You are on page 1of 6

International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-

nized by Karpagam Institute of Technology, Coimbatore

AI POWERED AUTHENTICATION FOR SMART


HOME SECURITY
Mrs. S. Suganya B.E, M.E,
Assistant Professor/CSE
Chettinad College of Engineering and Technology
Karur, India
suganya@chettinadtech.ac.in

Sriram .V Malishwaran .M
Computer Science Engineering Computer Science Engineering
Chettinad College of Engineering and Technology Chettinad College of Engineering and Technology
Karur, India Karur, India
sriiiram321@gmail.com mmalishwaran2002@gmail.com

Hariharan .T
Computer Science Engineering
Chettinad College of Engineering and Technology Eliza John .A
Karur, India Computer Science Engineering
hariharantamilselvan009@gmail.com Chettinad College of Engineering and Technology
Karur, India
johnkrp31@gmail.com

Abstract— The visitor management is a modern world member, while an alarm notification is displayed to the user in
problem with its application a numerous fraud, privacy issues, case of an outsider. Then the user wear face mask means,
etc. can be quickly identified and avoided. The visitor specify the alarm to remove the mask to recognize the facial
management system using face recognition is one of the most features. We can implement the framework using deep learning
secure systems even better than CCTV cameras and wake algorithm named as Convolutional neural network and
through gate methods. The primary decision that needs to be experiment this system in real time environments.
made in the project is whether the system's cost is consistent
with its scope. The size of operations and the security needs
Index Terms — Face recognition, security, deep learning, visitor
vary depending on the location, for example, household use
management, facial features.
versus industrial use. The majority of organizations, schools,
and colleges now employ visitor management systems, but I. Introduction
thanks to remarkable improvements, this can be expanded to
include railway stations, airports, toll stations, etc. Most Visitor management has become a critical concern in the
companies with large facilities integrate visitor management modern world due to the increasing instances of fraud and
systems into their total security, which is continually expanding privacy issues. One of the most secure and efficient solutions
at a constant rate. It is suggested that the face recognition to this problem is the implementation of a face recognition
visitors' management system (FRVMS) improve home security visitor management system (FRVMS). This system utilizes
by automatically identifying strangers. Having a centralized deep learning algorithms, such as Convolutional Neural
system will make managing and monitoring processes more
effective. Cost of development is also taking into consideration
Networks (CNNs), to automatically identify and verify
as this system is not requiring any extra devices. Face visitors based on their facial features. Unlike traditional
recognition is performed utilizing the computer's built-in web methods such as CCTV cameras and wake-through gate
camera. The family of face data contained in the monitoring methods, FRVMS eliminates the need for additional devices
system's database is compared with the detailed features that and can be implemented using a computer's built-in web
have been detected and security is cancelled in case of a camera. The system can be used in various settings, ranging

`
International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-
nized by Karpagam Institute of Technology, Coimbatore
from households to industrial facilities, schools, colleges, based on facial features, even in challenging situations,
railway stations, airports, toll stations, and other large FRVMS can significantly enhance security and streamline
organizations. With a centralized database of face data,

FRVMS offers a more effective and efficient approach to visitor management processes. As technology continues to
managing and monitoring visitors, improving overall security. advance, FRVMS has the potential for widespread adoption
In this paper, we will discuss the benefits and applications of and can greatly contribute to addressing the modern world
FRVMS, including its potential to enhance home security and problem of visitor management in a more efficient and
its scalability to different environments. We will also secure manner.
highlight the cost-effective nature of the system and its
potential for real-time implementation using deep learning II. RELATED WORKS
algorithms making it suitable for a diverse range of security The importance of face recognition systems for managing
needs. visitors for tackling security issues has been highlighted by
prior research and related publications. Compared to other
Furthermore, FRVMS offers enhanced security by techniques like surveillance footage and gate wake
automatically identifying strangers and triggering alarms to mechanisms, face recognition offers a superior level of
notify users in case of an outsider. Even if a visitor is wearing security. The system is affordable, adaptable for broad
a face mask, the system can prompt them to remove the mask operational and safety needs, and has the ability to be
for facial recognition, ensuring accurate identification. The expanded to other sites. Through the use of built-in web
use of deep learning algorithms, such as CNNs, allows for cameras, centralization improves management and monitoring
robust and reliable face recognition, even in complex and procedures while obviating the need for extra hardware.
dynamic environments. Quick verification is made possible by face detection and
database comparison, with security being cancelled for
recognized members and alert notifications for outsiders. For
enhancing the system's accuracy, face masks and the
application of deep learning techniques like Convolutional
Neural Networks, have been suggested. Additional
investigation and experimenting may be helpful.

