Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

Surveillance System with Human Intrusion Detection

Lawrence Rodriques1; Swati Padmanabhan2; Sumit Prasad3; Neha Auti4; Nupur Gaikwad5
Department of Computer Engineering Fr. C. Rodrigues Institute of Technology Navi Mumbai, India

Abstract:- Security in restricted areas is essential for particularly evident when distinguishing between authorized
protecting valuable assets, sensitive information, ensuring personnel and potential intruders, highlighting the urgent need
the safety from intruders. Traditional security systems have for a more sophisticated and efficient surveillance solution.
many limitations, where they cannot authenticate whether Traditional monitoring techniques compound these challenges,
the entered person is an intruder or not. To address this proving inefficient and inaccurate, especially in responding
challenge by implementing a real-time face recognition- promptly to intrusions. This critical gap underscores the
based surveillance system, is the goal of this project. pressing necessity for a novel and automated solution that
Realtime Intrusion detection system provides surveillance leverages cutting-edge Computer Vision and Machine Learn-
for restricted and confidential areas with help of face ing techniques. The project is designed to speed up the
recognition and detection, when an intruder or identification and timely notification of intrusions, placing
unauthorized person enters the area, this system will give emphasis on enhancing surveillance analytics, strengthening
an alert to the respective in charge through various security measures, and ensuring a swift response to potential
channels, including email, messaging services, and direct threats. The goal is to significantly advance the fields of
phone calls. In this system, the OpenCV python library surveillance analytics, security, and intrusion response through
along with several algorithms are used to abstract the facial the integration of facial recognition and real-time capabili-
features and to take the input dataset. For face detection, ties. By addressing the deficiencies in traditional monitoring
the system utilizes SCRFD and YOLO, and it employs methodologies, the project aims to redefine the effectiveness
Arcface for accurate face recognition. This technique of security operations, ensuring a proactive and responsive
ensures the system can distinguish between an intruder and approach to potential intrusions. The implementation of a real-
an authorized individual entering the secured area. This time face recognition-based surveillance system is poised to
proactive approach enhances surveillance efficiency and represent a transformative leap forward, promising to signifi-
reinforces the safety and integrity of restricted areas. For cantly impact and advance the realms of surveillance, security,
instance, when an individual enters the restricted area, the and emergency response operations.
system captures and analyses their face. It then verifies
whether the detected face matches any authorized faces in B. Motivation
the registered user database. If there’s no match, the system To addresses critical security concerns in restricted
identifies the person as an intruder and promptly sends an areas by introducing a real-time face recognition-based
alert to the designated authority. To enhance accessibility, surveillance system. Traditional security systems often struggle
a user-friendly graphical interface (GUI) has been to authenti-cate individuals, leaving valuable assets vulnerable.
developed using Python’s Tkinter. Unlike un-monitored CCTVs, this system employs advanced
face detec- tion and recognition algorithms to promptly identify
Keywords:- Computer Vision, Machine Learning, Deep intruders. The timely alert mechanism, utilizing various
Learning, Human Intrusion Detection, Face Recognition, Rule- communication channels, ensures a swift response from
based Notification, Real-time Monitoring, Surveillance System, designated authorities.
Image Processing.
By overcoming the limitations of manual video tracking,
I. INTRODUCTION the system enhances surveillance efficiency. Additionally, a
user- friendly graphical interface has been developed to make
A. Background the technology accessible, reinforcing the safety and integrity
The widespread integration of closed-circuit television of restricted areas.
(CCTV) cameras highlights the critical role of surveillance
systems in ensuring the security of private spaces. However, C. Aim and Objective
manually identifying authorized and unauthorized individuals, The project aims to develop a real-time intruder detection
along with detecting potential intruders within the extensive and alert system that enhances traditional CCTV systems by
CCTV network, presents significant challenges. Traditional integrating advanced technologies such as machine learning
security systems, responsible for safeguarding valuable assets and computer vision. The primary focus is on efficient face de-
and ensuring the security of restricted areas, face limita- tions tection and recognition, allowing the system to autonomously
in authenticating individuals. This challenge becomes identify unauthorized individuals within predefined zones. This

