Journal Title Volume,

Issue, ISSN:

Smart Security Surveillance

Prof. Shailesh P. Bendale1, Mr.Amarnath B Chavan2, Mr.Kaustubh S Kashale3, Mr. Omkar R
gattawar4, Miss. Khushi H Udasi5

Department of Computer Engineering, NBN Sinhgad School of Engineering, Pune, India.

Abstract
Due to the wide range of facial expressions, brightness, and coloration fringing, face detection in
unrestricted situations has been problematic for years. Recent research demonstrates that deep learning
strategy knowledge may achieve amazing performance in the recognition of various devices and patterns.
The primary objective of Smart Security Surveillance is to detect face masks, Motion detection, and
Name Detection of a person. Face recognition is a very difficult task in contemporary vision,
notwithstanding how distinctively different faces are. After Face detection, the method of detecting Face
masks is enhanced and made more dependable with the help of OpenCV. The structure is enthusiastic
about the character-based renaming of persons in various photographs or montage material. An improved
monitoring system makes use of motion detection technology and updated, globally accepted algorithms.
A powerful motion assessment and monitoring system has been created to enhance the capacity to detect
motion. Electronic or mechanical devices that interact with or measure changes in the surrounding
environment can detect motion.

Keywords: Face mask detection, Motion Detection, Name Detection, OpenCV, etc.

1. INTRODUCTION

Numerous face-detection methods have been suggested over the course of several years. While
there has been a significant advancement in the ability to recognize faces under tiny alterations in
lighting, facial expression, and stance, trustworthy methods for popularity under larger excessive
variances have been elusive. Particularly in recent years, face recognition systems have been one of the
most active research topics. It covers a wide range of uses, including data security, human laptop
intelligent interaction, access management, Master Card verification, criminal identification, social
control commerce, and digital libraries[1]. In this study, real-time face mask identification using deep
learning and OpenCV is investigated.

A subset of machine learning called "deep learning" makes use of algorithms inspired by the
structure of the human brain. Deep learning assists in the deconstruction of problems in various fields.
Deep learning (CNN) offers convolutional neural networks[2], image identification, and picture
categorization. Computer detection and classification tasks primarily utilize convolutional neural
networks (CNN). Recent research on the subject suggests combining both textual and visual content
rather than using just one alone. With this technique, the problem with face recognition is separated from
the problem with face-name affiliation. To incorporate the optimal amount of Convolutional Neural
Layers for precise detection, Convolutional Neural Networks (CNN) are utilized. To makes facial
recognition ideal for high-activity areas that are concerned with the general populace, such as airports,
train stations, ATMs, public transit, and organizations of all kinds. Recognition of faces reveals who was
present. In recent years, a variety of approaches for identifying people by name have been reported
utilizing databases.

In a continuous video stream, motion detection may be done in a variety of ways. Every single
one of them is based on comparing the current video frame to one from the previous frames or with a
 background object. In this study, four methods are employed, and a comparison is done to identify the
optimum detector for successful motion detection. A machine security system's artificial way of
detecting motion was based on statistical background modeling, in contrast to human vision. We would
occasionally use the camera to take a picture (make it the current picture), compare it to a previous
image, and keep both if there was a discernible change. If not, we would let go of memory from the
previous picture and make the new picture the current picture.

Motion detection includes identifying and recording the whole moving process. The objective of
motion detection is to automatically discover movement in a certain region[3]. The goal of motion
detection is to discover movement in a certain area. This region is always represented by the field of
view, a zone of awareness in in-camera geometry. It is often referred to as a monitored environment.
The environment in this case, with its moving items and activity, is what is being emphasized. A
moving item, also known as a zone of interest, might be a person, an animal, or an object.

1.1 Problem Statement

Regardless of size, every organization and the sensitive sector has difficulties in managing its
security. We created an innovative Smart Security Surveillance System to handle many functions and
provide excellent security in areas where no one is permitted to visit. This is intended to aid with
strategic planning and will guarantee that your organization or area has the proper degree of security
and information on unauthorized access. Our system includes remote access features, including Motion
Detection if anyone tries to sneak into the area, Face Mask Detection for the person with bad intentions
or to harm the property, and raising the alarm to the admin. Additionally, Name Detection is available if
a Person is verified and details will go to the admin. In the end, this system will let you have numerous
features in a single system and improve security in harsh environments.

