ADVANCED FACE-BASED ATTENDANCE MARKER USING DEEP

LEARNING

Asst. Prof. Mamta*1, Animesh*2, Sanyam Singh*3, Deepak Bhora*4

HMR INSTITUTE OF TECHNOLOGY AND MANAGEMENT

Hamidpur, New Delhi - 110036

(Affiliated to Guru Gobind Singh Indraprastha University, New Delhi)

Abstract

Facial recognition technology has gained significant popularity in many areas, including attendance
management systems. This article provides an in-depth analysis of facial recognition-based engagement
systems, focusing on their principles, benefits, limitations, and implementation strategies. The use of facial
recognition technology in attendance management has received much attention in recent years due to its
accuracy, efficiency, and non-intrusiveness.

The Interactive Facial Recognition System uses advanced algorithms to recognize and authenticate people based
on specific facial features captured by the camera. These systems offer many advantages over traditional
methods such as fingerprint biometrics or card-based systems. First of all, they provide a seamless and
uninterrupted experience by eliminating the need for physical contact with equipment, which is very beneficial
in terms of hygiene and public health. In addition, the facial recognition system provides a high degree of
accuracy and is not susceptible to spoofing or spoofing, increasing the integrity of the information provided.

In terms of application, recognition-based participation systems are widely used in a variety of sectors, including
education, environment, healthcare, and government agencies. These systems simplify attendance management,
record-keeping, and resource efficiency. This is very useful for tracking attendance in educational institutions,
workplaces, or any level where tracking is required.

This proposed project aims to facilitate participant tracking through automatic facial recognition, notifying
administrators via text message, and providing real-time voice updates. Smart Response System uses the power
of Deep Neural Networks, specifically CNNs, and SMTP (Simple Mail Processing Protocol) to integrate email
services (e.g. Gmail) into your system and send alerts or reports.

Keywords: Face Recognition, Audience Management, Computer Vision, Haar-Cascade function, binary pattern
histogram.

Introduction

The traditional way of looking at a blackboard is a tedious task in many schools and universities. The fact that
students have to show their participation in the manual interview, which they can take from the whole class for 5
minutes, creates an additional burden for the department. This takes time. There is an option to join the proxy.
That's why many companies use Radio Frequency Identification (RFID), iris recognition, fingerprint
recognition, etc. Started using many other recording methods such as. However, since the system is web-based,
it can be time-consuming and intrusive.

Facial Recognition has created an important biometric function that is easily accessible and inaccessible. The
facial recognition system does not seem to be affected by different faces. The facial recognition system consists
of two stages: control and facial recognition. Face recognition is a 1:1 matching process that compares face
image to face image while comparing face recognition problem is a 1:N problem.

The purpose of this system is to create a presence system based on facial recognition technology. Here, the
individual's face is evaluated to be shown to the participants. Nowadays, facial recognition is becoming popular
and widely used. In this paper, we proposed a system that identifies students' faces in school video footage.
Participants are tagged when a recognized face is found in the database. The new system will take less time than
the traditional system.

This advanced system not only increases security by ensuring reliable decisions are made but also simplifies the
process of asking participants to track and record data. Overall, facial recognition using deep learning is at the
forefront of modern biometric solutions, providing organizations with a powerful and effective tool for
attendance management and monitoring.

Normally, participants are identified as students sitting in the classroom, and the teacher calls out the students'
names one by one to confirm their presence. Attendance is usually marked with a pencil and recorded in a
notebook. Traditional attendance systems have many disadvantages: they are difficult to record, time-
consuming, error-prone, and waste resources.

Any other device this is near us is the biometrics system. Biometric systems have 3 ranges: registration, garage,
and identification. The registration segment can have a completely unique individual. The identification is saved
in the form of an photo or code. The identification phase extracts the biometric facts and compares it with the
database to peer if it suits. Facial popularity has many blessings over biometrics as it requires less user
movement and permits more contributors to be mounted concurrently.

