Smart Attendance System using Face-Recognition

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 03 | Mar 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 954
Smart Attendance System using Face-Recognition
Mr. Rajvardhan Shendge1, Mr. Aditya Patil2, Mrs. Tejashree Shendge3
1, 2Student, Computer Engineering, Ramrao Aidik Institute of Technology(India)
3Student, Electronics and Telecommunication Engineering, Fr. C. Rodrigues Institute of Technology(India)
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Every institute, college, and organization,large or
little, has an attendance marking system. We’ve built a
method that uses the latter to simplify the former, thanks to
huge advances in the field of image processing. Face
Recognition is becoming more popular than other biometric
verification methods due to its simplicity, non-invasiveness,
and lack of touch. The system’s major goal is to identify and
recognize faces in a real-time environment,matchthemwith
data in the database, and record their attendance. This is
intended to make the time-consuming manual attendance
process more efficient. This also overcomes the issue of
authentication and proxies because biometricsareone-of-a-
kind, and facial traits used for Face Recognition are one of
them. For face detection and recognition, the designed
system uses OpenCV, dlib, Face Recognition libraries, and
One-Shot Learning, which takes just oneimageperpersonin
the database and so saves space whencomparedtostandard
training-testing models.
Key Words: face recognition, image processing, face
detection, Siamese Networks
1.INTRODUCTION
There is an inherent positive relationship betweenstudents’
attendance in schools and colleges and their academic
performance, according to research [1]. And, in order to
maintain this relationship, it is necessary to encourage their
presence and performance in the classrooms, so that
students are motivated to keep up with the progress of the
subjects being taught in class, thereby increasing their
participation in school/college. Attendance management
systems have been implemented in schools, colleges, and
universities all over the world using a variety of methods.
Despite their high usability, the practicality of thesesystems
is a little questionable. The face recognition-based
attendance system is one such system that has recently
gained traction. Face recognition is a technique for
identifying, verifying, or distinguishinga subjectbasedon an
image or video of the subject’s face. It employs a biometric
identification method that uses facial and head
measurements to verify a person’s identity.Facerecognition
biometric systems use computer algorithms to pick out
specific, distinguishing features of a person’s face, such as
the space between the eyes or the shape of the face. These
characteristics are converted into a mathematical
representation, such as an array or matrix, and compared to
the characteristics of other faces in a face recognition
database. A face encoding is data about a specific face that
differs from a photograph in that it is designed to only
include certain details that can be used to distinguish one
face from another. This system requires any device with
digital photographic technology,suchasa webcamora CCTV
camera, to generate and obtain the images and data needed
to create and record the biometric facial pattern [11].
Various facial recognition algorithms have been developed
over the years to recognize people regardless of their
environment, lighting, angle, or facial expression. Based on
its performance in other security applications, it appears to
be a promising approach for student attendance systems
that can help solve problems associated with current
systems. A system that uses facial recognition to assess
students’ attendance using machine learning algorithms is
proposed in the proposed paper. The One-Shot Learning
approach is used in the proposed system, which requires
only one image of each student to train the system that will
be used to detect their faces, generate their face encodings,
and mark their attendance. The attendance will be recorded
on an Excel sheet, which staff and faculty members will be
able to assess and evaluate. We used a pre-trained deep
neural network called face-recognition, which was built
using dlib and has an accuracy of nearly99.38percentonthe
LFW dataset [12] to implement the OneShot Learning
approach. To test the system’s stability and robustness, we
tested it on a few students from our college in various light
settings, camera settings, and occlusions.
2. LITERATURE SURVEY
Various systems are currently in use to manage and assess
student attendance at universities. Even though these
systems are extremely usable, their practicality and
constraints pose a problem in the process, as previously
stated. The following are a few of the systems in place:
2.1 Manual attendance system
Manual attendance systems are traditional systemsinwhich
a teacher or lecturer takes students’ attendance by calling
names or signing an attendance sheet. Such attendance
systems rely entirely on students acting in a fair and
consistent manner. Although it is a low-cost system, it is
extremely vulnerable to human error or manipulation. A
student may be mistakenly marked present by the teacher if
another student answers it on a roll-call, or a student can
forge signatures on the sheet, resulting in ’proxyattendance’
[16].