IJISRT24APR1519 www.ijisrt.com 3231

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

approach aims to provide early alerts to administrators and Further research built upon these findings, proposing a
security professionals, thereby improving the overall effi- comprehensive automated surveillance and intrusion detection
ciency and reliability of security measures in private spaces. solution for warehouse environments. This system integrated
Our key objectives Involve: the YOLO (You Only Look Once) algorithm for object
detection with the ArcFace algorithm for face recognition intoa
 User-Friendly Interface: Design and develop an user- cohesive framework, enabling continuous monitoring of
friendly interface for administrators to monitor and man- individuals within a warehouse [16]. The system’s capability to
age the system effectively, providing insights into intruder promptly notify security personnel of unauthorized access
detection and overall security status. attempts played a pivotal role in mitigating security risks [12].
 Real-time Detection and Immediate Response: Implement a
system that delivers real-time detection of human intru- In addition to face recognition, advancements in real-time
sions and initiates immediate responses upon detection. The object detection and tracking technologies have led to more
system will promptly alert security personnel and sophisticated surveillance systems. A novel approach for
administrators through various channels, including email, tracking fast-moving objects within CCTV footage was
messaging services, and direct phone calls, ensuring rapid introduced, employing synthetic data generation and deep
reactions to potential security breaches. neural networks to achieve high accuracy in tracking, thereby
 Integrated Face Recognition Surveillance: Perform com- enhancing CCTV surveillance capabilities [15]. The study also
prehensive evaluations of the system’s functionality in a highlighted the critical aspect of optimizing computation time
variety of surveillance scenarios, covering both indoor and for real-time applications, essential for the effectiveness of
outdoor environments. Assess its performance under surveillance systems in dynamic settings [14].
different lighting conditions and potential occlusions to
ensure adaptability and robustness across diverse real- Concurrent research efforts have focused on integrating
world situations. attribute-based person search and recognition mechanisms
 Face Recognition API: Develop a versatile Face Recog- within surveillance systems. These systems utilize advanced
nition API with the objective of providing a scalable and machine learning algorithms and comprehensive attribute
interoperable solution, seamlessly integrable into multiple recognition models to enable detailed queries based on
systems. The API aims to offer reliable and efficient fa- cial individual attributes, time, and location [13]. Such innovations
recognition capabilities, enhancing security and user not only improve the efficiency of surveillance operations but
identification across diverse applications and platforms. also aid in proactive threat detection and incident prevention
[7].
II. STUDY OF THE SYSTEM
The field has also seen significant contributions toward
A. Related Work the development of facial recognition systems using deep
The domain of real-time face recognition-based learning neural networks. The VGG16 architecture, pre- trained
surveillancewithin CCTV systems has garnered considerable on ImageNet, has been effectively applied across various
interest due to the pressing need for enhanced security datasets for facial recognition tasks [6]. Smith and Chen (2019)
solutions in various settings, including warehouses. explored gender and age estimation models, achieving a gender
Traditional CCTV setups have been hampered by their identification accuracy of over 98% andan MAE of 4.1 years,
dependence on manual monitoring, leading to inefficiencies in demonstrating the model’s superiority over human performance
identifyingunauthorized individuals and potential intruders [1]. with VGGFace on the Morph-II dataset [6]. These studies
To overcome these challenges, recent advancements have collectively highlight the extensive applications and continuous
leveraged machine learning and deep learning techniques to progress in deep learning-based facial recognition systems [9].
automate recognition processes and improve security measures
[2].

A seminal study in this field introduced an efficient real-

time face recognition system designed for CCTV surveillance
applications, utilizing the ArcFace algorithm for its high
accuracy in face recognition tasks [17]. This system, when
combined with the SCRFD (Single-Shot Scale-Aware Network
for Face Detection), demonstrated robust performance in
detecting and recognizing faces from video streams in real-time
[4]. The system’s effectiveness was further validated under
various lighting conditions and angles, ensuring its reliability
in practical scenarios [5].

IJISRT24APR1519 www.ijisrt.com 3232

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

face within the CCTV footage, it leverages state-of-the-art

models such as YOLO (You Only Look Once) and SCRFD
(Sparse Convolutional Regression Forests for Face Detection)
to accurately identify and extract facial features. These features
undergo dimensionality reduction, optimizing efficiency and
reducing computational load.

