AUTOMATION OF ATTENDANCE USING DEEP LEARNING

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 11 | Nov 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 572
AUTOMATION OF ATTENDANCE USING DEEP LEARNING
Chiluka Harshith Reddy1, Lavu Nithish Chandra2, Ranabothu Sri Anand Reddy3
1,2,3Student, SENSE, VIT Vellore, Tamil Nadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Student Attendance mainframe structure is
defined for managing the student's class attendance data files
using the concept of face detection and recognition through
open computer vision.This approach is proposed primarily to
improve the existing university attendance practices and
prevent the waste of resources and time. The concept of
moving from the traditional attendance system to the digital
one using face detection and recognition techniques has been
driven by the automation world's pointing sides.Byaddingthe
dataset of an individual, the Student Attendance structure is
constructed in this way. In addition to lowering the long-term
time burden work, and disposables required, the main goal of
constructing this system was to increase the adaptability and
performance of the attendance system procedure.TheStudent
Attendance markup structure's primary functionistoaddand
modify a student's attendance notes, make an automatic
computation of the number of presentees and absentees
depending on the subject and affability of the class, and then
produce an automated document or spreadsheet.The concept
of open computer vision is used in this approach, which is
entirely based on the general-purpose language Python.For
face detection system we used haarcascade and for face
recognition, we used LBPH model;After training each
individual student, the system generated a spreadsheet that
included the number of students present in the classroom
along with a picture or video that was captured live.
Key Words: Principal Component Analysis, SVM, Dlib, LBP
Feature, API, Tensorflow
1. INTRODUCTION
Maintaining attendance is crucial at any institute for
monitoring the quality of education. Students' attendance is
routinely recorded by establishing attendance files or notes
provided by the departmental arch in the depths of the
institutions. The teacher manually takes attendance by
calling out each student's name and confirming whether or
not they are present in the class. This process is dull, time-
consuming, and unreliable because students frequently
make the wrong calls for their absent friends. Additionally,
this procedure makes it more difficult to alter every
student's attendance in a large classroom. In order to
automatically identify the students in a class and record
their attendance by collecting their frames, we designedthis
application and used a range of techniques, including facial
exposure and an understanding system. While some biotech
assimilation metrics can be increased properly, in the past,
students typically had to wait longer when they entered the
room for attendance. Face recognition is the best option
because of its non-intrusiveness and familiarity, as people
generally know other people by their facial features. This
facial biometric structure primarily consists of an enrolled
approach in which, following the process of detecting and
understanding, the key distinguishingcharacteristicsofeach
individual face will be saved in the dataset. The traditional
methods for analyzing student engagement in particular
subjects involve physically signing the attendance logs in a
PC framework for analysis. This approach is ineffective
because students would sign up for their absent classmates,
which is boring, unpleasant, and prone to errors. The use of
the face identification and acknowledgment framework in
place of the conventional methodswill providea quickerand
more effective method for accurately capturing student
participation while also providing secure, reliable, and
strong restrictions on the framework records. After these
restrictions have been approved, one can access the records
for any purpose, including for organization,guardianship, or
even for the student's own studies.
2. LITERATURE REVIEW
2.1 Face Recognition: From Traditional to Deep
Learning Methods
Early face recognition research centered on techniques that
matched basic features using image processing algorithms
de-tracing the faces' geometrical shapes.Nevertheless,these
techniques only functioned in incredibly restricted
circumstances,theydemonstrated.Computersarecapableof
recognizing faces automatically. Then, statistical sub spaces
techniques like principal component linear discriminant
analysis with principal component analysis (LDA) grew in
acceptance. These techniques are known as holistic because
they incorporate data from the full-face region. During this
time, advancements in other computervisionfieldsledtothe
creation of capable local feature extractorsthatelucidate the
texture of an image in several areas. Feature-basedmethods
for facial recognition involve comparing these regional
specifics over pictures of faces. Hybrid methods were
created by further developing and combining holistic and
feature-based approaches. Facerecognitionsoftwareusinga
combination of techniques Up until recently,state-of-the-art
technology was deep learning the most effective strategyfor
most computer vision applications, such as facial
identification. The remainder of this text summarizes some
of the most notable reviews Each of the aforementioned
kinds of strategies is supported by a search.