FIGURE 1. Security of smart home market prospect

The scalability of FRVMS is another notable advantage.


While t is already widely adopted in organizations, schools,
and colleges, it has the potential to be extended to various
other settings, such as railway stations, airports, toll stations, 1) KNOWLEDGE-BASED AUTHENTICATION
and more. This scalability is attributed to the continuous An authentication technique utilised in powered by AI au-
advancements and improvements in face recognition thentication systems for secure smart homes is knowledge-
technology, The cost-effectiveness of FRVMS is also a based authentication. It entails confirming a user's identifica-
significant factor to consider. Since it leverages existing tion based on their familiarity with specific data, like a pass-
hardware, such as built-in web cameras in computers, it
word, PIN, or the solutions to security questions. Know-
eliminates the need for additional expensive equipment,
ledge-based identification can be used as an additional layer
making it a cost-efficient solution compared to other security
methods. Moreover, by integrating FRVMS into an of protection in the context of security for smart homes to
organization's overall security system, it can streamline guarantee that only authorised users can access and control
visitor management processes and reduce the need for smart home systems and devices.
manual monitoring, resulting in cost savings in the long run.
Users may be asked to enter a password or PIN when check-
In conclusion, the implementation of a face recognition ing in to a mobile app or web portal for using their smart
visitor management system (FRVMS) using deep learning home controls in a knowledge-based authentication system
algorithms offers a secure, scalable, and cost-effective used for smart home security. They might also be asked to re-
solution for managing and monitoring visitors in various spond to security queries they've already set up. during the
settings. With its ability to automatically identify visitors initial setup process. The answers to these security questions

`
International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-
nized by Karpagam Institute of Technology, Coimbatore
are checked against the pre-registered answers in the system
to verify the user's identity at the time of initial setup. To
confirm a user's identification, the responses to III. DESIGN FOR PROPOSED SYSTEM & DEEP
LEARNING
A. SYSTEM STRUCTURE
these safety concerns are compared to the pre-registered re- The overall structure of the system proposed in this paper
sponses in the system. is shown in a block diagram as shown in Fig.5. The
composition of the system is largely divided into parts in
charge of registration and processing. In detail, it consists of
2) OWNERSHIP-BASED AUTHENTICATION
(1) deep learning model training, (2) visitor face detection, (3)
Ownership-based authentication in AI-powered smart home
face recognition and access verification, and (4) monitoring.
security uses ownership-based credentials to verify users' au-
thenticity and authorization. It involves unique ownership- 1) DESCRIPTION OF EACH COMPONENT
based attributes associated with users or smart home devices, (1) Part of a deep learning model used for training (derived
verified using AI-powered techniques. from a deep learning training model): Desktop - Save pictures
of loved ones and acquaintances in a dataset beforehand and
3) OBJECT CHARECTERISTIC BASED use the Tiny-YOLOv3 deep learning model to learn them.
AUTHENTICATION (2) Component for collecting video data: Uses a webcam
Object characteristic-based authentication is a type of authentic- CCTV to gather visitor video.
ation that verifies users by their attributes. A representative ob- (3) A database that stores and processes the gathered video
ject characteristic-based authentication technique is biometrics. pictures serves as the control and processing component.
Biometrics separate physical and behavioural characteristics. (4) Monitoring and post-processing: The
Physical characteristics include fingerprints, iris, vein, and vein manager's smartphone receives the processing result and
recognition, as well as other techniques like retina and hand processes it there.
form, which are used to recognize faces using facial shapes and
thermal images of each individual. Speech recognition, gait re- 2) HARDWARE CONFIGURATION DIAGRAM
cognition, and signature recognition are examples of behavi- The operation of the entire system for face detection, using
hardware specifications
oural traits. Recently, businesses have started utilizing behavi-
Processor : Intel / Ryzen processor
oural characteristics like voice, gait, and signature recognition
RAM : 4 GB
to improve biometric accuracy [12], [13], [14]. Compared to
Keyboard : standard keyboard
voice, vein, iris, and fingerprint recognition, face recognition is
Monitor : 14 / 15.6 inch monitor
more practical. The accuracy of iris recognition is higher than
that of other biometrics, and installation costs are lower than
those of speech recognition systems.
Recently, a permanent and inexpensive fingerprint re-
cognition system or a Face ID system based on the user's face
have been implemented and used nearly universally to identify
the phone's owner and enable smartphone use and access.