1.2 Objectives

The project's goal is to create a smart security monitoring system to take the place of several
traditional ways of authentication. It will offer a platform for private information and sensitive regions,
enhancing security and enabling legal access to the individual. To aid in strategic planning, a smart
security surveillance system will be created with remote access functions such as motion detection, face
mask identification, name detection, and intruder alarm. To accomplish these goals, cutting-edge digital
technologies will be used, and research, testing, and evaluation will be carried out to make sure the system
satisfies the high standards of security set by organizations like the military, government offices, and
others.

1.3 Project Scope

Smart Security Surveillance system refers to use of advanced technology to enhance the security
of homes, businesses, military, organization, government offices, and others. The project scope include
analysis, system design, privacy, maintenance and support, integration with existing system, etc. Overall
this system aim to provide a comprehensive, reliable and effective security solution for the target
environment.

2. RELATED REVIEW ON LITERATURE

Enhancing the precision of object recognition in the field of computer vision and foresight,
including the detection of the human face and eyes, is one of the most difficult and unpleasant jobs. A
system called a multi-task cascaded convolutional network (MTCNN) was created as a solution for both
face identification and face alignment. The method uses three degrees of convolutional networks, which
can recognize faces and facial landmarks like the eyes, nose, and mouth[1].

The industry has expanded to include business companies that are working together to establish
ecosystems, optimized for mobile technology, which allows IoT devices to become networked, from an
initial state of disjointed devices and no ecosystems at all. This system uses cutting-edge technology like
tensor flows deep learning and is built on a Raspberry Pi minicomputer with IOT connectivity.
Additionally, this technology offers users who access the internet a live video broadcast [2].
 Human monitoring approaches for motion analysis have utilized a wide range of human-machine
interaction learning methodologies. This uses a lot of visual and physiological elements, which often
makes training difficult. This research presents an efficient motion assessment monitoring system[3]. The
assessment takes into account a number of factors, including the subject's pace. This monitoring system is
crucial for handling subject data properly, delivering scientific data on object movement, and not just
supporting motion movement assessment.

The PIR detector is used to detect movement inside the designated area. The Raspberry Pi is then
able to take the images. The face will then be found and identified in the recorded image. The photographs
and notifications will finally be delivered to an IoT-based smartphone via a messaging application. The
anticipated systems are time-efficient, fast, and have minimal machine value. According to the
experimental findings, a true-time system employs the projected facial recognition system[4].

Biometric identifiers include DNA testing, individual markings, and to some extent, both. In any
case, facial acknowledgment stands out as something that does not impede or interfere with access. People
can recognize someone by their voice and face, thus a system that uses face and voice recognition is likely
to be more acceptable[5]. Image face recognition is easier than video face recognition. The situation is far
worse in the video. This demonstrates commotions to the character in the video. In addition, numerous
performers have played the same role at different times.

They identify important distinctions between those that had previously seen a commercial and
those that had not, and we suggest a connection to virality. The correlations between head movement and
facial behavior that were noted in the data are shown last. The architecture, data gathered, and analysis
provide a way for unobtrusively evaluating facial reactions to media material that is ecologically valid,
resistant to difficult real-world situations, and doesn't involve explicit participant recruiting or
payment[6].

The finding and identifying of 2D objects in pictures with grey levels are two significant
subproblems in computer vision. The development and training of models, computational methods for
effective implementation, and parallel implementations in physiologically relevant neural network
designs are all covered in this book. The method relies on statistical modeling and estimate and
prioritizes clarity, simplicity, and computational effectiveness [7]. Presents a variety of deformable
template models, from coarse sparse models that need discrete, quick calculations to finer-grained
models based on continuum formulations that require substantial optimization. Each model is described
in terms of a subset of reference grid points (the template), a set of permissible instantiations of these
points (deformations), and a statistical model for the data given a specific instantiation of the item shown
in the picture.

3. METHODOLOGY

In our project, we sought to create a surveillance system that, in addition to motion detection,
could send SMS messages to the user's mobile device, notify the user of an invasion, and capture video
from the moment the motion was detected.