The main purpose of the thing is to implement an engagement program using facial popularity. Face reputation
entails picture capture the usage of a webcam, face recognition the usage of the Viola-Jones algorithm the usage
of its cascade classifier, pre-processing, saving the photograph to the database, extracting features thru LBPH
and CNN algorithms (Convolutional Neural Network community), and evaluating it with the input furnished.
Consumer and if it matches the scenario it is highlighted.

Therefore, the proposed gadget goals to robotically perceive participants using facial recognition. The teacher
can without problems song attendance.

Literature Survey

The main purpose is to find the solutions suggested by other authors and to take into account the shortcomings
of the systems they propose to offer better solutions.

Kawaguchi developed a system for class participation using a new system called nonstop monitoring, and
pupil participation was demonstrated by a camera taking a print of the pupil in the classroom. The armature of
the system is simple as both cameras are classroom walls. The first is the recording camera used to capture the
image of a pupil in the classroom, and the second is the detector camera used to find the pupil's seat in front of
the class and prisoner the image of the pupil. The system compares prints taken with a videotape camera to
faces in a database developed over time that actors can recoup.

Another paper proposed incorporating real-time computer vision algorithms into automatic guidance systems.
The system has a built-in camera with intrusion detection that can record video from inside. Faces were
compared from classrooms and face camera images in the system.

This project presents the analysis and design of an employee tracking system using facial recognition. The
proposed system is designed to manage login and check-in transactions as well as payment systems using facial
recognition technology. The system generates wage and attendance information based on employee ID.
Although articles provide analysis and profile of a particular company, details or comparisons with other
companies are missing.

They focused on the use of real-time facial recognition algorithms integrated with the management system of
traditional universities to identify and recognize students' faces during lectures. The system is intended to take
less time and disrupt the class less than traditional marking methods. Although this article shares the goal of
implementing a facial recognition system, specific techniques, and different libraries from our project.

This project presents a biometric management system using Raspberry Pi. It involves using Raspberry Pi and
related technologies to develop turnkey systems. Although the details of the project are scarce, it seems to focus
on creating an automatic attendance system using biometrics. Although both projects aim to improve
engagement systems, our project uses facial recognition technology, while this article focuses on the use of
Raspberry Pi with biometric data.

The auto-attendance system uses deep learning for facial recognition. It provides a system that combines
machine learning and deep learning algorithms to recognize and identify students' faces in real-time during
lectures. The system is intended to be less time-consuming than traditional grading methods and does not
require any corrections or supervision by teachers. System delivery is up to 97%. Our project shares similarities
with this article in using facial recognition to provide automated responses. However, our project is different in
terms of the algorithms and specific technologies used, such as the use of OpenCV, Haar cascades, and HTML
CSS for the International Journal of Research and Publishing, Volume 5, Issue. 1 second. 4040-4044 January
2024 4042.

This presents a demonstrate for an programmed participation framework utilizing facial acknowledgment. The
creators highlight the significance of labor costs in organizations and the require for a more proficient and viable
way of taking care of member prove. The proposed show employments facial acknowledgment innovation to
control reaction signals and make strides effectiveness. In spite of the fact that the record presents a show, it
does not talk about the subtle elements of its execution or compare it with other ventures.

This presents the plan and improvement of an e-presence controller employing a facial acknowledgment
framework based on Histogram Introduction Slopes (Hoard) and Machine Vector Machine (SVM). The
framework can filter numerous faces in a ordinary classroom environment and accomplish 95.65curacy. The
answers gotten from the framework are sent to the database for endorsement. This article highlights the
significance of variables such as brightness, remove, and directional resistance in framework plan. Generally,
this venture is comparable to our own in that it employments face-to-face acknowledgment for school
participation administration.

