activities or areas requiring
monitoring. Ensure cameras cover a wide range of views to capture incidents effectively
2) Live Video Feed Processing
Utilize computer vision techniques to process live video feeds from CCTV cameras in real time.
Implement algorithms to detect and analyze various activities and behaviors.
3) Illegal Activity Detection:
Train machine learning models to recognize patterns associated with different types of illegal activities
such as theft, vandalism, assault, etc. Utilize object detection algorithms to identify suspicious objects
or movement
4) Risk Assessment:
48
Classify detected activities into low-risk, medium-risk, or high-risk categories based on predefined
criteria. Consider factors such as the severity of the crime, potential harm to individuals or property, and
the likelihood of escalation.
5) Alert Generation:
Develop a system to generate alerts when illegal or unusual activities are detected. Include information
about the nature of the crime and the assessed risk level in the alert.
6) Integration with Police Interface:
Create an interface for government organizations, specifically the police, to receive alerts from the
surveillance system. Ensure the interface provides real-time notifications and relevant details about the
detected activities.
7) Response Mechanism:
Define procedures for police response based on the nature and risk level of the detected activity.
Establish protocols for dispatching officers to the location, escalating response based on the severity of
14
the situation
8) Privacy and Data Security:
Implement measures to safeguard the privacy of individuals captured on camera. Ensure compliance
with data protection regulations and guidelines.
9) Testing and Deployment:
Conduct thorough testing of the system in controlled environments before deploying it in real-world
settings. Gradually deploy the solution, starting with pilot implementations in select locations before
scaling up.

66
1.6 ORGANISATION OF THESIS
chapter 1 is Introduction part where we describe about our project and what are the objectives of the
project.
Chapter 2 is literature study of our project Where we describe the details of our project like face
recognition and object detection.
Chapter 3 is methodology where we give the details about the methods which we use in our project.
Chapter 4 is result of the over all project.
Chapter 5 is the conclusions of our project.

15
 CHAPTER 2
LITERATURE STUDY

2.1 FACE RECOGNINITON

Face recognition technology has made significant strides in recent years, becoming a key component
of modern security systems. This technology's application in CCTV surveillance allows for real-time
detection and identification of individuals. In this project report, we explore the use of Python to
implement a smart CCTV surveillance system that incorporates face recognition. This technology
9
employs algorithms to identify and verify individuals based on their facial features by analyzing
9
images or video frames. Unique features such as the distance between the eyes, the shape of the nose,
and the contour of the jawline are extracted and compared to a database of known faces to determine
identity. The initial step in developing a smart CCTV surveillance system with face recognition
involves capturing high-quality video footage of the monitored area using strategically placed high-
resolution cameras. This footage is then processed with Python and OpenCV, a widely-used computer
vision library. OpenCV offers tools for face detection, utilizing Haar cascades, which are classifiers
trained with machine learning algorithms to detect facial patterns in images.

Fig 2.1: Showing working of Face Detection

Once faces are detected, the next step is to extract facial features using face landmark detection, which
identifies key points such as the corners of the eyes, the nose tip, and the edges of the lips. These
landmarks help calculate the unique features of a face. To identify faces in the footage, these features
21
are compared against a database of known faces using machine learning algorithms like Principal
Component Analysis (PCA) or Linear Discriminant Analysis (LDA). These algorithms create a
mathematical model from the facial features in the database to identify new faces.Upon identifying a
16
face, the system can trigger various actions based on the application. For instance, in a security system,
it might alert security personnel or automatically lock doors to prevent unauthorized access. Beyond
security, face recognition has applications in retail for tracking customer behavior and providing
personalized recommendations, and in healthcare for monitoring patients and detecting signs of illness
or distress.
In conclusion, face recognition technology is a versatile tool with numerous applications. This project
report discusses how Python and OpenCV can be used to implement a smart CCTV surveillance
system leveraging face recognition. Given the rapid technological advancements, many more
innovative uses of this technology are anticipated in the future.

6
The face recognition process involves two main steps: detecting faces in frames and predicting
identities using the LBPH face recognizer algorithm from a pre-trained model.

6
Detecting Faces in Frames

This task is accomplished using Haar cascade classifiers, available in the OpenCV Python module.
6
Cascade classifiers, a form of ensemble learning known as boosting, are typically based on Adaboost
classifiers (or variants like Real Adaboost, Gentle Adaboost, and Logitboost). These classifiers are
6
trained on hundreds of sample images containing the target object (in this case, human faces) and
images without the object. Common human facial features detected include:
- Dark regions around the eyes compared to the upper cheeks
- A bright nose bridge compared to the eyes
- Specific positioning of the eyes, mouth, and nose

17
 Fig 2.2: Explaining about Cascade Classifier

These features, known as Haar Features, are extracted through a process similar to applying
6
convolution kernels to detect their presence in the image. OpenCV’s cascade classifier function is used
to detect faces in frames.

6
Using LBPH for Face Recognition

Once faces are detected, the LBPH algorithm identifies them by comparing them against a trained
dataset. LBPH uses four parameters:
7
-Radius: Defines the circular local binary pattern radius around the central pixel, usually set to 1.
- Neighbors: The number of sample points for the circular local binary pattern, typically set to 8. More
sample points increase computational cost.
22 22
- Grid X: The number of horizontal cells in the grid, usually set to 8.

18
- Grid Y: The number of vertical cells in the grid, usually set to 8.

Fig 2.3: Explaining about LBPH Face Recognition works

27
The LBPH algorithm starts by creating an intermediate image highlighting facial features using a
sliding window based on the radius and neighbors parameters. Histograms are then extracted from this
40
image using the Grid X and Grid Y parameters, dividing the image into multiple grids. During
prediction, the same steps are applied, and the histograms are compared with the pre-trained model to
identify faces.

By following these steps, the model can accurately detect and recognize faces in video footage,
providing an effective solution for real-time surveillance applications.

10
2.2 OBJECT DETECTION

Object detection is a fundamental task in computer vision that involves identifying and localizing
objects of interest within an image or video. It plays a crucial role in many applications, including
smart CCTV surveillance systems. These systems integrate artificial intelligence (AI) algorithms to
enhance functionality, such as the ability to automatically detect objects and events of interest. Python
is a favored programming language for developing such systems due to its simplicity, extensive
libraries and frameworks, and versatility in handling data and implementing machine learning
algorithms. Object detection in images and videos typically involves three main stages: image
preprocessing, object detection, and post-processing. During image preprocessing, the input image or
video frames are prepared for object detection, which may involve resizing, normalizing, and color-
3
space conversion. The object detection stage involves identifying and locating objects within the image
or video frames. Post-processing refines the detected objects and removes any false positives.

19
 59
Object detection can be approached using various methods, including traditional computer vision
techniques and deep learning-based methods. Traditional methods, such as Haar cascades, rely on
handcrafted features and machine learning algorithms to detect objects, but these methods can be
3
limited in accuracy and robustness, especially for complex detection tasks. In contrast, deep learning-
based methods use convolutional neural networks (CNNs) to learn features from input data and
20
perform object detection, demonstrating superior performance compared to traditional methods and
currently representing the state-of-the-art in object detection. A popular deep learning-based method
10
is the region-based CNN (R-CNN) family of algorithms, which involves dividing the input image into
regions and processing each separately to generate object proposals. These proposals are then passed
through a CNN to extract features used for classification and refinement. This approach has evolved
with faster R-CNN, Mask R-CNN, and other variations, achieving high accuracy on benchmark
datasets such as COCO and PASCAL VOC.
55
Another prominent deep learning-based method for object detection is YOLO (You Only Look Once),
25
a real-time object detection algorithm that can detect objects in a single pass. YOLO divides the input
image into a grid and performs detection on each grid cell to generate object proposals. Over the years,
YOLO has improved with versions like YOLOv2 and YOLOv3, achieving high accuracy and real-
time performance. Python provides numerous libraries and frameworks for implementing object
detection algorithms, including OpenCV, TensorFlow, Keras, and PyTorch. These libraries offer pre-
trained models for object detection and tools for training custom models on specific datasets, along
with capabilities for image and video preprocessing, post-processing, and visualizing detection results.