2.2 Class Attendance System Using Viola-Jones
Algorithm and Principal Component Analysis
The outcome in Viola and Jones depends on the data and
unreliable classifiers. The uniformity of thetrainingsethasa
significant impact on the final detection's quality. Important
considerations include the size of the sets and the inter class
variability. When numerous peoplewithdifferentsequences
are considered, the analysis yields very poor results. Viola
and Jones algorithm is the employed algorithm inthispaper.
The data's training process should be carriedoutproperlyin
such a way that the quality of the final detection increase.
The system overview ought to include the overall building
design that will the concise and thorough details about the
project.
2.3 Automatic attendance system using Deep
Learning
The system is put into practice under the basic and
fundamental tenet of a digital camera in the classroom. In a
lecture that lasted 50 minutes, the digital camera wouldtake
2 pictures every 25 minutes. The system will now get the
image and extract all of the faces from it. Now, the existence
of the face would be determined by comparison with the
trained model of faces already in use. If a student's face is in
the current database, the system will save their unique ID in
the attendance database or discards them if the studentisn't
in the database. Student database We havetackleda number
of issues in this study, including real-time face detection,
multiple face detection, and integration with the computer
learning algorithm. The actual challenge in putting an idea
into practice was real-time face extraction from an image. In
order to resolve this problem, we employed using the Deep
Neural Network (DNN) of the Tensorflow estimator API,
which is also trained from the instantaneously extracted
photos. However, finding face-like patterns is only a small
portion of the issue. You must use face recognition
technology using an algorithm to successfully identify a
student from a database of pupils. We usedGoogle'sface net,
a model that has been pre-trained on 150 000 photos and
was inspired by the Google Pixel, to deal with this problem.
2.4 Face Recognition Using Neural Network
The computation is slow and the detecting method is
complicated. Performance compared to the Viola-Jones
algorithm is typically worse. The algorithm is neural
network-based. This strategy is onlyeffectiveifthelarge size
of the image was taught.
3. IMPLEMENTATION
3.1 Existing System
In the last few years, face recognition technology has
embraced a wide range of methodologies, but the classical
methodology still predominates. Component analysis,
discriminating analysis, discrete transformation, and
component analysis are categories for prestigious face
recognition. It is considered to be the most important
element in facial recognition technology. Numerous
researchers in the field of facial recognition technologies
employ the eigenfaces technique. The main element of this
technique is eigenfaces. Basically,itdivideda varietyofinput
variables into several classes (Li & Hua, 2015). The PCA
(Principal Component Analysis) algorithm can be used to
extract the image data in its original form. One of the
fundamental principles that PCA adheres to is that it can
recreate the image's original form from the original
collection by using eigenfaces. In face recognition
technology, Eigenfaces are regarded as the key component.
Eigenfaces typically depict the primaryfacial characteristics,
which the original image may not have had.
3.2 Proposed System
An attempt is made to develop the automated facial
attendance system utilizing SVM on the LBP feature taking
into account the drawbacks of some of the systems listed in
the previous works, as the LBP method provides good
accuracy in comparison to other systems. The suggested
method introduces an automated attendance system that
incorporates a facial recognition algorithm and an Android
app. Any device with a camera is capable of taking a picture
or a video, which it may then upload usinga webapplication.
The received file is put through face detection and
recognition processes, so the detected faces are extracted
from the image.
3.3 Gap Identified
A few drawbacks of facial recognition areimagequality,size,
angle of facing, and processing time. performance of the
facial recognition algorithmisfirstfundamentallyinfluenced
by the image quality. When comparedtoa digital camera, the
video scanning of an image has inferior quality. The method
of facial detection as a whole was impacted by the image
quality. Face recognition poses considerable challenges in
terms of storage and processing. To recognize a person's
true appearance, a certain angle is chosen ( Minaee & Wang,
2015). The process of detecting faces will be severely
hampered by the several angles that must be used in order
to obtain an adequate face while employing recognition
software. In essence, they adopted the 2D facial type'sphoto
structure. This format prevents facial recognition software
from currently detecting numerous faces. The facial
recognition technology will experience issuesbecauseofthe
person's movements, which resulted in erroneous
photographs being captured. It is necessary to use current
software, which is highly expensive on the market, for more
accuracy. The detection method can occasionally run into
trouble when the photos are blurry. The effectiveness of
facial recognition technology is also influenced by the
camera's perspective.