The real-time location of reaction instruments utilizing Haar cascades and Neighbourhood Double Designs
(LBP). The system captures faces employing a webcam, recognizes them utilizing the Haar arrange, and
extricates highlights utilizing LBP. Participation data is put away in a database for examination. In spite of the
fact that this venture offers the objective of utilizing facial acknowledgment for participatory administration, our
venture varies within the calculations and particular methods and subtle elements utilized (such as the utilize of
servers and SQLite databases).

Proposed System

1. Overview
The brilliant "Open Source Computer Vision Library," OpenCV, was used for this extend. We are going be
utilizing Python and Raspberry Pi (with Raspbian as the working system) for this lesson, but I tried the code on
my Mac as well, and it capacities just fine there as well.
"Many code alterations ought to be made for it to operate on a Mac. Do not stress, I'll comment. OpenCV was
made with a significant accentuation on real-time applications and handling execution. Subsequently, it's perfect
for camera-based real-time confront acknowledgment.

The Three Stages

To create a wrapped up Confront Acknowledgment extend, we got to center on three diverse stages:
 Face Location and Information Gathering
 Train the Recognizer
 Recognition of Faces

2. Installing OpenCV 3 Bundle

Make beyond any doubt you have the prerequisites, such as Python and a appropriate improvement
environment, sometime recently introducing the OpenCV 3 bundle. Another, you have got the choice to utilize
Python's bundle chief pip to introduce OpenCV 3. To introduce the specific form of OpenCV 3.4.14, do "pip
install OpenCV - python = 3.4.14". As an elective, you'll build OpenCV from the source, giving you more
control over establishment inclinations. For comprehensive instructions on building from source depending on
your operating system and prerequisites, see to the official OpenCV manual. After establishment, you'll utilize
the OpenCV library for a assortment of computer vision assignments by importing it into your Python ventures.

3. Examining Your Focal Point

After installing OpenCV on your Raspberry Pi, let's confirm that your camera is working accurately with a test.
I'm getting to expect that your Raspberry Pi as of now includes a PiCam introduced and turned on. The code will
record the video stream that your PiCam produces and play it back in both grayscale and BGR color. Take note
that since of the way my camera is put together, I turned it vertically. Comment out or expel the "flip" command
line on the off chance that that's not your circumstance. You ought to hit the console key [ESC] to exit the
program. To enact the video window, press it together with your mouse and at that point hit [ESC].

Pre-Processing: Amid this step, the image is grayscale. Taking after discovery, the photographs are included to

figure 1 Simple CamTest

the database.

The image above shows the result.

4. Recognition of Faces
Relating faces is, of course, the foremost essential errand in face recognition.
To identify a defy, you must first" prisoner" it( Stage 1) and after that compare it to a unused defy that will be
captured subsequently( Stage 3). The foremost popular system for relating a defy- or any kick- is the" Haar
Cascade classifier."
In their 2001 distribution" Quick Protest Detection employing a Boosted Cascade of Basic Highlights," Paul
Viola and Michael Jones presented an effective question distinguishing evidence strategy kick position
exercising Haar point- grounded waterfall classifiers. This strategy is grounded on machine literacy, and it
includes preparing a waterfall work with a huge number of both positive and negative images.
Next, it's connected to identify effects in further filmland. We will be exercising facial position in this
circumstance. To prepare the classifier, the strategy to begin with requires an extensive number of positive
filmland, or filmland with faces, and negative filmland, or filmland without faces. After that, we must liberate
its features. The good news is that OpenCV incorporates both a locator and a trainer. You will use OpenCV to
construct your classifier and prepare it for any object, similar as buses , aeroplanes, etc. You will discover all
the data nearly it then Cascade Classifier Preparing.
The result:

figure 2 face Detection

Classifiers for "eyes discovery" and indeed "grin place" may be protected as well. In sure situations, you'll
incorporate the classifier characteristic and draw a rectangle within the face loop, as it might be impossible to
become aware of an eye or a smile on its very own.