In smart CCTV surveillance systems, object detection is utilized to automatically detect and track
objects of interest within surveillance footage. This includes detecting people, vehicles, and other
objects, as well as identifying events such as abnormal behavior or incidents. Object detection can
trigger alarms or notifications and provide real-time insights into the surveillance footage. Overall,
51
object detection is a critical task in many computer vision applications, including smart CCTV
20
surveillance systems. With the availability of deep learning-based methods and the ease of use
provided by Python libraries and frameworks, implementing object detection in a wide range of
applications has become more accessible than ever.

One specific method used for object detection involves measuring structural similarity (SSIM) to find
differences between two frames—one captured before an event and another after the event. SSIM is a
metric for measuring similarity between two images, focusing on three key features: luminance,
6
contrast, and structure. Luminance is measured by averaging all pixel values, contrast is defined as the

20
difference between the highest and lowest intensity values, and structure is assessed by normalizing
the input signal to unit standard deviation for robust comparison. The SSIM method, facilitated by the
skimage package in Python, simplifies the process by providing built-in functions to perform these
calculations. By feeding two images or frames into the function, the skimage package outputs a masked
image with a similarity score, effectively highlighting the differences and enhancing the accuracy of
object detection in surveillance footage.

2.3 INTRUDER ALERT

CCTV (Closed Circuit Television) surveillance is an essential aspect of security systems in public
places, commercial buildings, residential areas, and other locations. With technological advancements,
CCTV surveillance systems have become smarter, more intelligent, and efficient. One such
advancement is the implementation of Python programming language to create a smart CCTV
19
surveillance system that provides an intruder alert. This smart CCTV surveillance system uses machine
learning algorithms and computer vision techniques to analyze the video feed from CCTV cameras,
detecting any unusual activity or behavior such as a person entering a restricted area, wandering around
suspiciously, or leaving an object unattended. The Intruder Alert system is a critical component of this
smart CCTV surveillance system, providing real-time alerts to security personnel in case of suspicious
activity. This report discusses the features of the Intruder Alert system and its implementation using
Python programming language.

The Intruder Alert system boasts several key features: real-time detection, automatic alerts, support
for multiple cameras, customizable settings, and easy integration with other security systems. Real-
19
time detection uses machine learning algorithms and computer vision techniques to analyze video
feeds from CCTV cameras continuously. The system automatically sends alerts to security personnel
via email, SMS, or other notification methods whenever suspicious activity is detected. It can support
multiple CCTV cameras simultaneously, analyzing video feeds from various sources and providing
alerts based on the detected activity. Users can customize the system's settings, such as sensitivity
level, threshold, and other parameters, to optimize performance. Additionally, the system can be easily
integrated with other security systems, including access control, fire alarms, and other surveillance
systems, enhancing overall security measures.

Implementing the Intruder Alert system using Python involves several steps and the use of various
libraries such as OpenCV, TensorFlow, and Keras. OpenCV is a computer vision library that offers
numerous algorithms for image and video processing, while TensorFlow and Keras are machine

21
learning libraries that provide tools for creating and training machine learning models. The
implementation process begins with video capture, where the video feed from the CCTV camera is
captured using OpenCV and either stored or processed in real time. Motion detection is then performed
using OpenCV to identify any movement in the video feed. Object detection follows, utilizing
13
TensorFlow or Keras to identify persons, vehicles, or other objects in the video feed. When suspicious
activity is detected, the system generates an alert that can be sent via email, SMS, or other notification
methods. The system’s settings can be customized based on specific user requirements, allowing for
adjustments in sensitivity, threshold, and other parameters to optimize performance. Additionally, the
13
system can be integrated with other security systems for a comprehensive security solution.

4
The real-time alert module of the Intruder Alert system uses information from other modules, such as
4
object detection, tracking, and classification, to detect user-specified alerts. The key feature of this
34
module is its extensible design, which is illustrated by the design of the directional motion alert. When
4
a user specifies a directional motion alert, a directional motion alert manager is instantiated to ensure
4
correct monitoring of the scene. This manager ensures that each tracked object has a corresponding
Object Track Observer, which is deleted once the object exits the scene. The Directional Motion Object
4
Track Observer measures the direction of the object's motion and compares it to the user-specified
direction. If the motion matches the specified direction, a real-time alert is issued, signaling that the
alert conditions have been met. The general structure of this module, involving the alert manager,
object track observer, and track manager, underpins the functionality of many alert types within the
system, providing robust and customizable surveillance capabilities

22
Sagar Pandey, Sarah Irshad, and Sanjay Kumar Singh, in their 2021 paper "Smart CCTV System" on
Research Gate, emphasize the critical role of security in contemporary life. With homes often left
unattended due to busy schedules, many individuals rely on CCTV cameras for protection. In smart
cities, surveillance footage is essential for preventing and investigating crimes. Traditional video
surveillance systems, which only record without analyzing the captured content, are less effective. The
paper introduces a sensor camera designed to detect motion and physical movements within its view.
This camera's monitoring system can identify approaching objects using adaptive background removal
technology, which adjusts to changes in the scene. The system updates the background frame based on
prior background intensity and analyzes the method's effectiveness. By recording only the frames with
movement, the system optimizes its storage capacity.

8
Dr. S. Hussain's 2015 paper "Smart Surveillance System using Thing Speak and Raspberry Pi" on
8
Research Gate proposes a portable monitoring system for home and office security. The design utilizes
14
hardware components such as a Raspberry Pi (model B), a gyro sensor, and a Raspberry Pi camera.
When motion is detected, the Raspberry Pi controls the camera to capture an image and sends an alert
email with the image to the user via a Wi-Fi adapter, as programmed in a Python environment.
8
Simultaneously, the sensor's real-time data is visualized in charts on Thing Speak. Once programmed,
the system operates independently without needing a PC.

In their 2016 paper "Smart CCTV Camera Surveillance System" published in the International Journal
of Science and Research (IJSR), Amol V. Nagime and Patange A. D. describe a surveillance system
that allows wireless monitoring and control of CCTV cameras using an RF module. The system
captures video through an embedded multitask operating system, compresses the digital video using
the MJPEG algorithm, and allows users to view the video directly on a TV. Authorized users can
control the camera and monitor motion detection via a common gateway interface.

30
Arun Hampapur, L. Brown, Jonathan Connell, S. Pankanti, Andrew Senior, and Y. Tian's 2004 paper
34
"Smart Surveillance: Applications, Technologies, and Implications" explores the use of automatic
video analysis technologies in surveillance applications. The paper addresses key questions about
smart surveillance, including its applications, system architectures, key technologies, technical
challenges, and implications for security and privacy.

18
K. Patel and M. Patel's 2021 paper "Smart Surveillance System using Deep Learning and Raspberry
57
Pi," presented at the 8th International Conference on Smart Computing and Communications (ICSCC),

23
discusses the enhancements of regular CCTV cameras using deep learning and IoT. The authors
developed customized deep learning models to upgrade CCTV cameras, enabling them to detect fire
and weapons, thus making surveillance systems more effective.

In their 2020 paper "Smart Surveillance and Tracking System" presented at the IEEE 23rd International
2
Multitopic Conference (INMIC), S. U. Ahmed, H. Khalid, M. Affan, T. A. Khan, and M. Ahmad
introduce an intelligent system for face recognition and tracking. The system processes real-time video
12
streams, detects motion, and uses dual-axis pan-tilt servos to track individuals. The recorded video is
12
synchronized with cloud storage, and mobile alerts are generated for unrecognized faces. In the absence
12
of internet connectivity, the system creates a database file and sends an audio notification to the security
room. Additional features include speech recognition and voice transmission to activate surrounding
12
lights, aiming to replace costly security systems using a Raspberry Pi 3B+ microcomputer.