4) MULTI-FACTOR AUTHENTICATION
knowledge-based, or password-based, authentication is vulner-
able Because it permits using short, insecure passwords that are
frequently never changed from the password that was used to
register, knowledge-based or password-based authentication is
weak. To circumvent the weakness of passwords, extra authen-
tication measures are therefore required. By first verifying us-
ing their ID and password and then utilizing a different addi-
tional authentication method, users of multi-factor authentica-
tion can increase security. Ownership-based authentication and
object characteristic-based authentication are further multifactor
authentication strategies. With capabilities like fingerprint, face,
OTP, and random number input through SMS, cell phones in
particular are being utilized as a way of authentication.

`
International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-
nized by Karpagam Institute of Technology, Coimbatore

IV. STRUCTURE AND PROTOCOL OF THE 3. If the visitor's facial region is discovered, the visitor's
PROPOSED SYSTEM face image is saved.
A. PROTOCOL:
An artificial intelligence algorithm known as Algorithm 1 [Step of face recognition]:
uses the CCTV camera that is mounted on the front door to 1. Recognize a face in the CCTV image.
identify visitors. It will notify the user to inspect the visitor's 2. Check whether a mask is being worn if a face is not
face if it cannot recognize them after a great distance or while being recognized, and if it is because of a mask, output the
they are wearing a mask. It takes a picture of the visitor message "Please pull down your mask" or say it aloud.
during this process to record their visit. When it recognizes a 3. If a mask is not being worn, go back to finding the face
visitor's face and determines that they are a family member, it area and taking another photo of it.
unlocks the door. It asks for additional identification if the 4. If the face is now recognized, see if a family member
initial identification is unsuccessful. who is stored in the database is present.

[Step of visitor recognition]: [Steps for post-authentication process]:


1. The visitor could be a member of the family or a guest 1. Open the front door automatically and use easy
who is just dropping by a bystander at the front door, or a authentication to verify the visitor is a registered family
courier bringing something. The guest is only recognized after member.
spending a specified amount of time within the proximity 2. The manager receives the visitor's face image in the
sensor's range while standing at the front door. The system case of a guest. After verification, the manager may then
then exits standby mode and begins to work. unlock the front door. If the system is unable to identify a face
2. The system uses the web camera to snap a photo of the registered in advance, the visitor may be authenticated before
visitor in order to identify them, and the learning model the front door is opened by entering the door lock password.
developed using deep learning is used to identify the visitor's In the suggested system, the front door is opened by a device
face. that inputs the password through a touchpad when the correct
3. Following the face detection, the system snaps a photo password, which was previously stored, is input. When the
of the face's centre and saves it. This is kept so that data about password typed on the touchpad as a string of numbers
suspect individuals can be gathered. matches the password that has been previously recorded, the
front door lock system authenticates guests.
Class name Class number
Eye_area 01 [Steps for CCTV image capturing]:
Mask 02 1. Using CCTV, faces are found in the image using a
Eyebrow 03 model that has been trained in advance with deep learning.
Face(head) 04 2. The number of faces in the recorded image is
R_eye 05 determined.
L_eye 06 The system determines whether one visitor is peeking if
mouth 07 there are two or more visitors.
Attack by shoulder surfing? It is assumed that the person
in the back is snooping if they approach the front door or peer
[Step of visitor decision]: over the front visitor's shoulder at the door lock. The alert
message "Please enter the password carefully so that no one
1. It could be a guest entering the home or a delivery
else can see" will then appear on the door lock display. After
person leaving a package at the front entrance. A family
that, the user defends against this social engineering assault
member or guest can ring the doorbell or enter the password
by carefully typing the password.
for the door lock if they want to enter the house. Based on
how long a person waits at the front entrance, the algorithm 3. The message "Please input the password" is displayed if
determines if they are a customer or a visitor. The doorbell is there are more than two visitors but no peeking is found, or if
pressed to let in guests, and the door lock password is pressed there is only one visitor.
to let in family members.
2. The system determines whether the visitor is a guest or [Steps for password input]:
a member of their family. The model locates a portion of the 1. The visitor inputs the password on the touchpad.
visitor's face to identify in order to check the visitor's face. 2. If the password entered matches the password stored in
The system will produce the message "Please come closer" or the database, the front door opens.
give that message through the speaker if it is unable to detect There are numerous assaults that might be made against
that part of the face. this system because it uses face recognition to decide whether

