Problem Statement — To design and implement attendance management system using deep learning to detect and recognize valid QR codes and facial features for a robust and a secure attendance.
Teachers taking attendance during a lecture in a class in the university is a time-consuming task. Traditionally a teacher calls out the student’s name or roll numbers in a class to mark their attendance. This roll call strategy has 2 main flaws. Firstly, it usually takes 5-10 valuable minutes for attendance marking, which could be used by the teacher in a productive manner. Secondly, there are many unauthorized attendance registrations that take place during this traditional approach as the teacher cannot focus his attention to measure the authenticity of each registration. This paper proposes an automated system for marking and handling of attendance by making use of the 2-factor authentication technique that can register a student’s attendance and can successfully prove the physical presence of that student in the class. The solution involves scanning of a QR code that is first displayed by the professor during the lecture and then face recognition performed on the handheld mobile device of the student himself. This process of using QR code and the face recognition provides a bridge to link the physical world (the classroom) to the physical presence of the student in the classroom via an intermediary device (student’s hand-held smartphone)
Every university or an institute has some criteria regarding a student’s attendance in a class. In order to get an accurate measure of the attendance and to minimize the time required for the lecturer to take the attendance, we need a robust method to automate this task. In the traditional method, the attendance is usually maintained on a piece of paper by the professor who calls out for a students’ roll number or name to mark his presence. This technique, although almost always provides a 90% accurate attendance on an average, is very time-consuming for the lecturer.
So we clearly see 2 problems with the approach:
What about the 10% inaccuracy – which is the unauthorized attendance registration
It is a time-consuming process
And one latent problem that is most often forgotten when giving a technological solution is:
The cost of the device used for the technological solution
Many solutions have been given to overcome both of these issues. But none of them is able to tackle both of these problems simultaneously in an almost perfect manner.
Let’s take for instance biometric attendance monitoring using fingerprint scanners. This clearly solves the 10% inaccuracy problem since the fingerprints can never be replicated for an unauthorized attendance registration. But this solution fails to address the time taken to measure the attendance. Imagine how much time it will take each student having to register his fingerprint in the beginning or at the end of each lecture. Even if the fingerprint recognition device is passed on from person to person in the class, imagine how much of a distraction it will be for students as well as the teacher.
Consider another example where the attendance is taken using facial recognition using a real-time video camera placed in each class. This system is very robust and solves both the problems, there is no inaccuracy as live feed is monitored and it does not waste even a single second of the lecture as this camera works in real time in the background. But this solution is not a viable solution because the cost of equipment including the camera (which has to be of a very good quality as it has to detect faces over a long distance) and the cost of GPU required to recognize live faces from a video stream (which is a computationally time-consuming process)
All these solutions have not yet been able to understand and take advantage of the wide-spread use of technology among students. According to  in 2014, 94.4% college students owned a smartphone. Intuitively, we can see that the number of students using smartphones now in 2018 will definitely be extremely close to 100%.
How can this be used for the benefit of attendance management?
Why don’t we let the students take the attendance instead of the professor?
But this clearly begs a question, how can we trust students? Won’t they just mark their attendance even if they are not present?
Most people would stop thinking about this right there. But this paper provides a solution that enables the student to mark their own attendance by linking their physical presence in the class to the physical world via the smartphone.
The proposed solution works in 2 steps:
Linking the physical world to a student’s smartphone: It makes use of a QR code that is displayed by the teacher’s application on either his/her laptop or on the projector. This QR code signifies the physical world and is scanned by the students using their smartphones. This process links the physical world to the smartphone.
Linking the smartphone to the student’s presence inside the classroom: This is done by facial recognition of students through the same smartphone application.
Isn’t it simple and intuitive?
PHYSICAL WORLD first linked to SMARTPHONE which is in turn linked to the STUDENT himself.
Let’s discuss how it solves all the 3 problems mentioned earlier:
Does it solve the unauthorized attendance registration issue?
Yes, it definitely does make it much better! The linking of the physical world to the smartphone to the student is a robust 2-factor authentication system that is very difficult to fool if the QR code is properly encoded and the facial recognition system is strong enough as to not detect false faces.
Does it take attendance quickly?
Of course, it does. It hardly takes a second to scan a QR code and another second to recognize the face on a smartphone so powerful these days.
How cost effective is it?
This solution can be implemented by simply using a students smartphone that only needs 2 cameras (Front-facing or selfie camera and Rear-facing camera)
Since it does not require any other hardware and more than 95% of students own such smartphones, the cost can be practically assumed to be 0 which is just brilliant!