3.4 Principal Component Analysis (PCA)
It is derived from the transition of Karhunen-Loeve.
Principal Component Analysis (PCA) often locates a t-
dimensional subspacewhosebasisvectorscorrespondto the
largest variance directionin theoriginal imagespacegivinga
dimensional vector representation of each face in a training
set of photos. Typically, this new subspace has a lower
dimension (ts). The PCA basis vectors are defined as
eigenvectors of the scatter matrix if the image elements are
thought of as random variables. For dimensionality
reduction, the Eigenface technique use PCA to identify the
vectors that best capture the distribution of face pictures
throughout the whole image space. The subspace of face
images is defined by these vectors, and it is known as face
space. To determine a set of weights that accurately
represents the contribution of each vector in the face space,
all of the faces in the training set are projected
onto the face space. To generate the appropriate set of
weights for identifying a test picture, the test image must be
projected into the face space. The face in the test image can
be recognized by comparing the weights of the test image
with the set of weights of the faces in the training set. The
foundation of PCA's primary methodistheKarhumen-Loeve
transformation. The image might be viewed as a sample of a
stochastic process if the image's constituent parts are
assumed to represent random variables.Theeigenvectorsof
the scatter matrix ST, ST=Σ N i=1 (xi - μ)(xi - μ)T, are what
are known as the PCA basis vectors.
4. SYSTEM OVERVIEW
The ultimate objective of a face recognition system is image
understanding, or the ability to recognize an image's
meaning in addition to its structure. A general definition of
automatic face recognition is as follows:givenstill ormoving
photographs of a scene, identify or confirm one or more
people in the scene using a database of faces that have as
been previously recorded. The challenge can be solved by
segmenting faces (facial detection) from cluttered scenes,
extracting features from the face regions, and then
identifying or verifying the faces. The inputforidentification
is an unidentified face, and the system returns the identity it
has deduced from a database ofwell-knownpeople, whereas
in issues with verification: the system must accept or reject
the claim of the input's identity.
4.1 Data Set Creation
1. Obtain the class's student roster.
2. Look through each image folder associated with each
student.
3. The parent folder name appears as the class label for each
image.
4. Determine whether a face is there by applying a face
detection algorithm to the face.
5. Get the bounding box coordinates if a face is found.
6. Crop the image so that it only contains a face using the
bounding box coordinates.
7. Use Dlib to determine the direction of the face, then apply
transforms to align the eyes, mouth, andotherfacial features
at a specific angle. This guarantees the neural network's
input is of high quality.
8. Feed the neural network with the aligned image. The
network produces 128 measurements of each face.
9. At this point, the image's data have been extracted.Traina
classifier using this embedding vector.
10. The classifier is finished off and saved asanobject,which
can later be imported from the database and used to
make predictions.
4.2 Applying the LBP operation
The initial computational stage of the LBPH is to produce an
intermediate image that, by emphasizing thefacefeatures of
the original image, more accurately describes the original
image. The method does this by utilizing a sliding window
idea depending on radius and neighbour. The procedure is
shown in the image.

5. EXPERIMENTAL RESULTS
6. CONCLUSION
This method has been suggested to keep up attendance. The
paperwork and stationery are replaced with an automated
system that is quick, effective, cost-effective, and time-
saving. The proposed system, however, is anticipated to
produce the intended outcomes. Integrating other effective
methods could also increase efficiency. Here, we've covered
a variety of face identification techniquesthattheresearcher
utilized. These techniques might be applied by educational
or business institutions to track students' attendance at
lectures by identifying their faces. In order to classify faces,
we are attempting to develop a system using Improved
Support Vector Machines (IVSM) on LBP features in the
following phase.
FUTURE DEVELOPMENT
In the future, we'll connect the system with the email
address and mobile number so that if anybody doesn'tshow
up, an automatic SMS or MAIL will be sent to them.
REFERENCES
[1] P. Mehta, “An Efficient Attendance Management System
based on Face Recognition using Matlab and Raspberry
Pi 2,” Int. J. Eng. Technol. Sci. Res. IJETSR, vol. 3,no.5,pp.
71–78, 2016..
[2] Yang Zhong Josephine Sullivan Haibo L KTH Royal
Institute of Technology, 100 44 Stockholm, Sweden
"Face Attribute Prediction
[3] Max Ehrlich, Timothy J. Shields, Timur Almaev, and
Mohamed R. Amer, "Facial Attributes Classification
using Multi-Task Representation Learning",2016 IEEE
Conference on ComputerVisionandPatternRecognition
Workshops.
[4] Yun-Fu Liu, Member, IEEE, Jing-Ming Guo, Senior
Member, IEEE, Po-Hsien Liu, Jiann-Der Lee, and Chen-
Chieh Yao, "Panoramic Face Recognition", IEEE,
Transactions on Circuits and Systems for Video
Technology.
[5] Hamdi Dibeklio˘glu, Member, IEEE, Fares Alnajar,
Student Member, IEEE, Albert Ali Salah, Member, IEEE,
and Theo Gevers, Member, IEEE, "Combining Facial
Dynamics With Appearance for Age Estimation", IEEE
TRANSACTIONSON IMAGEPROCESSING,VOL.24,NO.6,
JUNE 2015.
[6] W. Haider, H. Bashir, A. Sharif, I. Sharif, and A. Wahab,“A
Survey on Face Detection and Recognition Techniques,”
Res. J. Recent Sci., 2014.
[7] I. Dagher, “Incremental PCA-LDA algorithm,” in CIMSA
2010 - IEEE International Conference onComputational

Intelligence for MeasurementSystemsandApplications,
Proceedings, 2010.
[8] N. Kar, M. K. Debbarma, A. Saha, and D. R. Pal, “Study of
ImplementingAutomatedAttendanceSystemUsingFace
Recognition Technique,” Int. J. Comput. Commun. Eng.,
vol. 1, no. 2, pp. 100–103, 2012.

AUTOMATION OF ATTENDANCE USING DEEP LEARNING

More Related Content

Similar to AUTOMATION OF ATTENDANCE USING DEEP LEARNING (20)

More from IRJET Journal (20)

Recently uploaded (20)

AUTOMATION OF ATTENDANCE USING DEEP LEARNING