`
International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-
nized by Karpagam Institute of Technology, Coimbatore
or not to unlock the door lock. Users can, for instance, be
verified using pictures of their relatives. As a result, the

suggested system recognizes when feature points on a face


change and determines whether or not the face is a photo, in specified amount of time, it is determined that the group of
which case user authentication is prohibited. people is a passerby if its size reduces. If not, it is assumed to
be a guest or an intruder.
3. The process of employing CCTV at the front door to Second, the person behind the visitor can sneak a peek if
verify that a guest is a family member. The suggested model the visitor at the front touches the password for user
uses a deep learning-trained model to identify a visitor's face authentication. The guest who is standing behind the lead
in a CCTV image. When a face image captured by CCTV is person must therefore be identified as either a family member
compared to the face data of family members who have or a companion.
previously registered, and the comparison value exceeds the The CCTV outputs a message to the front visitor on the
predetermined threshold, authentication is complete. display alerting them of this if the visitor's face is perceived as
Following that, the subsequent step is carried out after the approaching the door or if the CCTV detects that the visitor's
visitor information is recorded in the database access log eyes are attempting to steal the password. As a result, when
table. If more authentication is needed, the front door lock can entering a registered password, the visitor in front is informed
be unlocked by entering the door lock password, which that their password is too short if the person behind is
completes authentication. determined to be peeping.
Third, relatives register ahead of time. It gains knowledge
V. ANALYSIS AND EXPERIMENTAL RESULTS from the stored images, assesses whether a face has been pre-
OF THE PROPOSED SYSTEM registered from the video image entered into the CCTV, and
verifies the visitor's identity.
A. ANALYSIS OF USER AUTHENTICATION By matching the CCTV image with a previously recorded
SYSTEM face, an authentication result is generated probabilistically. If
When the visitor’s face is recognized with a CCTV image, the calculated result exceeds the threshold, primary
81 feature points are detected using the model that was authentication is carried out; however, if it is less than 1,
trained on the file shape_predictor_81_face_landmarks.dat. A secondary authentication is necessary due to the possibility of
linear classifier is used to locate facial landmarks including incorrect authentication. Although secondary authentication
the eyes, nose, mouth, chin line, and eyebrows using the HoG can be accomplished in a number of ways, in this study it can
feature as a face detector. The faces of registered users or be done by entering a brief password. The door is opened
family members are already saved in the database for user once this authentication is finished.
authentication. The familyTbl table in the database contains Fourth, the manager's phone can receive the CCTV image
the vector values of the feature points in the family members' in the proposed model via the smart home service. The
face photos. This function collects the feature point vector manager can see if someone is there by looking at the image.
values from the visitor's face image and compares them to the The guest cannot enter the residence with authentication if the
feature point vector values of family members that are kept in password authentication fails or if a family member's visit is
familyTbl. When the threshold value falls inside the not recorded.
acceptable range for authentication, authentication is carried In this instance, it uses a smartphone to check the guest
out. before remotely opening the door. The important image is
part of Python code that extracts 81 facial features. To saved on the server and the verified visitor's name is recorded
extract and store features from the family photo, it makes use in a log file as a result of the suggested procedure. When
of the Dlib library. The feature in the video image captured by members of the family return home, access information, use
the webcam when a visitor is waiting at the main door is pattern analysis, or have a prolonged absence, this log file
extracted by the source code, and it then verifies that the may also offer extra services.
image represents the same person by comparing the vector
value of the previously saved feature of the family to the VII. CONCLUSION
extracted vector value.
Deep learning, a key artificial intelligence technology, is
B. EVALUATION OF ANALYSIS OF SECURITY expanding quickly as it is used for speech and picture
FIELD identification. Particularly, autonomous driving and crime
The following conditions are met by the requirements prevention monitoring systems, which are emerging as future
specified in the introduction. industries, are using deep learning technology in the field of
First, based on faces discovered by CCTV, the suggested image recognition as a core technology. Regarding the use of
model can be used to count the number of visitors. After a deep learning models for image recognition, there have been