2.2 RFID-Based attendance system
Rfid is the abbreviation for Radio Frequency Identification.
Students are given RFID cards, which are scanned using an
RFID reader to mark their attendance at universities. The
RFID system’s main flaw is its lack of practicality. RFID tag
cards are prohibitively expensive, and purchasing them for
an entire university is not feasible. Another disadvantage is
that if students are not supervised, they can scan multiple
RFID cards on the reader, resulting in proxy attendance. If
not supervised or cared for properly, the RFID reader is also
susceptible to damage [17].
2.3 Bluetooth-based attendance system
Bluetooth-based systems collect information about the
students present in the class and record their attendance
using Bluetooth signals from their phones. This system
appears to be very practical, as nearly 95% of college
students have their phones with them. To make the system
perfect, it can also implement proxy removal methods.
However, the system’s main flaw is its lack of usability. A
Bluetooth-based device can only connect to 8 other devices
at a time. This is due to the Master and Slave concept, which
limits a device’s connection to only eight other devices at a
time. As a result, this system can only be used when the
number of students in a classroom is in the single digits[18].
After more thought, it was discovered that every existing
attendance management system had flaws that tainted the
process. Problems caused by ’proxy attendances’ will be
eliminated by using facial detection and recognition as a
parameter of attendance generation, as only those students
present in the lecture will be marked present.Becauseevery
classroom has a laptop and a webcam, the components are
also inexpensive. The main strategy is to compare the face
encodings of the image captured in real-time with those
already stored in the database, which can then be used to
mark attendance if a match is found. A Real-Time Multiple
Face Recognition using Deep Learning on Embedded GPU
System was proposed in the paper by author Saypadithet al.
[9]. Face detection and tracking were implemented using a
Convolutional Neural Network (CNN). Author Deeba et al.
[10] used a similar approach to develop a Local Binary
Pattern Histogram (LBPH)-based Enhanced Real-Time Face
Recognition system that can recognize faces in low and
highlevel images in real time. A model of an automated
attendance system was proposed by authors Akbar etal.[7].
Their system detects and counts students as they enter and
exit the classroom using a combination of Radio Frequency
Identification (RFID) and Face Recognition. It keeps track of
each student’s attendance records and provides pertinent
information as needed. Author Smitha et al. [8] used Haar-
Cascade Classifier and Local Binary Pattern Histogram
(LBPH) for face detection andrecognitionintheirautomated
attendance system. Faces were captured using a live stream
video of the class in their system, and attendance was
recorded, which could be accessed as a CSV file. For face
recognition, author Sawhney et al. [3]usea hybridalgorithm
that combines Eigenface, Principal Component Analysis
(PCA), and Linear Discriminant Analysis (LDA). The facial
features obtained through thesealgorithmscanthen beused
to identify students and, as a result, mark their attendance.
Authors Kiran et al. [6] developed a face recognition
attendance system that employs Eigenface, Haar Cascade
Classifier, and Principal Component Analysis (PCA)
algorithms. Their method was to take real-time images of
students, compare the extracted Eigenvalues to those in the
database, and mark attendance based on the recognition
result from PCA analysis. This system had a 97% accuracy
rate when tested on a database with images from 70
students. The attendance system proposed by D’Souza et al.
[4] is based on the Haar Cascade Classifier and the Local
Binary Pattern Histogram (LBPH)algorithm.Their proposed
system would take group photos of students during class
hours, perform facial segmentation and identification, and
update attendance accordingly. Harikrishnan et al. [5]
created an attendance system using the Haar Cascade
Classifier and the Local Binary Pattern Histogram (LBPH)
algorithm in another implementation of a similar system.
Their system achieved a maximum accuracy of 74% when
used in various environments such as lighting and
occlusions.
3. METHODOLOGY
3.1 One-Shot Learning Model
The One-Shot Learning Model is the foundation of our
system. It’s a classification task in which one sample is used
to classify a large number of future samples. Face-
recognition systems based on the One-Shot Learning Model
learn a rich low-dimensional feature representation known
as a face encoding, which can be easily calculated for faces
and compared for verification and identification tasks [14].
Consider the case of a face recognition system for a
timekeeping system. Images of multiple faces make up the
input dataset.Convolutional Neural Networks(CNN)-trained
models require a large number of images to train and
achieve high accuracy. If there are a few minor changes in
the dataset, these models must be trained iteratively. If a
student drops out of college, the dataset must be updated by
deleting the student’s images, and the model must be
retrained. In addition, when a new student is admitted to
college, their images must be collected, and the model must
be retrained. This procedure consumes a significantamount
of time and manpower. To address this, the One-Shot
Learning model can be used, which takes a lot less time to
train because only one image of the student is required. The
Siamese Network is widely used to implement the One-Shot
Learning Model. Figure 1 shows how the Siamese Network
performs facial recognition. A similarity function is the
foundation of any Siamese Network. A Siamese Network’s
architecture consists of two parallel networks, each takinga
different input, and combining their outputs to generate a
prediction. A Siamese Network for face recognition is a
neural network that learns a function f(d) that takes two