Following feature extraction, the system utilizes the

Arcface model to compare the extracted features against the
stored features of authorized individuals. If a match is
identified, confirming the presence of an authorized individual,
an entry is meticulously logged with a timestamp, documenting
their activity within the monitored area. However, in the
event of no match, signaling the presence of an unauthorized
user, the system promptly triggers an alert mechanism. This
alert mechanism encompasses notifications sent via call and
email to designated personnel, facilitating swift responses to
potential security breaches. Overall, this comprehensive
workflow seamlessly integrates cutting-edge technology with
precise training and vigilant surveillance processes to elevate
the security posture of restricted areas.

Finally, the workflow seamlessly incorporates the human

element by integrating the security authority. This authority
assumes responsibility for real-time validation and oversight,
ensuring the system operates effectively. A unique feature
involves the security administrator or system owner receiving
intruder images via email for confirmation of intrusion
authenticity. This direct involvement allows for swift
validation, where the administrator can confirm whether the
detected intruder is genuine or not. Such a mechanism enhances
the system’s accuracy and reliability by leveraging human
judgment in critical situations. By combining human oversight
Fig 1: Workflow of Surveillance System with Human with advanced technological capabilities, the workflow
Intrusion Detection establishes a robust surveillance system that efficiently
responds to intrusion events while maintaining a high level of
III. EXPERIMENTAL SETUP accuracy and accountability.

The workflow of this surveillance system for detecting In this research, we embark on a comparative analysis
intrusions in restricted areas begins with the security focusing on two prominent deep learning models for face
administrator or system owner capturing or uploading images detection in videos: YOLOv5-face (You Only Look Once
of authorized individuals and meticulously training themodel version 5) and MTCNN (Multi-task Cascaded Convolutional
with these images. This initial step is paramount as it Networks). The rationale behind evaluating these models lies in
establishes the foundation of the system’s capability to their widespread adoption and specialized functionalities within
accurately distinguish between authorized and unauthorized the realms of real-time object detection and face detection,
persons. The hybrid model employed for training combines respectively. YOLOv5-face is acclaimed for its exceptional
advanced techniques, including Haar cascade frontal face real-time object detection capabilities, while MTCNN excels
detection and Arc face weights, ensuring a robust and precise specifically in face detection tasks. Through this comparative
facial recognition process. study, our aim is to elucidate the distinct strengths and
weaknesses inherent in each model when applied to the
Upon completion of the training phase, the extracted nuanced domain of face detection.
features from the authorized person images are stored,
constructing a reference database for comparison during In our research methodology, we deployed both YOLOv5-
surveillance operations. Users are then empowered to add or face and MTCNN for face detection within video data.
select the CCTV cameras through which intrusion detection
will be conducted. When the surveillance system detectsa

IJISRT24APR1519 www.ijisrt.com 3233

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

To ensure a comprehensive evaluation, we employed a Our approach involves the strategic application of various
standardized dataset comprising diverse video sequences cutting-edge technologies to enhance face detection and
characterized by varying lighting conditions, viewing angles, recognition capabilities within surveillance systems. Haar
and occlusions. Prior to analysis, we pre-processed the videos Cascade serves as a foundational feature-based object detection
to extract individual frames, subsequently applying both algorithm tailored for real-time applications, offering efficient
YOLOv5-face and MTCNN to each frame for face detection. computation. Trained on a diverse dataset encompassing
To facilitate experimentation with the models, we partitioned positive and negative images, Haar cascade frontal face
the CCTV footage into smaller sections, enabling focused detection excels in accurate facial recognition. By analyzing
evaluation of their performance under specific conditions and intensity patterns against predefined facial patterns in a sliding
scenarios. window fashion, the algorithm precisely locates faces and
delineates bounding boxes during user registration, ensuring
Through this systematic approach, we aim to provide precise identification and registration of authorized individuals.
valuable insights into the comparative effectiveness of
YOLOv5-face and MTCNN for face detection in video data. By Furthermore, our methodology integrates advanced
subjecting both models to rigorous testing against a YOLOv5-face (You Only Look Once version 5 specifically
standardized dataset and diverse environmental conditions, we for face detection) and SCRFD (Sparse Convolutional
seek to elucidate their respective capabilities, limitations, and Regression Forests for Face Detection) technologies for CCTV
potential applications within the domain of surveillance and face detection, renowned for their exceptional speed, accuracy,
video analytics. Furthermore, our research endeavors to offer and adaptability. YOLOv5-face’s simultaneous prediction of
practical recommendations for leveraging the strengths of bounding boxes and class probabilities makes it highly
YOLOv5-face and MTCNN to enhance face detection accuracy effective for intrusion detection, even across considerable
and efficiency in real-world video surveillance systems. distances, while SCRFD excels in face detection across diverse
scales, ensuring robust performance. This seamless integration
Table 1: Accuracy Comparison significantly enhances facial recognition capabilities, enabling
Model Accuracy more accurate user identification, particularly in scenarios with
YOLOV5-Face 99-97% distant subjects or varying facial scales.
MTCNN 97-16%