3.1. Comparing phase

Motion detection may be used to have a job run automatically when motion is detected. For
instance, motion detection may be used in security systems to spot illegal activities or to light up a room
when people enter it. Motion may be detected using a variety of techniques, some of which are
hardware, such as infrared sensors (IR sensors)[9]. We compare the webcam's live photos to one another
in order to detect changes in these frames and predict the presence of motion in order to determine
whether any motion is present in the live room.
 Figure 1. Flow Diagram

3.2. Convolutional Neural Network

A type of feedforward neural network with convolutional computation and deep structure is a
convolutional neural network. They are one of the deep learning algorithms that serve as examples. By
stacking a series of processes including convolution, convergence, and nonlinear activation function, a
convolutional neural network (CNN) may extract high-stage semantic statistics from raw statistics entered
layer by layer11. The position of the pooling layer is intended to make the characteristic map's size
smaller. The convolutional neural network uses a pooling layer behind the convolutional layer to reduce
the amount of information, speed up network training, and decrease the amount of computational data.
The pooling operation can not only make the characteristic size extracted by the convolution layer
smaller, and reduce the amount of computing data, but can also somewhat reduce the degree of
over-becoming of the network. The completely related layer's job is to use the overall connection to map
any size function map into any size vector, allowing for the mapping of two-dimensional feature maps
into one-dimensional feature vectors.

3.2.1. Building blocks of CNN architecture

A description of the architecture of a convolutional neural network (CNN) and how it is trained.
Convolution layers, pooling layers (such as max pooling), and fully connected (FC) layers are some of the
building pieces that make up a CNN. A learnable parameter, such as kernels and weights, is updated in
accordance with the loss value through backpropagation with the gradient descent optimization algorithm.
A model's performance under specific kernels and weights is calculated with a loss function through
forward propagation on a training dataset. rectified linear unit (ReLU).

Figure 2. CNN Architecture

3.4. OpenCV

Image processing's goal is to assist a computer in comprehending an image's content. A set of

computer tools called OpenCV is mostly used for image processing. It offers de facto industry-standard
API for computer vision software. The infrared radiation that human bodies release is detected using an
infrared sensor. A free software library for computer vision and machine learning is called OpenCV
(Open Source Computer Vision Library). Computer vision is essential to artificial intelligence or serves
as its main foundation. An infrared sensor can pick up the infrared radiation that people's bodies emit.
OpenCV (Open Source Computer Vision Library) is a free software library for computer vision and
machine learning. Artificial intelligence depends on computer vision or is built mostly on it. Robots,
self-driving cars, and picture editing software all rely on computer vision. Using the OpenCV application
subsystem for Android, images were taken, processed, and analyzed.
OpenCV was created to be used in image processing. The image processing programmer was
taken into consideration when creating each function and data structure. Matlab, on the other hand, is a
reasonably flexible programming language. Almost anything in the world is available to us in the form of
toolboxes. An algorithm evaluates subsequent video frames and outputs the movement of targets between
the frames to perform video tracking.
Additionally, MATLAB, Java, and Python bindings are available for these algorithms. It focuses
mostly on facial recognition, image processing, and video recording. Open CV benefits from the
underlying heterogeneous computing architecture's hardware acceleration. Through Open CV, about 2,500
algorithms for different computer vision techniques are accessible. Any OS may utilize Open CV because
it is a cross-platform library.

3.5. Mask Detection Methodology

The two phases that make up the detection of a face mask are as follows: The first two are face
recognition and mask detection. Face mask recognition is achieved using OpenCV and the Viola-Jones
method, a machine-learning object identification technique, using Haar feature-based cascade classifiers.
Quick Object Recognition The face identification method was first proposed in a 2001 paper by Paul Viola
and Michael Jones titled Using a Boosted Cascade of Simple Features. To achieve real-time face
recognition in a video, face detection must be done for each frame of the movie. Figure 3 shows the
strategy of face detection.

When a face is found in the picture, it is preprocessed and shrunk to 224x224 pixels. In order
to assess whether a person is wearing a cover, the mouth and nose are also identified using OpenCV
spretrainedHaar course classifiers. It is uncertain whether someone is covered if their lips and nose are
readily visible since they are not donning a veil.

Figure 3. Proposed framework with detailed view.