5. Data Gathering
It said that, let's get our project's initial phase underway. Starting with the last phase (facial detection), we will
simply establish a dataset in which we keep a set of grayscale images for each ID, together with the section of
the photos that were utilized for face detection.
Keep in mind that you need to import the "os" library to save the aforementioned file. The names of each file
will adhere to this format:

User.face_id.count.jpg

For instance, the fourth sample image in the dataset/directory for a person with face_id = 1 would look
something like:

User.1.4.jpg

as seen in the picture taken with my Pi:

figure 3 User Dataset

I'm collecting 30 samples from each ID using my code. On the final "elif," you can modify it. The loop where
the face samples are taken is broken using the number of samples.

figure 4 Train the Recognizer

Execute the Python script and obtain several IDs. Every time you want to add a new user (or update the images
for an existing member), you have to run the script.

6. The trainer
We need to "teach" the OpenCV Recognizer using all of the user statistics from our dataset in this 2nd section.
An OpenCV characteristic specifically handles this. A.yml file with the end result may be stored within the
"instructor/" directory. So, let's start constructing a subdirectory in which we will shop the trained information:

mkdir instructor

verify that you have the PIL library hooked up in your Raspberry Pi. If now not, do the following command in
Terminal:

pip deploy pillow

As a recognizer, we will make use of the LBPH (Local Binary Pattern Histogram) Face Recognizer, that is
supplied in the OpenCV bundle. We accomplish this inside the following line:

recognizer=cv2.face.LBPHFaceRecognizer_create()

The function "getImagesAndLabels(route)" will retrieve all snap shots from the directory "dataset/" and go back
arrays: "Ids" and "faces". With those arrays as input, we will "train our recognizer":

recognizer.educate(faces, IDs)

As an end result, we are able to keep a record titled "instructor.yml" within the trainer listing we in advance
generated. This is, it! I covered the closing print assertion wherein I showed the quantity of person faces we
have educated. On every occasion you do segment 1, you must also run phase 2.

7. Recognizer
We've now completed the remaining level of our challenge. here, we are able to take a photograph of a brand
new face the use of our digicam. If this character has already had his face taken and educated, our recognizer
will make a “prediction” and provide its identity and an index, indicating the recognizer's level of confidence on
this match.
Due to the fact we are including a new array on this area, names may be displayed as opposed to numbered IDs:

names = ['None', 'Animesh', 'Deepak', 'Sanyam', 'Z', 'W']

Animesh, for instance, will be the user with id = 1, Deepak, id = 2, etc.

The next step may be to apply the haasCascade classifier to detect faces, just as we did formerly. As soon as a
face has been recognized, we may additionally call the most essential feature:

id, confidence = recognizer.predict(grey portion of the face)

A captured section of the face to be studied can be handed as enter to the recognizer.expect() feature, in order to
go back the likely owner alongside the identity and the recognizer's stage of confidence for this suit. Remember
the fact that if its miles deemed a great fit, the self-belief index will go back "0."
In the end, we overlay a textual content with the possibly identity and the proportion of the suit that is accurate
("chance" = 100 - self-belief index) if the recognizer become able to discover a face. If now not, the face is
classified as "unknown."

Final result:

figure 5 face Recognition Result

8. Attendance and matching

Each face image in the training dataset is compared, one by one, with each image from the input set. Student
facial photos were compared to those in the training dataset; if the two sets of photos match, the student is
recorded as present in class, and those that don't match are recorded as absent.
After identifying faces, the system will record the identified faces as present in a database and the unidentified
faces as absent. A list of absent students will also be created and emailed to the relevant faculty members. An
updated monthly attendance sheet will be sent to the faculty after each month.

figure 6 Mark Attendance

Future Scope
 ATM fraud can be avoided. As it continues to be improved, facial recognition technology will become
more reliable in the future. Only consumers will be granted admission.
 It can also be used to set oneself apart from the competition during tests like the MBBS, SSC, IIT, and
Civil Services Exam.
 There are reports of double voters everywhere in the world. To prevent this, voting will only be
permitted for those with valid identification.
 Globally, government and corporate organizations will make extensive use of this architecture.
Result
The initial stage of the face recognition model entails taking a webcam picture. The Haar cascade classifier is
used in the second step of the Viola-Jones method for face detection. Pre-processing, which takes place in the
third stage, mostly entails turning the colored image into a grayscale. By running the image through the median
and Gaussian filters, this is accomplished. The image is stored in the database in the fourth stage. Next, using
LBPH convolutional neural networks for feature extraction (CNN). Following that, it will be compared to the
user's input; if they match, attendance will be noted in the Data file; if not, it won't be noted.