input images, one from the dataset called the actual image
and the other from outside the dataset called the candidate
image, and the output is the similarity between the two
images. If the two images passedthroughthenetwork havea
small distance between them, they can be classified as the
dataset’s actual image [21].
Fig. 1. Siamese Network for Face Recognition
These images are passed through similar networks called
sister networks, which are similar in terms of their
parameters and shared weights, in order to train the neural
network to learn how to compute similarities between two
images, the actual image and candidate image. These sister
networks are made up of a series of convolutional, pooling,
and fullyconnected layers that produce a fixed-size feature
vector denoted by h1 as an encoding of the actual image
Image1. The difference —h2- h1—betweenthe encodings of
the two images passed is the distance between their
encodings. The value of —h2 -h1— is relatively small if the
two images passed are of the same person. The workings of
sister networks are depicted in Figure 2.
Fig. 2. Feature vector extraction in sister networks
3.2 dlib’s HoG Face Detection
Histogram of Oriented Gradients (HoG) is an abbreviation
for Histogram of Oriented Gradients. HoG’s main idea is to
turn facial features from an image (or a real-time video)into
a vector and feed it into a classifier like SVM (SupportVector
Machine) to detect the presence of a face in an image. The
histograms of directions of gradients, or oriented gradients
of the image, are the names giventotheseextractedfeatures.
Gradients are large around edgesandcornersingeneral, and
they allow us to detect regions of interest (ROI) [20]. This
method for detecting human bodies was developed by Dalal
et al. [19]. The images are first preprocessed by being
cropped and scaled to the appropriate size. The image
gradients must be calculated as the first step in face
detection. These gradients are calculated to remove all non-
essential elements from an image, suchasbackgroundnoise,
leaving only the region of interest (ROI). Kernels are used to
compute the horizontal and vertical gradients, as shown in
fig.3. [20].
Fig. 3. Kernels applied to compute gradients [20]
After gradient computation, the image is divided into 8x8
cells to create a compact representation, making our HoG
more noise resistant. Then, for each of these cells, HoG is
calculated. The gradient’s direction inside a region is
estimated by creating a histogram from the 64 gradient
directions and their magnitudes within each region. The
histogram is divided into nine categories that correspond to
angles ranging from 0 to 180 degrees. The temperatures are
0°, 20°, 40°, 80°, and 160°[20].
While building an HoG, 3 subcases arise as follows:
1.)If the angle is smaller than 160° and it is not halfway
between 2 classes, the angle will be categorized in the right
category of HoG [20].
2.)If the angle is smaller than 160°and it is exactly between
2 classes, then the angle contributes equally to both the
bounds, and the magnitude is divided by 2 [20].
3.)If the angle is greater than 160°, the pixel isconsidered to
contribute proportionally to 160° and 0° [20].
Finally, a 16x16 block is used to normalize the image,
making it insensitive to things like lighting. The value of the
8x8 sized HoG is divided by the L2- norm of the HoG of the
16x16 block that contains it, which is a vector of length 36.
The feature vector is created by concatenatingall ofthe36x1
vectors into a single large vector that can be used to train an
SVM (Support Vector Machine) classifier and used for face
detection using the dlib library [20].

Fig. 4. Subcase 2 of HoG building [20]
Fig. 5. Subcase 3 of HoG building [20]
4. CONCEPTUAL ARCHITECTURE
Our method entails assessing student attendance using only
one image per student from the class, captured using a
webcam connected to a laptop or desktop computer. All of
the students in the class must register on the device by
entering their information, and each student’s image will be
captured and saved in the dataset. The student will be asked
to register their attendance using the device that the system
is running on before each lecture. The system will then
detect the face, calculate the encodings of the face, and
compare them to those in the dataset. The student will be
marked present for the lecture if there is a match. This
attendance information will be saved in a CSV file that the
class’s faculty/lecturer can easily access.
This process can be primarily divided into 4 stages:
4.1 Image acquisition for dataset creation
We used images of 50 students from our own college to
create our dataset. These photographs only show the
students frontal faces. Only one image is used per student.
After that, the images in the dataset are preprocessed. The
images are first cropped in preprocessing so that only the
Region of Interest (ROI) is available for further detection.
After that, the cropped images are resized to a specific pixel
position. After resizing the images, the cv2 module of the
OpenCV library is used to convert them from BGR to RGB.
Finally, the processed images are saved in the dataset along
with the student’s name.
Fig. 6. Conceptual Architecture of the System