The following sections provide in-depth methodologies and techniques used to achieve this.
Educational institutions always make it mandatory for any student to have a minimum attendance of 75%. Monitoring the attendance has always been one of the important aspects of any lecture. The basic aim of this criterion is beneficial for the students because as said almost 30-40% of the overall learning is done in the classroom itself.
As discussed in the introduction, different attendance monitoring systems are currently being deployed. But each system has some cost associated with it. Our institution uses the traditional ‘Roll-Call’ method. The Lecturer in every lecture has to perform the ‘Roll-Call’ in order to record the attendance. The Lecturer already has to speak continuously throughout the lecture. This additional ‘Roll-Call’ obviously uses their energy. Looking from their point of view, this is definitely a tiresome, tedious process. Proxy is one of the most important concerns. In a class of 60-80 students finding out the proxies proves to be an additional overhead for the Lecturer. She/he might have to count the number of students in the lecture and the number of recorder attendances and cross check each and every record. Those students who attend the Lecture regularly to complete the criterion of 75% attendance and Proxies are clearly doing them an injustice.
“To contribute to the community, to solve real-life problems” is what defines Engineering. Soon to be Engineers we were aware of this real-life problem. To help lecturers as well as to help students improve their academics was the driving force which led us to provide the following solution of “2-factor attendance monitoring using QR code and facial recognition”.
III. LITERATURE SURVEY
The various attendance systems that were analyzed before coming up with the proposed solutions are:
Mobile Attendance using Near Field Communication and One-Time Password
Biometric Smart Attendance Kit with Fingerprint Scanner by Using Microcontroller
Design and Implementation of Iris Recognition Based Attendance Management System
FaceTime – Deep Learning Based Face Recognition Attendance System
Summaries of each research paper:
Paper  proposes a system that makes use of the Near Field Communication (NFC) tag that is inbuilt in the handheld smartphone owned by students and also the use of One Time Password (OTP). According to the paper, the student taps his NFC enabled phone to the teacher’s phone, in-turn receiving an OTP which needs to be verified through the use of the college’s Wi-Fi network. This system is robust in itself but it makes some assumptions. One of which is that every phone is NFC enabled and the other being that the college Wi-Fi is readily available to be used by a large number of students at the same time. And one disadvantage that is mentioned by the paper itself is that the OTPs might take time to deliver since “SMS traffic is not sent point to point; it is queued and then sent on to the required network cell where it is again queued and finally sent to the end user’s phone”. Another visible problem of using such system is that sending SMS over the network is always a priced service just like it is mentioned in the paper “Irrespective of delay to delivery or non-usage of the SMS OTP, each and every SMS OTP will be charged”
Paper  gives us a method to take attendance that is fail-proof in most of the cases. It proposes to make use of “intelligent system based on fingerprint scanner”. This kind of a system makes use of a hardware, that is used to scan and recognize the fingerprints of the students in a classroom, which is placed outside the classroom. The paper says that, at first, the course code for which the attendance has to be taken is fed to the hardware. This hardware itself is then used to scan a student’s fingerprint. The paper says that this system is “portable, handy, cheap and reliable”. It is easy to understand that it really is portable, handy and reliable, but is it really cheap? The paper mentions the use of a fingerprint scanner module, an RTC module, ATmega2560 (processing unit), and a digital display. It is clearly visible that this solution is REALLY not cost-friendly because of the use of so many components. Another problem with the solution is that the students have to stand-up from the seat, make a queue and give their fingerprint attendance after each lecture. Is this really viable?
Similarly, paper  proposes to make use of an “iris scanner” instead of the “fingerprint scanner”. According to the paper “Iris recognition is regarded as one of the most reliable, accurate and efficient biometric identification systems due to the inner characteristics of iris, such as uniqueness, immovability, and time invariance”. Unlike the previous fingerprint scanner, this system does not really require any additional hardware, but it makes use of the webcam that is either inbuilt in the laptop or the computer that is present in the university. This method, although making use of some brilliant algorithms to detect and recognize iris, is very assuming. It forgets the fact that not all students have computers in their lectures. Consider the students studying art, we cannot expect them to sit in front of their computers and laptops during an art lecture just for marking their attendance. Plus, even though Iris is a very reliable source of verification, it requires an even more reliable algorithm to recognize iris which is a very complicated process.
