Ladage Aditya et al.; International Journal of Advance Research, Ideas and Innovations in Technology © 2019, www.IJARIIT.com All Rights Reserved Page |1439 ISSN: 2454-132X Impact factor: 4.295 (Volume 5, Issue 2) Available online at: www.ijariit.com E-attendance system using face recognition Aditya Ladage aditya.ladage@somaiya.edu K. J. Somaiya College of Engineering, Mumbai, Maharashtra Pooja Pache pooja.pache@somaiya.edu K. J. Somaiya College of Engineering, Mumbai, Maharashtra Sahil Maniar sahil.maniar@somaiya.edu K. J. Somaiya College of Engineering, Mumbai, Maharashtra Sheetal Pereira sheetalpereira@somaiya.edu K. J. Somaiya College of Engineering, Mumbai, Maharashtra ABSTRACT Traditionally, the attendance of the students has been a major concern for the colleges and the faculties have to spend quite some time of their lectures for taking the attendance manually. In this paper, we are introducing a new way of attendance monitoring by making use of smartphones available with the teachers. We have suggested the use of YOLO algorithm for face detection and Siamese network for face recognition. This system will automatically mark the attendance of the students and thereby save the time and efforts for the faculties. The designed system will be quite efficient and reliable as Siamese network has proven to render high accuracies in face recognition. Keywords— YOLO (You Only Look Once), Siamese, CNN (Convolutional Neural Network), R-CNN (Region based Convolutional Neural Network), DNN (Deep Neural Networks), Darknet, CUDA 1. INTRODUCTION One of the considerably important issues for any college or university today is the attendance of the students. Attendance monitoring has been a tedious task for the faculties as they have to take attendance manually to avoid malpractices. At the end of each month, it is also the faculties’ duty to generate attendance report of every student. We are introducing a new paradigm for the attendance system which will automate the process eventually reducing the work of the faculties. We have suggested the use of cameras in the classroom to capture the images of the students subsequently detecting and recognizing the students in the image and marking their respective attendance. For detecting the faces, we are making use of YOLO algorithm which is actually used for object detection but we will tune it for detecting faces. The recognition of the faces will be done using Siamese network which works on facial features of the detected faces for identification. The identified students will be marked and reports will be generated for every student. At the end of the month, our system will be notifying the students, their parents and their respective mentors about their attendance. In this way, the attendance monitoring system can be made with very little human interference. 2. RELATED WORK The biometric system for attendance is well known and widely used these days but the problem with the biometric system is that it is delay based system where people have to form a queue to scan their fingerprints and get their attendance marked. Thus, we intend to remove this delay factor by making use of image processing. In recent years, image processing has been used to process important information from the image which involves interpretation of the image for extracting useful information from the image. Also with the increasing amount of popularity of smartphones among the common people, there has been a demand to exploit the mobility of these devices and create easily accessible systems. This is why we have decided to create out attendance system which would make use of smartphones rather than any dedicated setup for attendance. The face of a person is unique but in a manner that it possesses a set of features that might resemble with some other face in one or the other way. Our system will make use of this unique set of features possessed by each student to recognize faces and mark attendance. There have been many systems created for face detection and subsequently recognition for attendance system but there have been certain drawbacks in each of them. One way is using Viola- Jones algorithm and Back Propagation Neural Network (BPNN). In BPNN, there are two weighted propagations. In the forward propagation, the input is fed through the network to generate output activation of the propagation and in backward propagation, a feedback network is formed by feeding the output as the input in order to generate a difference between target and actual output. Hence, while training, two propagations are required for every epoch instead of one as in almost all other