Fig. 7. Modular Diagram of the system
4.2 Face Detection
The face-recognition model [12], which is heavily based on
dlib, is used for face detection. The color and size of the slant
eyes, the gap between the eyebrows, the distance between
the lips and the chin, and other details are noted in this
model. When all of these values are added together, a face
encoding is created, which is a vector array with 128 values.
This model is looped through the dataset in our system to
calculate the face encodings of each image.Inthe nextstepof
face recognition, this face encoding aids in identifying the
students. 128-valued face encoding vector array.
Fig. 9. 128-valued face encoding vector array
4.3 Face Recognition
Real-time image processing and detection are involved in
this step. The student’s face is detected and the student is
recognized using a webcam to record live video of the
student. Because the image is captured in real time, image
distortion can occur if a student is not fully facing the
camera. Face landmark estimation [15] is used to detect the
pose of the face, which solves the problem. There are 68
distinct landmarks on every face. The top of the chin, the
outside edge of each eye, the inner edge of each eyebrow,
and other landmarks can be found.
Fig. 10. 68 landmarks present on the face
By rotating, scaling, and shearing the face image, these
landmarks can be used to center it. This image can now be
used to calculate face encodings,whicharethencomparedto
encodings already stored in the database, and the student is
identified as such.
4.4 Attendance Generation
Fig. 11. Attendance List as observed using Google Sheets
The recognized faces are then marked present on a CSV file,
which can be generated and assessed in a soft copy format
on Excel, following the recognition process.

5. IMPLEMENTATION SETUP
Students and faculty can use our PyQT-based GUI to interact
with the system. When the students open the app, they will
be taken to a screen where they can see the livevideofeedas
well as the date and time when the attendance is taken. To
begin their attendance process, the student must first clock
in. If the system recognizes the student’s face when they
clock in, a label with the student’s name and the matchindex
will be generated around the face.
Fig. 12. GUI of application capturing real-time video
6. RESULT AND ANALYSIS
According to the implementation setup, the match index is
the smallest difference between the face encodings of the
student’s face in the live video capture and those in the
database. Figure 13 depicts the system’s implementation as
it is being worked on using real-time video capture. The
matchindex’s confidence threshold is set to 0.6 by default. If
the match index value is less than this confidence threshold,
the face will be recognized and identified asthesame.Figure
14 shows the relationship between the live image capture’s
confidence score and the face distance (i.e. match index) of
the images in the dataset. Afterthat,thestudent’sattendance
is recorded in a CSV file. On the attendance sheet, only the
names of students whosefaceswerescannedandrecognized
will be written. Any app that supports CSV files, such as
Excel, Google Sheets, or Numbers, can then be used to
evaluate this file.
Fig. 13. Live webcam capture of the Students with
Identification. The match index is mapped to the identified
student’s label in our system.
7. CONCLUSIONS
Individual classroom attendance is currently feasible using
the system we developed. It can be widely used at the
collegiate level with the necessary enhancements and the
creation of a proper database containing all of the details of
each student in the college or university. This system can be
used to manage not only students, but also faculty, staff, and
nonstaff members’ students. Another development that we
want to make sure of is the system’s complete automation.
To avoid any discrepancies, such as tampering with the
devices, the system must currentlybesupervised. Ourgoal is
to completely automate the process by using a real-time live
feed captured by a CCTV camera to mark students’
attendance without the need for manual supervision,
resulting in legitimate and untampered attendance reports.
8. FUTURE SCOPE
The system that we have developed is currently viable for
individual classroom attendance. With the required
enhancements and creation of a proper database consisting
of all the details of each student in the college or university,
it can be widely used at the collegiate level. This system can
also be used to manage the students of not only the students
but also the faculty members, staff, and non-staff members
as well. Another developmentwhichwe wishtoensureisthe
complete automation of the system. Currently, the system
has to be supervised to avoid any discrepancies such as
tampering with the devices. Our goal is to completely
automate the process by using a real-time live feed capture
using a CCTV camera, which can mark the attendance of
studentswithoutanymanual supervision,therebyproducing
legitimate and untampered attendance reports.
REFERENCES
[1] T. Fadelelmoula and Almaarefa Colleges, Riyadh, Saudi
Arabia, “The impact of class attendance on student
performance,” Int. Res. J.med.Med.Sci.,pp.47–49,2018.