`
International Conference on Latest Trends in Science, Engineering and Technology (ICLTSET’23) on May 5&6, 2023 orga-
nized by Karpagam Institute of Technology, Coimbatore
many proposed algorithms that have enhanced and developed [2] L. L. Chambino, J. S. Silva, and A. Bernardino,
CNNs that can process images. ‘‘Multispectral facial recognition: A review,’’ IEEE Access,
We introduce several object detection algorithms, vol. 8, pp. 207871–207883, 2020
including CNN, in this study. A visitor's face is detected by

[3] Waseem, Muhammad, et al. "Face recognition for


CCTV, which then recognizes 81 feature points to produce a smart door lock system using hierarchical network." 2020
collection of vector values depending on the features. Family International Conference on Computational Intelligence
members have face images registered in advance. In this (ICCI). IEEE, 2020.
work, we built and put into practice a system that, upon [4] Li, Huoyou, et al. "UFaceNet: Research on Multi-Task
identifying a new visitor as a member of the family, unlocks Face Recognition Algorithm Based on
the front door when the gap between the recognized facial CNN." Algorithms 14.9 (2021): 268.
feature vector values between the CCTV image and the image [5] Lee, Min-hye, and Hyung-Jin Mun. "Comparison
stored in the database is smaller than the predetermined analysis and case study for deep learning-based object
threshold. detection algorithm." Int. J. Adv. Sci. Converg 2.4 (2020): 7-
Future study will focus on creating and researching a 16.
system that can identify faces while regulating entry to stores [6] Jiang, Xinbei, et al. "Real-time face mask detection
or restaurants, automatically track repeat customers, count method based on YOLOv3." Electronics 10.7 (2021): 837.
visitors, identify the amount of visitors, store visitor [7] Shah, Syed W., and Salil S. Kanhere. "Recent trends in
information, and regulate access based on visitors. user authentication–a survey." IEEE access 7 (2019): 112505-
Additionally, a system for access authentication must be 112519.
created that has a low false detection rate and can enable [8] Renuka, K. M., et al. "Design of a secure password-
simple access using face recognition rather than fingerprints based authentication scheme for M2M networks in IoT
or passwords. enabled cyber-physical systems." IEEE Access 7 (2019):
51014-51027.
REFERENCES [9] Fang, Wei, Lin Wang, and Peiming Ren. "Tinier-
YOLO: A real-time object detection method for constrained
[1] Tsai, Tsung-Han, et al. "Design of wireless vision environments." IEEE Access 8 (2019): 1935-1944.
sensor network for smart home." IEEE Access 8 (2020): [10] M. Alsawwaf, Z. Chaczko, M. Kulbacki, and N.
60455-60467. Sarathy, ‘‘In your face: Person identification through ratios
and distances between facial features,’’ Vietnam J. Comput.
Sci., vol. 9, no. 2, pp. 187–202, May 2022.

You might also like