2.4 INFERENCE FROM LITERATURE SURVEY

58
The literature survey provides valuable insights into the current state of research and development in the
field of smart CCTV surveillance systems. It indicates that smart surveillance systems have diverse
applications, including home and office security, crime prevention, and investigation. Unlike traditional
video surveillance, which merely records footage without analysis, smart surveillance systems utilize
automatic video analysis technologies, such as motion detection and face recognition, to analyze video
footage in real-time. The surveyed papers highlight various proposed smart surveillance systems that
46
differ in their hardware, software components, and functionalities. For example, the Smart Surveillance
5
System using Thing Speak and Raspberry Pi, proposed by Dr. S. Hussain, employs a Raspberry Pi model
B, a gyro sensor, and a Raspberry Pi camera to monitor motion and send alert emails with captured
images. The system also visualizes real-time sensor data through charts on Thing Speak. In contrast, the
49
Smart CCTV Camera Surveillance System by Amol V. Nagime and Patange A. D uses an RF module
for wireless monitoring and control of the CCTV camera. This system captures digital video through an
embedded multitask operating system and compresses it using the MJPEG algorithm, allowing users to
view the monitored video on a TV. Authorized users can control the camera and observe motion
detection via a common gateway interface.
The surveyed papers also suggest that deep learning and IoT can significantly enhance the functionality
5
of traditional CCTV cameras. For instance, the Smart Surveillance System using Deep Learning and
Raspberry Pi, proposed by K. Patel and M. Patel, upgrades regular CCTV cameras with customized
deep learning models, enabling them to detect fire, weapons, and face masks, which is particularly useful
24
during the COVID-19 pandemic. This system also incorporates IoT (Raspberry Pi) and AWS EC2 Cloud
Architecture to enhance user-friendliness. Additionally, the surveyed literature addresses the technical
challenges associated with smart surveillance, such as system architecture, key technologies, and
security and privacy issues. Hampapur et al. (2004) discuss these challenges, including scalability,
robustness, and integration with other systems, as well as potential security and privacy implications,
such as unauthorized access to personal data and privacy violations.
Overall, the literature survey suggests that smart CCTV surveillance systems offer numerous
advantages, including enhanced security, crime prevention, and investigation capabilities. Despite
variations in hardware, software components, and functionalities among the proposed systems, they all
utilize automatic video analysis technologies like motion detection and face recognition for real-time
video analysis. To develop effective and secure smart surveillance systems, it is essential to address
technical challenges, including system architecture, key technologies, and security and privacy
concerns.

The field of smart CCTV surveillance systems is rapidly evolving and presents several open
problems that researchers and developers are currently working on addressing. In this section, we
will discuss some of the open problems in smart CCTV surveillance systems.

Privacy Concerns
13
 Violation of Privacy: Smart surveillance systems use advanced technologies such as facial
recognition, object detection, and tracking to identify and track individuals, which raises
significant privacy concerns.

 Illegal Surveillance: There is a risk that these systems can be misused for unauthorized
monitoring or illegal surveillance.

 Privacy Regulations: Ensuring that proper privacy regulations and laws are in place to protect
individuals' rights is essential to address these concerns.

Scalability

 High Costs: The installation and deployment of CCTV cameras in public areas can be
expensive, making it challenging to implement on a large scale.

 Cost-Effective Solutions: There is a need to develop cost-effective solutions that can be

easily deployed in public areas without incurring high costs.
25
Real-time Analysis

 Need for Immediate Response: Unlike traditional CCTV systems that record footage for later
analysis, smart surveillance systems require real-time analysis to detect and respond to
potential threats quickly.

 Advanced Algorithms: Implementing advanced algorithms and processing power for real-
time analysis can be challenging, especially on a large scale.

Integration with IoT

6
 Data Management: Smart surveillance systems generate large amounts of data, which must
be efficiently processed and analyzed.

 IoT Devices:Integration with IoT devices can improve the accuracy of the analysis and
detection of potential threats, but it presents its own set of challenges.

Robustness
5
 Environmental Conditions: The system must be robust enough to handle challenging
environmental conditions such as low light, changing weather conditions, and occlusions.

 False Alarms: The system must be capable of managing false alarms and preventing
unnecessary alerts to maintain effectiveness.

Power Consumption

 Constant Power Supply: Smart CCTV surveillance systems require a constant power supply,
which can be particularly challenging in remote areas.

 Energy-Efficient Solutions: Developing energy-efficient solutions that can operate using

renewable energy sources such as solar power is crucial.

User-Friendliness

 Ease of Use: The system must be easy to use and operate, especially for non-technical users.

 Intuitive Interface: The user interface must be intuitive, providing clear instructions and
26
 feedback to ensure usability.

In conclusion, smart CCTV surveillance systems present several open problems that researchers and
developers are currently working on addressing. These problems include privacy concerns, scalability,
real-time analysis, integration with IoT, robustness, power consumption, and user-friendliness.
Addressing these issues will require innovative solutions and collaborations between researchers,
developers, and policymakers to develop effective and efficient smart surveillance systems that can
improve public safety while ensuring privacy and security.

27
 CHAPTER 3

METHODOLOGY

The combination of hardware, software, and database components are fundamental in designing a
sophisticated surveillance system to detect illegal activities and swiftly alerting law enforcement
agencies. Aspects of the camera hardware, for instance, are capturing live footage from different points
of view with various types of cameras. Switches and routers work as intermediaries in facilitating
network connectivity and the transfer of data seamlessly between cameras, storage units, and other
ancillary components. Storage units store video footage in the archive where the video data might be
analyzed in future operations. Monitors display live feeds through the real-time monitoring screen.
Power supplies provide all the required electricity to system components without any interruptions. GPS
sensors add another dimension of enhanced situational awareness to the system. On the software side, it
involves both the operating system, which controls the distribution of resources by hardware resources
and the software application running on that computer, and the networking capability that will allow
easy communication between the system components over network channels. Load balancers help
balance out the incoming traffic to multiple servers, which makes the system efficient and available.
Web servers, along with UI and API access, host both the database setup. Socket programming provides
real-time communication between the disparate components of the system. Database component plays
a critical role in information management: RDBMS stores structured data like user information and
historical activity logs, while S3 storage can store vast volumes of unstructured data, like recorded video
footage and images. These load balancers ensure better query distribution across several servers and are
helpful in system performance and scalability. Cloud servers provide the necessary flexibility and
scalability for the hosting environment of databases and connected components. Comprehensive logging
mechanisms keep track of all system events, errors, and user actions, providing audit, troubleshooting,
7
and analysis purposes for ensuring the reliability and integrity of the surveillance system.
Of course! Designing an advanced crime detection system using computer vision and AI demands
certain components. Below follows a detailed outline of the system requirements and architecture for
your project.

28
 NO

BEHAVIOUR ABNORMAL
ACTION DETECTION
MODELLING

YES

ANALYZING
LEVELS

WEB INTERFACE
(CCVT MONITORING, MOBILE APP
DATA STORING, (LOCATION
CREATE REPROT) MONITORING)

Fig 3.1: Abnormal Detection process.

 Smart Camera:
The smart camera is equipped with high-resolution imaging capabilities, ensuring clear and
detailed video footage both during the day and at night, thanks to its advanced night vision
features. This camera boasts sophisticated image processing capabilities that facilitate real-time
analysis of the captured footage. It seamlessly integrates with various computer vision
algorithms, enabling it to perform object detection, facial recognition, and anomaly detection
with high accuracy. Additionally, the smart camera supports network connectivity via Wi-Fi or
Ethernet, allowing for efficient data transmission to central servers or cloud storage systems for
further analysis and storage.