Fig 2: Haar Cascade Frontal Face

Moreover, Arcface emerges as a pivotal component in hypersphere. Arcface excels in scenarios requiring precise
user registration, being a state-of-the-art face recognition differentiation of subtle facial variations, thereby contributing
algorithm designed for precision. Trained on captured images, to heightened face recognition accuracy within surveillance
Arcface enhances accuracy by focusing on angular systems.
relationships between facial features, mapping them into a
hypersphere. Through the introduction of an angular margin
during training, increased separation between features of
distinct identities is fostered, resulting in unique feature vectors
for each face and ensuring accurate grouping within the

IJISRT24APR1519 www.ijisrt.com 3234

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

IV. RESULTS & DISCUSSION register authorized users by either capturing images directly
through the system or uploading images for training the facial
Upon logging in, users are greeted with the default camera recognition model. This process ensures a tailored approach
feed, providing real-time surveillance for intrusion detection. to user authentication, enhancing the system’s accuracy in
The security administrator or owner holds the authority to recognizing authorized personnel.

Fig 3: System Homepage

Fig 4: Authorised User Registration

Users are granted flexibility in choosing between two in identifying individuals, while the latter focuses on detecting
primary modes of operation: Face Recognition Intrusion any unauthorized intrusions within the monitored area. This
Detection and Intrusion Detection. The former employs versatility caters to varying security requirements and
advanced facial recognition algorithms for enhanced accuracy preferences of users.

IJISRT24APR1519 www.ijisrt.com 3235

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

Fig 5: Intrusion Detection through Webcam

Fig 6: CCTV Feed Output

Adding new CCTV cameras to the surveillance manage their camera inventory, with options to add or delete
network is made effortless through the intuitive interface, cameras as needed. This feature ensures scalability and
allowing users to input the IP address of the camera for adaptability to evolving surveillance needs.
seamless integration. Additionally, users have the capability to

IJISRT24APR1519 www.ijisrt.com 3236

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

Fig 7: Adding Security Admin/Owner Details

Fig 8: Email & Call Alert

In the event of a security breach, timely alerts are entering relevant contact details and specifying notification
dispatched to designated security personnel via email and preferences. This proactive approach enables swift response
phone call. Administrators can easily configure alert settings by measures to mitigate potential threats effectively.

IJISRT24APR1519 www.ijisrt.com 3237

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

Fig 9: Logs & Recordings

V. CONCLUSION REFERENCES

The Surveillance System with Human Intrusion Detection [1]. Akula Surya Teja, Ginni Chandra Mohini, Dannana
project has effectively addressed the limitations of conventional Dhanunjay, Dr. P M Manohar, ”Realtime Intrusion
security systems by introducing a real- time face Detection System Using Open CV,” Journal of Survey
recognition-based surveillance system. Through the in Fisheries Sciences, Visakhapatnam, India, 2023, pp.
utilization of advanced algorithms and the OpenCV Python 2734-2740.
library, the system ensures robust surveillance in restricted [2]. G. Mallikharjuna Rao, Haseena Palle, Pragna Dasari,
areas. Its distinguishing feature lies in its ability to Shivani Jan- naikode, Dr. P M Manohar,
differentiate between intruders and authorized individuals, ”Implementation of Low Cost IoT Based Intruder
thereby significantly enhancing the security and integrity of Detection System by Face Recognition using Machine
monitored zones. By integrating SCRFD and YOLO for face Learn- ing,” Turkish Journal of Computer and
detection, complemented by ArcFace for recognition, the Mathematics Education, Vol. 12, No. 13, 2021, pp. 353-
system has achieved an impressive accuracy rate surpassing 362.
90%. However, it’s imperative to recognize that the accuracy is [3]. K. S. Krishnendu, ”Analysis of Recent Trends in Face
subject to the quality of the dataset and surveillance camera Recognition Systems,” 2023, arXiv:2304.11725.
employed. Furthermore, the development of a user-friendly [4]. Rehmat Ullah et al., ”A Real-Time Framework for
graphical interface using Python’s Tkinter enhances the Human Face De- tection and Recognition in CCTV
system’s accessibility, making it applicable across various Images,” Mathematical Problems in Engineering, 2022,
domains. This project’s success highlights the rapid evolution Hindawi.
of face recognition technology, with continual efforts aimed [5]. Edwin Jose, Greeshma M, Mithun Haridas T. P, ”Face
at improving accuracy while reducing processing overheads. Recognition based Surveillance System Using FaceNet
The API developed as part of this project holds potentialfor and MTCNN on Jetson TX2,” in Proceedings of the 5th
widespread adoption in systems requiring robust face International Conference on Advanced Computing &
recognition capabilities. Communication Systems (ICACCS), 2019.