[2] A. G. Menezes, J. M. D. da C. Sa, E. Llapa, and C. A.
EstombeloMontesco, “Automatic attendance
management system based on deep oneshot learning,”
in 2020 International Conference on Systems, Signals
and Image Processing (IWSSIP), 2020, pp. 137–142.
[3] S. Sawhney, K. Kacker, S. Jain, S. N. Singh, and R. Garg,
“Realtime smart attendance system using face
recognition techniques,” in 2019 9th International
Conference on Cloud Computing, Data Science
Engineering (Confluence), 2019, pp. 522–525.
[4] J. W. S. D’Souza, S. Jothi, and A. Chandrasekar,
“Automated attendance marking and management
system by facial recognition using histogram,” in 2019
5th International Conference on Advanced Computing
Communication Systems (ICACCS), 2019, pp. 66–69.
[5] J. Harikrishnan, A. Sudarsan, A. Sadashiv, and R. A. S.
Ajai, “Visionface recognition attendance monitoring
system for surveillance using deep learning technology
and computer vision,” in 2019 International Conference
on Vision Towards Emerging Trends in Communication
and Networking (ViTECoN), 2019, pp. 1–5.
[6] T. A. Kiran, N. D. K. Reddy, A. I. Ninan, P. Krishnan, D. J.
Aravindhar, and A. Geetha, “PCA based Facial
Recognition for Attendance System,” in 2020
International Conference on Smart Electronics and
Communication (ICOSEC), 2020, pp. 248–252.
[7] M. S. Akbar, P. Sarker, A. T. Mansoor, A. M.Al Ashray,and
J. Uddin, “Face recognition and RFID verifiedattendance
system,” in 2018 International Conference on
Computing, Electronics Communications Engineering
(iCCECE), 2018, pp. 168–172.
[8] Smitha, P. S. Hegde, Afshin, and Yenepoya Institute of
Technology, “Face Recognition based Attendance
Management System,” Int. J. Eng. Res. Technol.
(Ahmedabad), vol. V9, no. 05, 2020.
[9] S. Saypadith and S. Aramvith, “Real-time multiple face
recognition using deep learning on embedded GPU
system,” in 2018 Asia-Pacific Signal and Information
Processing Association Annual Summit and Conference
(APSIPA ASC), 2018, pp. 1318–1324.
[10] F. Deeba, H. Memon, F. Ali, A. Ahmed, and A. Ghaffar,
“LBPH-based enhancedreal-timeface recognition,”Int. J.
Adv. Comput. Sci. Appl., vol. 10, no. 5, 2019.
[11] https://www.electronicid.eu/en/blog/post/face-
recognition/en
[12] https://pypi.org/project/face-recognition/
[13] https://medium.com/@ageitgey/machine-learning-is-
fun-part-4-modern-face-recognition-with-deep-
learning-c3cffc121d78
[14] https://machinelearningmastery.com/one-shot-
learning-with-siamese-networks-contrastive-and-
triplet-loss-for-face-recognition/
[15] V. Kazemi and J. Sullivan, “One millisecond face
alignment with an ensemble of regression trees,” in
2014 IEEE Conference on Computer Vision and Pattern
Recognition, 2014, pp. 1867–1874.
[16] https://www.softwaresuggest.com/blog/manual-vs-
automated-attendance-system-pros-cons/
[17] https://www.edusys.co/blog/rfid-attendance-system
[18] S. K. Baharin, Z. Zulkifli, and S. B. Ahmad, “Student
absenteeism monitoring system using Bluetooth smart
location-based technique,” in 2020 International
Conference on Computational Intelligence (ICCI), 2020,
pp. 109–114.
[19] N. Dalal and B. Triggs, “Histograms oforientedgradients
for human detection,” in 2005 IEEE Computer Society
Conference on ComputerVisionandPatternRecognition
(CVPR’05), 2005, vol. 1, pp. 886–893 vol. 1.
[20] https://maelfabien.github.io/tutorials/face-
detection/1-theory-1
[21] https://heartbeat.comet.ml/one-shot-learning-part-1-2-
definitions-and-fundamental-techniques-1df944e5836a

Smart Attendance System using Face-Recognition

More Related Content

Similar to Smart Attendance System using Face-Recognition (20)

More from IRJET Journal (20)

Recently uploaded (20)

Smart Attendance System using Face-Recognition