 AI and Computer Vision Algorithms:

19
The AI component of the system employs state-of-the-art computer vision algorithms designed
to detect and identify suspicious objects or activities within the surveillance area. These
algorithms include advanced object detection techniques that can recognize potential threats and
unusual items. Facial recognition algorithms are also incorporated to identify individuals who
29
 may be involved in criminal activities, aiding in quick and accurate identification. Anomaly
detection algorithms play a crucial role in identifying abnormal behavior patterns that could
indicate potential security threats. The system's machine learning models are continuously
updated and improved, ensuring that detection accuracy enhances over time as the system learns
from new data.
 Alert Notification System:
The alert notification system is integral to the smart surveillance solution, facilitating immediate
responses to detected threats. It integrates seamlessly with various alerting mechanisms, ensuring
that alerts can be promptly communicated to relevant authorities such as the police, fire brigade,
and ambulance services. The system utilizes advanced communication protocols to transmit
alerts, which include geo-tagging to provide precise location information. This ensures that
response teams can quickly and accurately locate the incident. Additionally, the alert notification
system supports a priority-based alerting framework, allowing different types of incidents to be
categorized and responded to based on their severity and urgency.

 Web Application:
The web application provides a secure portal through which authorized users can access
comprehensive crime data. The application features a user-friendly dashboard that displays both
real-time and historical data, making it easier for users to monitor and analyze security events.
26
Advanced search and filter functionalities are included to enable efficient data retrieval, allowing
users to quickly find specific information. The web application also incorporates role-based
access control, ensuring that user permissions are managed effectively and that sensitive
information is only accessible to authorized personnel.

 Mobile Application for Authorities:

A dedicated mobile application is available for law enforcement and other authorities, delivering
real-time alerts directly to their devices. This application integrates GPS functionality, aiding in
the efficient navigation to crime scenes and ensuring a swift response. The app can also provide
detailed information about criminal acts, including multimedia evidence such as photos and
3
videos. Secure authentication and access control features are implemented to protect the data and
ensure that only authorized personnel can access sensitive information.

 Cloud Storage:
The system utilizes cloud storage to manage and store large volumes of crime data securely. This
30
 approach ensures data integrity and compliance with privacy regulations, safeguarding sensitive
information from unauthorized access and breaches. Regular backups are performed, and
redundancy measures are in place to prevent data loss, ensuring that critical data is always
available when needed.

 Machine Learning Training Model Pipeline:

The machine learning training model pipeline is designed to ensure continuous improvement of
the AI models used within the system. This pipeline facilitates the regular updating of models
with new data, enhancing their accuracy and effectiveness over time. By continuously training
the models on recent and relevant data, the system can adapt to new types of threats and improve
its ability to detect and respond to various security incidents effectively.

Fig 3.2: Abnormal Detection Training Process

31
3.2 DEVELOPMENT ENVIRONMENT SOFTWARE

Operating System: Windows 11

Windows 11 has been selected as the operating system for development due to its extensive and diverse
software ecosystem, which surpasses that of any other platform. The wide variety of available software
provides users with a multitude of options, fostering healthy competition among software developers.
This competitive environment motivates developers to push boundaries and create the best possible
programs, as anything less than the best will lead users to choose alternative solutions. This drive for
excellence results in high-quality software that effectively meets users' needs. Additionally, Windows
41
11 offers a user-friendly interface, robust security features, and seamless integration with a variety of
hardware and software, making it an ideal platform for development.

Software Used: Python

Python has been selected as the primary programming language for this project due to its high-level,
interpreted nature and its emphasis on readability and simplicity. Created by Guido Van Rossum in
1991, Python has rapidly grown to become one of the most popular programming languages
worldwide. Its power, flexibility, and user-friendliness make it suitable for a variety of applications.
69
Python boasts an active community that provides extensive support and resources, which contributes
45
to its widespread adoption. The language supports multiple programming paradigms, such as
procedural, object-oriented, and functional programming, making it highly versatile. Additionally,
Python's automatic memory management enhances efficiency and simplifies development.

Machine Learning and Neural Networks

27
Python is especially well-suited for developing machine learning and neural network models due to
42
its comprehensive libraries and frameworks like TensorFlow, Keras, and PyTorch. These tools offer
powerful capabilities for building, training, and deploying machine learning models. Python’s
simplicity and readability facilitate experimentation with different algorithms and architectures,
speeding up the development of innovative solutions.

44
Machine learning involves algorithms that enable computers to learn from data and make predictions
6 3
or decisions. Neural networks, a subset of machine learning, are computational models inspired by
32
the human brain's structure and function. These networks consist of interconnected layers of nodes
31
(neurons) that process and transform input data to generate outputs. Neural networks are particularly
effective for tasks such as image and speech recognition, natural language processing, and predictive
analytics.

28
Generative Adversarial Networks (GANs)

Generative Adversarial Networks (GANs), introduced by Ian Goodfellow in 2014, are a type of
1
neural network architecture. GANs consist of two neural networks: the generator and the
discriminator. These networks are trained simultaneously in a competitive environment. The
generator creates synthetic data samples, while the discriminator evaluates these samples against real
20
data, providing feedback to the generator. This adversarial process continues until the generator
65
produces highly realistic synthetic data. GANs have significantly impacted fields such as image
generation, video synthesis, and data augmentation, enabling the creation of high-quality, realistic
content from limited data sources.

Overall, using Windows 11 as the operating system alongside Python as the programming language
offers a robust and versatile foundation for developing advanced applications. This combination is
particularly effective for machine learning, neural networks, and GANs. The supportive ecosystem
and abundant resources available for these technologies empower developers to create innovative
and effective solutions to meet diverse needs.

3.3 MODULES
 MAIN MODULE
 FIND_MOTION MODULE
 IDENTIFY MODULE
 IN-OUT MODULE
 MOTION MODULE
 RECORD MODULE
 RECT_NOISE MODULE
 SUSPICIOUS MODULE

MAIN MODULE

The Smart CCTV Surveillance System is designed to provide enhanced security and monitoring
through an advanced graphical user interface (GUI). The system leverages Python's Tkinter library for

33
GUI development and the Python Imaging Library (PIL) for image processing. The system's main
functionalities include motion detection, noise detection, recording, and neural network-based
identification, making it a comprehensive solution for modern surveillance needs.

System Architecture
The Smart CCTV Surveillance System consists of several key components:

1. User Interface (UI)

2. Functional Modules
3. Image Processing and Handling

User Interface (UI)

The UI is the central component that allows users to interact with the system. It is built using the
Tkinter library, which provides tools to create windows, frames, labels, and buttons. The UI is designed
to be intuitive and user-friendly, with clearly labeled buttons and visual icons for each functionality.

Window Initialization
Main Window: The main window is created using `tk.Tk()`, which serves as the root window for the
application. The title, icon, and size of the window are set to provide a professional and cohesive
appearance.
Frame: A frame (`frame1`) is used to organize the layout of various UI elements. It acts as a container
for labels and buttons.

Label Title: A label with the text "Neural" is added to the frame. The label uses a bold, large font to
make the title prominent and clear.
Application Icon: An icon image is loaded and resized using PIL. This icon is displayed next to the
title to enhance the visual appeal of the UI.

Each button in the UI is associated with a specific functionality. Buttons are created using `tk.Button()`
and are placed within the frame using a grid layout. Images are loaded and resized using PIL, and then
set as icons for the buttons. The buttons are linked to their respective functions using the `command`
parameter.

34
Monitor Button: Linked to the `find_motion` function, this button initiates the motion detection
feature.
Rectangle Button: Associated with the `rect_noise` function, it enables noise detection within a
specified rectangular area.
Noise Button: Executes the `noise` function to detect disturbances or noise in the surveillance feed.
Record Button: Activates the `record` function, allowing the system to start recording the video feed.
In Out Button: Calls the `in_out` function to track entry and exit movements within the monitored
area.
Identify Button: Linked to the `maincall` function, it utilizes neural networks to identify objects or
individuals in the video feed.
Exit Button: This button calls `window.quit`, which closes the application.

Functional Modules
The system integrates several functional modules, each designed to perform a specific task related to
surveillance.