IJISRT24APR1519 www.ijisrt.com 3238

Volume 9, Issue 4, April – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24APR1519

[6]. Philip Smith, Cuixian Chen, ”Transfer Learning with

Deep CNNs for Gender Recognition and Age
Estimation,” 2018, arXiv:1811.07344v1 [cs.CV], 2018.
[7]. Franz Cardoz et al., ”Visual Sentinel: Data Analytics for
Missing Subject Identification,” in Proceedings of the
2023 Pune International Conference, Pune, India, 2023.
[8]. Kaipeng Zhang, Zhanpeng Zhang, Zhifeng Li, Yu Qiao,
”Joint Face Detection and Alignment Using Multitask
Cascaded Convolutional Net- works,” IEEE SIGNAL
PROCESSING LETTERS, October 10, 2016.
[9]. F. Schroff, D. Kalenichenko, J. Philbin, ”FaceNet: A
unified embed- ding for face recognition and
clustering,” in Proceedings of the 2015 IEEE
Conference on Computer Vision and Pattern
Recognition (CVPR), Boston, MA, 2015, pp. 815-823.
doi: 10.1109/CVPR.2015.7298682.
[10]. T. Sanjay, W. Deva Priya, ”Efficient System for
Criminal Face De- tection Technique on Innovative
Facial Features To Improve Accu- racy Using LBPH
In Comparison With CNN,” Journal of Pharma-
ceutical Negative Results, Volume 13, Special Issue 4,
2022. DOI: 10.47750/pnr.2022.13.S03.085.
[11]. Arnab Pushilal, Sulakshana Chakraborty, Raunak
Singhania, P. Ma-halakshmi, ”Implementation of Facial
Recognition for Home Security Systems,” International
Journal of Engineering & Technology, Vol. 7, No. 4.10,
2018, pp. 55-58.
[12]. Mfundo Zuma et al., ”Intrusion Detection System using
Raspberry Pi and Telegram Integration,” in
Proceedings of icARTi ’21, 2021.
[13]. R. Prakash, P. Chithaluru, ”Active Security by
Implementing Intrusion Detection and Facial
Recognition,” in Nanoelectronics, Circuits and
Communication Systems, Lecture Notes in Electrical
Engineering, vol 692, Springer, Singapore, 2021.
[14]. Bazama A, Mansur F, Alsharef N. ”Security System by
Face Recogni- tion,” Alq J Med App Sci, 2021;4(2):58-
67.
[15]. Kajenthani Kanthaseelan et al., ”CCTV Intelligent
Surveillance on Intruder Detection,” International
Journal of Computer Applications, Volume 174, Issue
14, 2021.
[16]. R. Menaka et al., ”Enhanced Missing Object Detection
System using YOLO,” in Proceedings of the 2020 6th
International Conference on Advanced Computing and
Communication Systems (ICACCS), Coim- batore,
India, 2020.
[17]. J. Deng et al., ”ArcFace: Additive Angular Margin Loss
for Deep Face Recognition,” in Proceedings of the 2019
IEEE/CVF Conference on Computer Vision and Pattern
Recognition (CVPR), Long Beach, CA, USA, 2019, pp.
4685-4694.

IJISRT24APR1519 www.ijisrt.com 3239