3.5.1 Image-Based and Facial Features

It's challenging to get a high level of accuracy when labeling appearances for pictures taken in
abandoned situations. Eyeglasses, make-up, facial hair, hairstyle, and a few other individual and
environment-related variations on the face, in addition to brightness, posture, and center resolution of
photographs, cause recognition to be challenging for analysts of the range.
The program incorporates an alarm function since, if motion is detected, the user would want to
get an SMS message right away saying that the software has discovered an incursion. When it detects
any kind of motion, this alarm system immediately produces an aural alert sound in wav file format. This
aids in preventing security breaches at present. When motion is detected, the user receives an SMS and
is also given a location with pictures of the trespasser at the same time.
The face recognition phase consists of the following steps: preprocessing the face image,
vectorizing the image matrix, building the database, and classification. The data are classified using a
FeedForward Neural Network (FFNN). Figure 4 shows how the facial recognition process works.
The video he or she needs to employ as the goal. The tracker's speed, vigor, and precision are
often the key factors. Speed is not the most important problem, similar to the case of many video
conferencing preparation tools, because most video organizing and arranging tasks permit detached
management. However, a continuous face tracker is essential if the target document depends on a
sizable volume of video, such as 24 hours of continuous video recording that calls for daily organizing
[11].

Figure 4. Algorithm for Face detection part.

4. THEORETICAL AND TECHNICAL ANALYSIS

Theoretical Analysis: Using a variety of theoretical concepts and principles, the topic of smart
security monitoring is one that is continuously growing. Smart security surveillance is fundamentally
based on the monitoring and analysis of data from many sources, including video cameras, sensors, and
other devices, using cutting-edge technology. The objective is to reduce the risk of harm to persons or
property by quickly identifying and responding to security risks.
Situational awareness is one of the fundamental theoretical ideas that guides smart security
surveillance. This is a reference to the capacity to comprehend one's surroundings and recognise potential
dangers or risks. A smart security surveillance system may give real-time situational awareness by
utilising a variety of sensors and cameras, enabling security staff to react swiftly and efficiently to any
possible security threats.
Data analytics is another key idea in smart security surveillance. In order to do this, video
footage and sensor data from a variety of sources will be analysed utilising algorithms and machine
learning. Smart security surveillance systems can quickly respond and take action by analysing this data to
find trends and anomalies that may point to a security problem.

Technical Feasibility:- Technical feasibility is one of the first studies that must be conducted
after a project has been identified. The technical aspect explores if the project is within the limits of
current technology and does the technology exist at all.From a technical perspective, smart security
surveillance relies on a range of hardware and software components.

These include
Cameras: To record video of the target environment, high-resolution cameras are utilized.
Network infrastructure: To transmit data and enable real-time communication between the many
components of smart security monitoring systems, a reliable network infrastructure is needed.
Software and algorithms: Sophisticated software and algorithms are used by smart security
surveillance systems to analyze data and spot possible threats. This includes artificial intelligence (AI)
algorithms that can pick up on and adjust to shifting security contexts.

5. RESULT
5.1 Motion detection

A home or business security system called a "Smart webcam motion detection surveillance
system" may be very helpful in situations when security is an issue. The tactics that thieves and other
criminals employ to steal have significantly increased because to technological breakthroughs in the
modern world. As a result, surveillance systems need to develop to keep up with the world's changing
needs. The most cutting-edge tools in the war against theft and destruction are video monitoring and
surveillance.
This monitoring system's capacity to offer a user-friendly and efficient way gives it an edge
over competing systems. The assessment must be taken into account in all respects, including pace and
background. This approach, which can track development over an extended period, provides the most
appropriate monitoring currently available.
Motion detection is the most crucial component of digital video surveillance systems. It allows
the camera to only take pictures when absolutely necessary, saving a lot of storage space. When
unexpected motion is observed, an alert could be heard. Personnel are no longer required to continuously
monitor as a result.The outcome suggests that the morph filter performs better in the monitoring system's
motion detection. Smooth operation and quick motion detection are also features of the detection.
Depending on the imagination, there is a lot that can be done with this motion detection
monitoring system. Video surveillance is one simple but not the only application for this technology.
Additionally, it may be used for amazing things like gait analysis, behavioural analysis, and gesture
recognition.

Figure 5. The Output Format

5.2 Mask detection

To develop a depth-wise separable convolution-driven MobileNet solution for mask or no-mask
detection in face image classification is the main goal of our approach. The system divides the input
facial picture into two categories, namely mask and no-mask data, based on the source data it gets from
a number of datasets.
 Figure 6. Mask detection and No Mask image.