Paper  proposes a 4 step pipeline for the face detection and recognition process before the attendance itself is marked. The first step in the pipeline is face detection which is done using a CNN cascade consisting of 3 CNN’s for binary classification (face and non-face) and 3 CNN’s for proper adjustment of the bounding box for the face. The objective of the second step is to detect and extract the facial landmarks from the detected face even when the face is in different positions. The third step is the generation of 128-byte embeddings per face and using a pre-trained FaceNet network to train our network which has triplets consisting of the face image of the target person, the test face image of the same person and face image of another person. The last step of the pipeline presents applying an SVM to develop the facial recognition model using the previously generated embedding from the dataset for tracking the attendance. Even though the system is very robust, it requires the creation and generation of a large dataset which itself is a time-consuming process which only goes on increasing as the number of faces increase.
IV. GAP ANALYSIS
V. ISSUES / CHALLENGES
Mobile Attendance using Near Field Communication and One-Time Password:
OTPs might take time to deliver since “SMS traffic is not sent point to point; it is queued and then sent on to the required network cell where it is again queued and finally sent to the end user’s phone” 
Sending SMS is a costly service, “Irrespective of delay to delivery or non-usage of the SMS OTP, each and every SMS OTP will be charged”. 
NFC is a radio-frequency based component. Hence there is a potential security risk of eavesdropping.
Biometric Smart Attendance Kit with Fingerprint Scanner by Using Microcontroller:
Fingerprint scanning for every individual during every single lecture is definitely a time-consuming task.
It requires the use of a fingerprint scanner module, an RTC module, ATmega2560 (processing unit), and a digital display, hence it definitely does not count as a cheap solution.
Design and Implementation of Iris Recognition Based Attendance Management System:
This entire system is very assuming in itself. The most basic and false assumption it makes is that each student owns a laptop or there is one computer for every student in the class.
In the case of an art student, he/she does not require to do any computational work and hence is almost never around a computer. Attendance management using this proposed system will not be feasible.
The algorithm itself takes time to design and maintain because the Iris is a small and difficult thing to track.
FaceTime – Deep Learning Based Face Recognition Attendance System:
Even though the system is very robust, it requires the creation and generation of a large dataset which itself is a time-consuming process which only goes on increasing as the number of faces increase.
Constantly requires internet connection due to the use of a face API.
Unpredictable recognition due to different lighting conditions was observed.
VI. PROPOSED WORK
The proposed system consists of 3 main components, a backend server to collect and store the attendance data for each student, a client-side application on the teacher’s computer to generate and display the QR codes and a client-side mobile application on the student’s phone to scan the QR code and the student’s face for verification and marking of attendance.
The backend server acts as the central repository for all the student, teacher and attendance related data like the student and teacher profiles, the information related to the lectures associated with the students and teachers, the attendance data itself, and deep learning models for facial recognition of the students.
The client-side application on the teacher’s computer will have the sole responsibility of generating and displaying the QR codes for the students to scan to mark their attendance in each lecture. The QR code will be generated using two main components, one will be the parameters pertaining to the lecture whose attendance has to be taken and the second will be the timestamp of when the QR code has been generated. This entire information will be encoded before a QR code is generated. This generated QR code will be displayed via a projector or any big screen available to the teacher at the time through the computer of the teacher itself.
The last part of this system is a client-side mobile application which will be installed on the phones of all students. The role of the students would be to scan the QR code displayed by the teacher through the application itself. Once the QR code has been scanned, the information it carries will be decoded. Out of this decoded information, the timestamp will be extracted and verified for its validity. Once the QR code verification is successful, the student will be prompted to scan and verify their faces via the front-facing cameras on their smartphones using facial recognition. Once the QR and face verification is done, the student’s attendance for the particular lecture is considered marked and then synced with the backend server.
VII. CONCLUSION & FUTURE WORK
In this paper, we have explored and studied various attendance management systems implemented using various technologies that are available in the market today. At the same time, we have explored the advantages and drawbacks of these systems and proposed our own system that attempts to overcome these limitations. The proposed system also aims to be very cost effective, amongst other factors, when compared to the existing systems, techniques, and technologies used by the current implementations of such systems. Finally, the proposed system comes with its own set of limitations which gives us further scope and room for improvement in the near future.
 Mobile Attendance using Near Field Communication and One-Time Password
 Biometric Smart Attendance Kit with Fingerprint Scanner by Using Microcontroller
 Design and Implementation of Iris Recognition Based Attendance Management System
 FaceTime – Deep Learning Based Face Recognition Attendance System
 Automated attendance management system based on face recognition algorithms
 Load Balanced GANs for Multi-view Face Image Synthesis
 Smartphones usage among university students