Motion Detection (`find_motion`)

This module analyzes the video feed to detect any motion. It likely compares consecutive frames to
identify changes, signaling movement. This is crucial for real-time monitoring and alerting.

Rectangle Noise Detection (`rect_noise`)

This module focuses on detecting noise or disturbances within a predefined rectangular area of the
video feed. It is useful for targeted surveillance of specific zones.

Noise Detection (`noise`)

This module scans the video feed for any unusual noise or activity. It helps in filtering out irrelevant
movements or sounds, ensuring that only significant events are captured.

Recording (`record`)
The recording module enables the system to capture and store video footage. This functionality is
essential for reviewing incidents and maintaining a record of events.

In Out Tracking (`in_out`)

This module tracks the entry and exit of objects or individuals within the surveillance area. It is

35
 beneficial for monitoring movements and ensuring security in restricted zones.

Identification (`maincall`)
This advanced module employs neural networks to recognize and identify objects or individuals in the
video feed. It enhances the system’s capability to provide accurate and real-time identification, adding
a layer of intelligence to the surveillance system.

Image Processing and Handling

The system uses the Python Imaging Library (PIL) for handling and displaying images. Images are
loaded, resized using the `Resampling.LANCZOS` method for high quality, and then converted to a
format compatible with Tkinter. This ensures that the UI remains visually appealing and functional.

2
Conclusion

The Smart CCTV Surveillance System is a robust solution that integrates multiple functionalities into
a cohesive and user-friendly interface. By leveraging Python’s Tkinter library and PIL, the system
provides advanced surveillance features such as motion detection, noise detection, recording, and
neural network-based identification. This comprehensive approach ensures that the system can
effectively monitor and secure various environments, making it an ideal choice for modern
surveillance needs.

FIND_MOTION MODULE

Introduction

63
The `find_motion` function is designed to detect motion in a video feed captured by a webcam. This
function is a crucial part of a CCTV surveillance system, providing real-time monitoring and alerting
32
capabilities. The function uses the OpenCV library for video capture and image processing, making it
an efficient and effective solution for motion detection.

System Architecture

The `find_motion` function operates through several key steps:

36
1. Initialization
2. Frame Capture and Preprocessing
3. Motion Detection
4. Motion Confirmation and Action
5. Display and Cleanup

Detailed Explanation

Initialization

The function begins by initializing key variables and the video capture object.

-Variables: `motion_detected` and `is_start_done` are boolean flags used to manage the state of motion
detection and timing.
Video Capture: `cap = cv2.VideoCapture(0)` initializes the video capture from the default webcam.
Frame Buffer: `check` is an unused list, which might be intended for storing frames or checks in future
implementations.

Frame Capture and Preprocessing

Initial Delay: `time.sleep(2)` provides a 2-second delay to allow the camera to stabilize before capturing
frames.
Frame Capture: The first frame (`frame1`) and its grayscale version (`frm1`) are captured to serve as the
baseline for motion detection.

Motion Detection

The function enters a loop where it continuously captures and processes frames to detect motion.

Frame Capture and Conversion: Each new frame (`frm2`) is captured and converted to grayscale.

Thresholding: The difference image is thresholded to create a binary image where changes are
highlighted.
Contour Detection: Contours are detected in the binary image to identify regions of motion.

37
Motion Confirmation and Action

Contour Filtering: Contours with an area less than 25 pixels are ignored to reduce noise.
Motion Detected: If more than 5 contours are detected, motion is flagged, and a message is displayed.
Motion End Detection: If motion was previously detected but the number of contours drops below 3, a
timer is started to confirm the end of motion.

Display and Cleanup

Display: The threshold image is displayed in a window named "winname".

Frame Update: The current frame (`frm2`) is updated to be the previous frame (`frm1`) for the next
iteration.
Exit Condition: The loop breaks if the 'Escape' key (key code 27) is pressed.
Conclusion

The `find_motion` function efficiently detects motion by comparing consecutive video frames. It uses
frame differencing, thresholding, and contour detection to identify changes. The function also includes
logic to confirm the start and end of motion, providing robust and accurate motion detection. This makes
it a vital component of the Smart CCTV Surveillance System, ensuring real-time monitoring and
security.

IDENTIFY MODULE

Introduction

The Face Recognition System for CCTV Surveillance is designed to identify and recognize individuals
captured by a camera feed in real-time. This system utilizes the OpenCV library in Python for face
detection, image processing, and recognition tasks. The system allows for the collection of training data
from users and subsequent identification of known individuals from the trained model.

System Components

38
The system consists of several key components:

1. Data Collection: Allows users to input their name and ID, capture facial images, and store them as
training data.
2. Model Training: Trains a face recognition model using the collected data.
3. Face Identification: Identifies known individuals from the trained model in real-time camera feeds.

Detailed Explanation

Data Collection (`collect_data` Function)

This function captures facial images of individuals and stores them as training data for the face
recognition model.

Input: The user provides their name and ID through a graphical interface.
Capture: The system captures facial images using the webcam and detects faces using the Haar cascade
classifier.
Storage: Captured facial images are saved in a directory named "persons" with filenames containing the
user's name and ID.
Limitation: Data collection stops either when the user presses the 'Escape' key or when the specified
number of images (300) is reached.

Model Training (`train` Function)

This function trains a face recognition model using the collected facial images.

Model Creation: Initializes a LBPH (Local Binary Patterns Histograms) face recognizer.
Data Loading: Loads facial images from the "persons" directory and extracts labels and IDs from
filenames.
Training: Trains the recognizer using the loaded facial images and their corresponding IDs.
Model Saving: Saves the trained model to a file named "model.yml".

Face Identification (`identify` Function)

39
This function performs real-time face recognition on camera feeds to identify known individuals.

Model Loading: Loads the pre-trained face recognition model from the "model.yml" file.
Face Detection: Detects faces in the camera feed using the Haar cascade classifier.
Recognition: Recognizes faces using the trained face recognition model and assigns labels to the
detected faces.
Display: Displays the camera feed with bounding boxes around recognized faces and their corresponding
labels.
Limitation: Face identification continues until the user presses the 'Escape' key to exit the system.

User Interface (`maincall` Function)

This function creates a graphical user interface (GUI) using Tkinter to interact with the system.

Inputs: Provides input fields for users to enter their name and ID.
Buttons: Includes buttons to add a new member (collect data) and start identification with known
members.
Interface Layout: Organizes the elements in a structured layout for ease of use.
Interaction: Allows users to input their data and initiate data collection or face identification processes.

Conclusion

The Face Recognition System for CCTV Surveillance leverages computer vision techniques to provide
real-time identification of individuals captured by camera feeds. By collecting training data, training a
face recognition model, and implementing face identification algorithms, the system enhances security
and monitoring capabilities. The integration of a user-friendly interface using Tkinter ensures easy
interaction and operation of the system.

IN-OUT MODULE

Theoretical Explanation of the In-Out Motion Detection System for CCTV Surveillance

Introduction

40
The In-Out Motion Detection System is designed to monitor movement at a specific location captured
by a webcam in real-time. This system can detect whether an object is moving towards the left or right
direction and records the entry or exit of individuals based on their direction of motion. It utilizes the
OpenCV library in Python for video processing and motion detection.

System Components

The system comprises the following key components:

1. Video Capture: Captures video frames from the webcam.

2. Motion Detection: Detects motion by comparing consecutive video frames.
3. Directional Detection: Determines the direction of motion based on the detected contours.
4. Entry/Exit Recording: Records the entry or exit of individuals and saves corresponding images with
timestamps.

Detailed Explanation

Video Capture and Motion Detection

The system captures consecutive frames from the webcam and detects motion between them.

- Frame Capture: Captures two consecutive frames (`frame1` and `frame2`) from the webcam.
- Frame Differencing: Computes the absolute difference between the two frames to detect changes
(motion).
- Thresholding: Converts the difference image to grayscale and applies a threshold to create a binary
image highlighting areas of significant change (motion).