5.3. Name Identification

The creation of humanoid robots with more specific uses, such as Guides, Guards, Office, etc.,
is the ultimate objective of the Name detection project. In the study, Local Binary Pattern (LBP) and
NFS, an improvement of a multi-choice level blend based on consolidated components for face
recognition methods, were used. Since using the most up-to-date techniques to handle Automatic facial
Naming Using Image Processing 363 facial recognition, the histogram equalization (HE) method has
been used to remove the illumination effect from pictures during pre-processing. As a result, finding
image testers for them is a somewhat less difficult task than finding image testers for people who appear
less frequently on the internet. Prominent people, such as big names or government officials, appear
online more frequently. Smartphones are used to send notifications to the mail when a known individual
visits, and to take pictures of unexpected visitors and send them to the mail when they do. Data about
known persons have already been entered into the system. In order for known person information to be
sent via mobile devices. We get the required results as shown in Figure 7.

Figure 7. The output of Name detection.

 6. CONCLUSION AND FUTURE SCOPE

Video monitoring and surveillance are the most cutting-edge technology employed in the fight
against theft and destruction. Every square inch and passing second of the region of interest may be seen
and recorded with the aid of technology.
Motion detection is the most crucial component of digital video surveillance systems. It allows
the camera to only take pictures when absolutely necessary, saving a lot of storage space. When
unexpected motion is observed, an alert could be heard. Staff are relieved of continual monitoring as a
result. The motion detector fills the requirement for a practical, cheap security solution. Computerized
home security has a very bright future. With modern technologies, the future seems bright and easier.

This application may be used to keep an eye on any area or important object to prevent
unauthorized entry as well as to monitor how long an unauthorized person or object tries to breach the
boundary or enter the restricted area. The processes might potentially be improved by using better
approaches and algorithms, such as those that follow the item that causes motion events or make use of
optical flows, also known as picture flows. There are traces of human activity here. The prototype system's
algorithm may have caught a human body gesture or position that is desired to recognize in the future.

In the future, this system will be enhanced to mobilise resources to supply the required supplies,
equipment, and infrastructure to guarantee that every assessment is utilising the full potential of motion
analysis. To improve the dependability and efficiency of this monitoring system, this is crucial. An
efficient motion detection and assessment tool has been created for monitoring systems. The monitoring
system uses the level of motion as an input to produce an assessment for the motion detection of the
individual person. data and conclusions demonstrate that the monitoring system consistently produces data
that are pertinent to the motion that has been identified.

References:

[1] Face Detection in Extreme Conditions: A Machine-learning Approach ,Sameer Aqib Hashmi,
Department of Electrical and Computer Engineering

[2] An Intelligent Motion Detection Using OpenCV. International Journal of Scientific Research in
Science, Engineering and Technology Print ISSN: 2395-1990

[3] Motion detection analysis with four different detectors, ching Yee Yong, Rubita Sundirman. Faculty
of Electrical Engineering.

[4] Nashwan Adnan, Othman, Ilhan Aydin, “A face recognition method in the Internet of Things for
security applications in smart homes and cities”, 6th International Istanbul Smart Grids and Cities
Congress and Fair (ICSG), IEEE, 2018

[5] Automatic Face Naming Using Image Processing: A review study based on face-name detection in
images or videos, Pragya Baluni, Department of Computer Science and Engineering, Researchgate, July
2017.

[6] Crowdsourcing facial responses to online videos. IEEE Trans. Affect. Comput. McDuff, D., el
Kaliouby, R., Picard, R.W.: 99, 456–468 (2012).

[7] Amit, Y. (2002). 2D Object Detection and Recognition: Models,Algorithms, and Networks. Chicago:
Cambridge, Mass. MIT Press,2002.

[8] Design a face recognition system , The 15th International Conference on Machine Design and
Production June 19 – 22, 2012, Pamukkale, Denizli, Turkey.
[9] Detecting the presence of people in a room using
motion detection, Linus Granath, Computer Science 2016-05-30.

[10] Lu, K., Dong, L.: Using LBP histogram for face recognition on Android platform, pp. 266–268.
IEEE (2011).

[11] Le, D.-D., Satoh, S.: Unsupervised face annotation by mining the web. In: ICDM, pp. 383–392
(2008)