Table 1. CSV Data file of Attendance

S.No. Enrollment number Name of students
1 00296502720 Animesh
2 01996502720 Sanyam
3 20196502720 Deepak
4 XXX96502720 Z
5 XXX96502720 W

Conclusion
The suggested solution uses face recognition technology to generate an effective system for tracking student
attendance. The proposed technology will be able to use facial recognition to record attendance. It will use a
webcam to recognize faces and then identify the users. It will document the recognized student's attendance and
update the attendance records on the outcome of the successful recognition process. For schools and colleges to
record the attendance of instructors, staff members, and students, this technique is quite helpful.

When compared to previous systems that use eigenfaces or fisher faces, this system performs far better since it
makes use of the most effective algorithms available: the LBPH algorithm, the Convolutional Neural Network
(CNN) for feature extraction, and the Viola-Jones method for face detection.

The LBPH algorithm yields an accuracy of 85%. With CNN, an accuracy of 95% can be achieved. Additionally,
this system has confidence built in, so it only records attendance when it is at least 70%; otherwise, it does not
record attendance. Additionally, this technology can identify numerous faces, allowing attendance to be taken
simultaneously by the entire class. A student must be physically present for their attendance to be recorded; an
image from a cell phone or a photocopy will not be accepted. Proxy attendance via mobile phone photos can be
avoided in this way.

Therefore, any organization or place of business needs to install a facial recognition attendance system because
fingerprint systems are unable to record several attendances at once and may cause long lines of people.

References

[1] Poornima, Sripriya, Vijayalakshmi, Vishnupriya Attendance Monitoring System using Facial Recognition
with Audio Output and Gender Classification, IEEE International Conference on Computer,
Communication, and Signal Processing (ISP-2017) https://ieeexplore.ieee.org/document/7944103
[2] Shivam Singh, Prof. S. Grace line Jasmine Face Recognition System, International Journal of Engineering
Research & Technology (IJERT)ISSN: 2278-0181 Vol. 8 Issue 05, May 2019
[3] Kennedy Ok pujie, Etinosa Noma-Osaghae, Samuel John, Kalu-Anyah Grace, Imhade Okokpujie “ A face
Recognition Attendance system with GSM Notification” in IEEE NIGERICON 2017.
https://ieeexplore.ieee.org/document/8281895
[4] https://www.ijeat.org/wpcontent/uploads/papers/v8i3S/C
[5] Shreya K. Sawhney, Karan Kakkar, Samyak Jain, Shailendra Narayan, Rakesh Garg “Real Time Smart
Attendance system using face recognition techniques “ in the international conference on cloud computing
data science and engineering 2019. https://ieeexplore.ieee.org/abstract/document/8776934
[6] Kritika Shrivastava, Shweta Manda, P.S Chavan, “Conceptual model for proficient automated attendance
system based on face recognition and gender classification using Haar- cascade” in IJEAT 2018.
https://www.ripublication.com/ijaer18/ijaerv13n10_111.pdf
[7] Aruna Katara, Sudesh V, Amar P, Nikhil D, Bhele
https://www.scribd.com/document/350875725/Attendance-System-Using-Face-Recognition-and-Class-
Monitoring-System
[8] Omar Abdul Rehman Salim, Rashida Olanrewaju, Wasiu Balogun “Class attendance management system
using face recognition” in ICE 2018. https://www.researchgate.net/publication/329067820_Class