Directional Detection

The system determines the direction of motion based on the detected contours in the binary image.

- Contour Detection: Identifies contours in the binary image using the `cv2.findContours()` function.
- Bounding Rectangle: Finds the bounding rectangle of the largest contour, representing the moving
object.

41
 47
- Directional Thresholds: Sets directional thresholds to determine whether the object is moving towards
the left or right side of the frame.

Entry/Exit Recording

The system records the entry or exit of individuals based on their direction of motion and saves
corresponding images with timestamps.

- Directional Thresholds Check: Checks if the object's position exceeds predefined thresholds to
determine its direction (left or right).
- Entry Recording: If the object moves towards the left (exits), the system records the event, saves the
frame with a timestamp, and stores it in the "visitors/out" directory.
- Exit Recording: If the object moves towards the right (enters), the system records the event, saves the
frame with a timestamp, and stores it in the "visitors/in" directory.

User Interaction

The system operates autonomously without user interaction. However, it can be terminated by pressing
the 'Escape' key, which releases the webcam and closes all windows.

Conclusion

The In-Out Motion Detection System enhances CCTV surveillance capabilities by providing real-time
monitoring and recording of entry and exit events. By leveraging motion detection and directional
analysis, the system accurately identifies the direction of motion and records corresponding events with
timestamps. This system can be deployed in various security and monitoring applications to improve
situational awareness and event logging.

MOTION MODULE

Theoretical Explanation of the Noise Detection System for CCTV Surveillance

Introduction

42
The Noise Detection System is designed to monitor video streams from a webcam and detect noise or
movement in the environment. This system can differentiate between frames with motion (indicating
potential noise or activity) and frames without motion. It utilizes the OpenCV library in Python for video
processing and motion detection.

System Components

The system comprises the following key components:

1. Video Capture: Captures consecutive frames from the webcam.

2. Frame Comparison: Compares consecutive frames to detect differences (potential noise or motion).
3. Motion Detection: Identifies motion by analyzing differences between frames.
4. Visualization: Visualizes detected motion by drawing bounding boxes around moving objects.
5. User Interaction: Allows users to terminate the system by pressing the 'Escape' key.

Detailed Explanation

Video Capture

The system captures consecutive frames from the webcam to analyze changes between frames.

- Frame Acquisition: Captures two consecutive frames (`frame1` and `frame2`) from the webcam.
- Frame Difference: Computes the absolute difference between the two frames to detect
changes (potential motion or noise).

Motion Detection

The system analyzes the difference image to identify regions with potential motion.

Grayscale Conversion: Converts the difference image to grayscale for simplifying further processing.
Blurring: Applies blurring to the grayscale image to reduce noise and enhance motion detection
accuracy.
Thresholding: Applies a binary threshold to create a binary image, highlighting regions with significant
differences (potential motion).

43
Visualization

The system visualizes detected motion by drawing bounding boxes around regions with significant
differences.

Contour Detection: Identifies contours in the binary image using the `cv2.findContours()` function.
Bounding Rectangle: Finds the bounding rectangle of the largest contour, representing the area of
detected motion.
Drawing: Draws a bounding box around the detected motion area and labels it as "MOTION". If no
motion is detected, the system displays "NO-MOTION".

User Interaction

The system operates autonomously but allows users to terminate the process by pressing the 'Escape'
key.

Termination: Monitors keyboard input and releases the webcam and closes all windows when the
'Escape' key is pressed.

Conclusion

The Noise Detection System enhances CCTV surveillance capabilities by providing real-time
monitoring of motion or noise in the environment. By leveraging video processing techniques, motion
detection algorithms, and visualization methods, the system accurately identifies and visualizes regions
with potential motion. This system can be deployed in various security and monitoring applications to
improve situational awareness and prompt response to detected events.

RECORD MODULE

Introduction

The Video Recording System is designed to capture video streams from a webcam and save them as

44
video files for later review or analysis. This system allows users to record ongoing activities and events
in the monitored area. It utilizes the OpenCV library in Python for video capture and processing.

System Components

The system comprises the following key components:

1. Video Capture: Captures consecutive frames from the webcam.

2. Video Encoding: Encodes the captured frames into a video format.
3. Video Writing: Writes the encoded frames to a video file.
4. Timestamping: Adds timestamps to the captured frames for reference.
5. User Interaction: Allows users to terminate the recording process.

Detailed Explanation

Video Capture

The system continuously captures frames from the webcam to create a video stream.

Frame Acquisition: Retrieves frames from the webcam using the `cv2.VideoCapture()` function.

Video Encoding and Writing

The system encodes the captured frames and writes them to a video file.

FourCC Codec: Initializes a FourCC codec for video encoding using the `cv2.VideoWriter_fourcc()`
function.
Video Writer Initialization: Initializes a `cv2.VideoWriter` object to write the captured frames to a video
file.
Video Writing: Writes the captured frames to the video file using the `write()` method of the
`cv2.VideoWriter` object.

Timestamping

45
The system adds timestamps to the captured frames for reference.
Timestamp Generation: Retrieves the current date and time using the `datetime.now()` function.
Text Overlay: Adds the timestamp as text overlay to each frame using the `cv2.putText()` function.

User Interaction

The system operates autonomously but allows users to terminate the recording process.

Termination: Monitors keyboard input and releases the webcam and closes all windows when the
'Escape' key is pressed.

Conclusion

The Video Recording System enhances CCTV surveillance capabilities by providing the ability to record
ongoing activities and events in the monitored area. By leveraging video capture, encoding, and writing
techniques, the system creates video files containing captured frames with timestamps for reference.
This system can be deployed in various security and monitoring applications to record evidence, analyze
incidents, and ensure comprehensive surveillance coverage.

RECT_NOISE MODULE

Theoretical Explanation of the Region Selection and Motion Detection System for CCTV Surveillance

Introduction

The Region Selection and Motion Detection System is designed to allow users to select a specific region
of interest in the video feed and detect motion within that region. This system provides users with the
flexibility to focus on particular areas of the monitored environment and receive alerts or notifications
when motion is detected within those regions. It utilizes the OpenCV library in Python for video
processing, region selection, and motion detection.

System Components

The system comprises the following key components:

46
1. Region Selection: Allows users to select a region of interest within the video feed.
2. Motion Detection: Detects motion within the selected region and highlights it in the video feed.
3. User Interaction: Allows users to interact with the system by selecting the region of interest and
terminating the process.

Detailed Explanation

Region Selection

The system enables users to select a specific region of interest within the video feed.

Mouse Events: Monitors mouse events to capture the coordinates of the selected region.
Left Mouse Button: Records the starting coordinates (topleft corner) of the selected region when the
left mouse button is clicked (`cv2.EVENT_LBUTTONDOWN`).
Right Mouse Button: Records the ending coordinates (bottomright corner) of the selected region when
the right mouse button is clicked (`cv2.EVENT_RBUTTONDOWN`).
Drawing: Draws a rectangle on the video feed to visually indicate the selected region.

Motion Detection

The system detects motion within the selected region and highlights it in the video feed.

Frame Difference: Computes the absolute difference between consecutive frames to detect changes
(potential motion).
Blurring and Thresholding: Applies blurring and thresholding techniques to the difference image to
reduce noise and highlight regions with significant differences (potential motion).
Contour Detection: Identifies contours in the thresholded image using the `cv2.findContours()`
function.
Bounding Rectangle: Finds the bounding rectangle of the largest contour within the selected region,
representing the area of detected motion.
Visualization: Draws a bounding box around the detected motion area and labels it as "MOTION" or
"NOMOTION" based on the presence of motion.

47
User Interaction

The system allows users to interact with the region selection and motion detection process.

Termination: Monitors keyboard input and releases the webcam and closes all windows when the
'Escape' key is pressed (`cv2.waitKey(1) == 27`).

Conclusion

The Region Selection and Motion Detection System enhances CCTV surveillance capabilities by
enabling users to focus on specific areas of interest and receive alerts or notifications when motion is
detected within those regions. By combining region selection and motion detection techniques, this
system provides users with greater control and situational awareness in monitoring their environments.
It can be deployed in various security and monitoring applications to enhance the effectiveness and
efficiency of surveillance operations.

SUSPICIOUS MODULE

2
This script utilizes the YOLO (You Only Look Once) object detection model to detect objects,
specifically focusing on identifying knives and guns in a video feed from the webcam. Here's a
breakdown of the code:

1. Loading YOLO Model: The YOLO model is loaded using `cv2.dnn.readNet()` function with the
weights and configuration file paths provided as arguments. The COCO class names are also loaded
from the `coco.names` file.

2. Identifying Output Layers: The script identifies the output layers of the YOLO model. If the model's
layers are unconnected, it finds the output layers using `getUnconnectedOutLayers()`. Otherwise, it
retrieves the last layers of the network.

3. Capturing Video: The script captures video frames from the webcam using `cv2.VideoCapture(0)`.

4. Object Detection: For each frame captured, the script performs object detection using YOLO. It

48
preprocesses the frame, passes it through the network, and retrieves the outputs. It then iterates through
the detections, identifies objects with confidence higher than 0.5, and checks if the detected object is a
"knife" or "gun".

5. Drawing Bounding Boxes: If a "knife" or "gun" is detected, the script draws a bounding box around
the object and labels it accordingly.

6. Displaying Results: The script displays the processed frame with bounding boxes and labels using
`cv2.imshow()`.

7. Exiting the Program: The script exits when the 'q' key is pressed.

Overall, this script provides a real time object detection system for identifying knives and guns in a
video feed.

49
 CHAPTER 4

RESULT

FIGURE (4.1)In this we have two options in python file, first is Add member and second in Start
recognition. We also have the option if the new person is arrived then they have to add their name and
give some id based on this, our model will train on their face data which captures approx 300 frames.
After successfully training model will able to recognize the person. So when the person comes next time
then after clicking on start recognition face will be recognized by model every time persons come in
front of camera.

4.1. Output of face recognition

4.2 Output of knife recognition

50
FIGURE (4.2) This is output when any which carry suspicious things like knife and gun then the model
will able to recognize the knife and gun. Which model will draw a rectangle on the suspicious things
which will able to detect the knife and gun. This detection is indicated by the green bounding box around
the knife and the label "knife".

4.3 Motion Detected (NIGHT VISION)

FIGURE (4.3) The Model is designed to detect and highlight areas of the image where motion has
been detected. The output image has been processed to emphasize the detected motion. The white
areas in the image correspond to regions where motion has been detected. The black background
represents no detected motion, while the white areas represent the detected movement of the person.

The project focuses on enhancing surveillance by integrating advanced intelligence into traditional
CCTV cameras. This enables the detection of various movements in crowded areas, including
sophisticated object detection and motion tracking. The system autonomously identifies dangerous
activities, sending immediate alerts to nearby security stations to enhance area safety and security.
33
Leveraging cutting-edge technology, the project significantly reduces the need for human intervention,
providing a cost-effective and efficient public safety solution.

51
 4.4 Motion Detection (In Light)
FIGURE ( 4.4)This output demonstrates its ability to detect and highlight areas of motion within a
video frame. The Model successfully detects motion, as evidenced by the green bounding box around
the hand of the person in the frame. This indicates that the hand's movement was recognized as
motion.
The green bounding box highlights the detected motion, making it easy to see which part of the frame
contains movement. The text "MOTION" at the top left further reinforces that motion has been
detected.

Designed for robustness, the system operates independently, making real-time decisions without
human oversight. This autonomy enhances security by ensuring continuous monitoring, minimizing
human error, and reducing operational costs. The integration process employs machine learning and
AI algorithms to, ensuring high reliability and performance. The project demonstrates how traditional
surveillance systems can be transformed into intelligent, autonomous safety networks. By minimizing
human involvement and costs, the system offers a sustainable and scalable solution for improving
public safety. This advancement highlights the potential of intelligent surveillance to revolutionize
security measures in crowded public providing a reproactive approach to threat detection and response.

52
 CHAPTER 5

CONCLUSION

In this project report, we delved into the utilization of Python and OpenCV for crafting a sophisticated
CCTV surveillance system. Our exploration encompassed various system features, including object
detection, face recognition, intruder alert, person counter, wireless transmission, and night vision. We
also underscored the significance of employing smart surveillance systems for bolstering public safety
and security.

4
The smart CCTV surveillance system offers a myriad of advantages over traditional surveillance
systems. Firstly, it enables real-time detection of unusual activities, facilitating prompt and efficient
responses from security personnel. Secondly, the system has the capability to learn from historical data,
discerning patterns indicative of potential threats. Moreover, its scalability and modularity allow for
seamless adaptation to diverse environments and configurations.

Our examination also delved into the integration of face recognition technology within CCTV
9
surveillance systems. This technology employs algorithms to ascertain and verify individuals' identities
based on their facial features. By analyzing images or videos of faces and extracting unique features
such as eye spacing and jawline contour, these algorithms facilitate accurate identification.

The implementation of a smart CCTV surveillance system utilizing face recognition entails acquiring
high-quality video footage, which is subsequently processed using Python and OpenCV. OpenCV
provides an array of tools for face detection, employing techniques like Haar cascades to detect faces.
Subsequently, facial landmarks are identified, enabling the calculation of unique facial features.
21
Machine learning algorithms such as Principal Component Analysis (PCA) or Linear Discriminant
9
Analysis (LDA) are then employed to compare these features with a database of known faces, thereby
identifying individuals.

The deployment of such a system opens up a plethora of possibilities, ranging from security applications
to retail and healthcare. For instance, in security systems, alerts can be sent to security personnel or
doors can be automatically locked upon identifying unauthorized individuals. In retail, customer
behavior can be analyzed to provide personalized recommendations, while in healthcare, patients can be
monitored remotely for signs of illness or distress.

53
However, amidst the myriad benefits, concerns surrounding privacy and security loom large. As these
systems amass vast amounts of data, there is a risk of misuse or hacking. Addressing these concerns
3
necessitates implementing robust security measures such as encryption and access controls, alongside
ensuring compliance with pertinent privacy laws and regulations.

In conclusion, smart CCTV surveillance systems leveraging Python and other technologies herald a
paradigm shift in public safety and security. With their real-time monitoring, advanced analytics, and
machine learning capabilities, these systems offer a comprehensive and effective approach to security.
The face recognition model will be able to correctly recognize the face of the person with almost 70-
80% accuracy of the model. Which we can further improve with the advancement of the model.
Moving forward, it is imperative to address privacy and security concerns to foster responsible and
ethical usage of these systems, thereby enhancing public safety and quality of life. Continued research
and development promise to unveil even more innovative applications, further elevating the efficacy and
impact of smart surveillance systems.

54
 REFERENCES

1. Sagar Pandey, Sarah Irshad, and Sanjay Kumar Singh, “Smart CCTV System”, Research Gate,
2021.
36
2. Hampapur, Arun, L. Brown, Jonathan Connell, S. Pankanti, Andrew Senior, and Y. Tian, “Smart
Surveillance: Applications, Technologies, and Implications”, 2004.
18
3. K. Patel and M. Patel, "Smart Surveillance System using Deep Learning and Raspberry Pi", 2021
8th International Conference on Smart Computing and Communications (ICSCC), 2021.
2
4. S. U. Ahmed, H. Khalid, M. Affan, T. A. Khan, and M. Ahmad, "Smart Surveillance and Tracking
System," 2020 IEEE 23rd International Multitopic Conference (INMIC), 2020.
5. S. Choudhary, S. Dubey, R. K. Tripathi, and A. Tiwari, "Python-Based Smart Surveillance System
2
Using CCTV Camera," 2018 International Conference on Advances in Computing, Communications
and Informatics (ICACCI), Bangalore, India, 2018, pp. 974-977. doi: 10.1109/ICACCI.2018.8554805
6. S. Saha, S. Kundu, S. Sarkar, S. Saha, and S. Bandyopadhyay, "Smart Surveillance System Using
17
Raspberry Pi and OpenCV," 2016 IEEE 1st International Conference on Power Electronics, Intelligent
Control and Energy Systems (ICPEICES), Delhi, India, 2016, pp. 1-6. doi:
10.1109/ICPEICES.2016.7853665
7. D. D. Anh, N. H. Anh, V. D. Hieu, and P. T. Luong, "Python-Based Surveillance System Using
24
Motion Detection and Notification," 2018 International Conference on Advanced Technologies for
Communications (ATC), Ho Chi Minh City, Vietnam, 2018, pp. 211-216. doi:
10.1109/ATC.2018.8573877
29
8. M. C. Paul and M. R. Haque, "Python-Based Intelligent Video Surveillance System," 2019 IEEE
International Conference on Imaging, Vision and Pattern Recognition (icIVPR), Dhaka, Bangladesh,
2019, pp. 1-6. doi: 10.1109/icIVPR.2019.8880485
9. M. T. Miah, M. S. Islam, M. A. Hossain, and K. S. Ahmed, "Smart Surveillance System for
23
Intrusion Detection and Tracking Using Python," 2019 International Conference on Robotics,
Electrical and Signal Processing Techniques (ICREST), Dhaka, Bangladesh, 2019, pp. 7-12. doi:
10.1109/ICREST.2019.877701
8
10. Singh, A. K., & Yadav, A., "Smart CCTV Surveillance System Using Raspberry Pi and OpenCV,"
International Journal of Computer Science and Information Security, 18(4), 1-8, 2020.
11. Kumar, A., & Sharma, P., "Real-Time Object Detection and Tracking Using Smart CCTV
52
Surveillance System," Journal of Ambient Intelligence and Humanized Computing, 11(11), 5223-
5238, 2020.
7
12. Khayyam, H., Abbasi, M., & Mohseni, M., "A Smart CCTV Surveillance System Based on Object
55
Detection and Tracking Using YOLO Algorithm," Journal of Applied Research and Technology,
18(4), 351-359, 2020.
13. Sathish, V., Sathyamoorthy, R., & Sundaram, R., "Smart Surveillance System Using Python and
56
OpenCV," International Journal of Innovative Technology and Exploring Engineering, 10(9), 1151-
1156, 2021.
14. Soni, H., Choudhary, S., & Kaur, P., "Smart CCTV Surveillance System Using Python and
53
OpenCV," International Journal of Advanced Computer Science and Applications, 11(11), 100-106,
2020.
15. Ali, M., & Malik, A. S., "A Smart CCTV Surveillance System Using Python and Deep Learning
for Traffic Monitoring," Journal of Advanced Transportation, 2021, 1-11.
16. Pradhan, S. K., & Swain, S. K., "Smart CCTV Surveillance System for Crime Detection and
Prevention Using Python and OpenCV," International Journal of Computer Applications, 180(13), 7-
13, 2021.

56
 Similarity Report ID: oid:3618:60680463

12% Overall Similarity

Top sources found in the following databases:
10% Internet database 7% Publications database
Crossref database Crossref Posted Content database

TOP SOURCES
The sources with the highest number of matches within the submission. Overlapping sources will not be
displayed.

coursehero.com
1 <1%
Internet

ijraset.com
2 <1%
Internet

fastercapital.com
3 <1%
Internet

jetir.org
4 <1%
Internet

mdpi.com
5 <1%
Internet

"ICT for Intelligent Systems", Springer Science and Business Media LL...
6 <1%
Crossref

svce.ac.in
7 <1%
Internet

elibrary.buse.ac.zw:8080
8 <1%
Internet

Vijayalakshmi S Katti, Dandu Sweshitha, SK Mahendra, S M Akash. "An...

9 <1%
Crossref

Sources overview
 Similarity Report ID: oid:3618:60680463

dergipark.org.tr
10 <1%
Internet

escholarship.org
11 <1%
Internet

Syed Umaid Ahmed, Hamza Khalid, Muhammad Affan, Tauqeer Ali Kha...
12 <1%
Crossref

Moez Balti, Ghada Somrani, Abderrazak Jemai, Meriem. Bouhachem. "...

13 <1%
Crossref

iaescore.com
14 <1%
Internet

content.iospress.com
15 <1%
Internet

iaeme.com
16 <1%
Internet

web.mitsgwalior.in
17 <1%
Internet

katinamichael.com
18 <1%
Internet

ijisrt.com
19 <1%
Internet

"Pattern Recognition and Computer Vision", Springer Science and Busi...

20 <1%
Crossref

ifatet.uni-magdeburg.de
21 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:60680463

Istvan Nagy, Yernar Kenzhetayev, Mate Szakalos. "Facial Recognition a...

22 <1%
Crossref

al-kindipublisher.com
23 <1%
Internet

revistas.ucp.edu.co
24 <1%
Internet

isteonline.in
25 <1%
Internet

oa.upm.es
26 <1%
Internet

irjmets.com
27 <1%
Internet

geeksforgeeks.org
28 <1%
Internet

Tulay Karayilan, Ozkan Kilic. "Sign language recognition", 2017 Internat...

29 <1%
Crossref

e3s-conferences.org
30 <1%
Internet

nec-labs.com
31 <1%
Internet

www2.mdpi.com
32 <1%
Internet

Roberto Moro-Visconti. "Artificial Intelligence Valuation", Springer Scie...

33 <1%
Crossref

Sources overview
 Similarity Report ID: oid:3618:60680463

Yingli Tian. "Smart surveillance: applications, technologies and implica...

34 <1%
Crossref

docplayer.info
35 <1%
Internet

mafiadoc.com
36 <1%
Internet

clausiuspress.com
37 <1%
Internet

spandidos-publications.com
38 <1%
Internet

artemis.cslab.ece.ntua.gr:8080
39 <1%
Internet

ijrpr.com
40 <1%
Internet

management.org
41 <1%
Internet

hackerrank.com
42 <1%
Internet

researchgate.net
43 <1%
Internet

sscnotes.com
44 <1%
Internet

theseus.fi
45 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:60680463

docplayer.net
46 <1%
Internet

mdu.diva-portal.org
47 <1%
Internet

prodcraftsa01.blob.core.windows.net
48 <1%
Internet

ijsr.net
49 <1%
Internet

socher.org
50 <1%
Internet

Mohammad Salman Khan, Ayesha Imran. "The Art of Seeing: A Compu...

51 <1%
Crossref posted content

ciencia.iscte-iul.pt
52 <1%
Internet

eprints.bournemouth.ac.uk
53 <1%
Internet

ichgcp.net
54 <1%
Internet

stars.library.ucf.edu
55 <1%
Internet

vjit.ac.in
56 <1%
Internet

scholarshipsinindia.com
57 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:60680463

d-nb.info
58 <1%
Internet

ijarcce.com
59 <1%
Internet

ijmrr.com
60 <1%
Internet

limsforum.com
61 <1%
Internet

sure.sunderland.ac.uk
62 <1%
Internet

"Image and Graphics", Springer Science and Business Media LLC, 2017
63 <1%
Crossref

Samira Pouyanfar, Shu-Ching Chen. "Automatic Video Event Detection ...

64 <1%
Crossref

aiforsocialgood.ca
65 <1%
Internet

dspace.dtu.ac.in:8080
66 <1%
Internet

elibrary.tucl.edu.np
67 <1%
Internet

ir.lib.uwo.ca
68 <1%
Internet

open-innovation-projects.org
69 <1%
Internet

Sources overview
 Similarity Report ID: oid:3618:60680463

scholarworks.iupui.edu
70 <1%
Internet

frontiersin.org
71 <1%
Internet

Gunjan Pahuja, T.N. Nagabhushan, Bhanu Prasad, Ravi Pushkarna. "Ear...

72 <1%
Crossref

Sources overview