See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/342331838

Smart Attendance System Using Face Recognition

Article in Journal of Advanced Research in Dynamical and Control Systems · May 2020
DOI: 10.5373/JARDCS/V12SP5/20201860

CITATIONS READS
5 3,758

5 authors, including:

Dr.Sabeenian R.S
Sona College of Technology
107 PUBLICATIONS 497 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Machine and deep learning in image processing platform View project

tumor segmentation View project

All content following this page was uploaded by Dr.Sabeenian R.S on 28 June 2020.

The user has requested enhancement of the downloaded file.

Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

Smart Attendance System Using Face

Recognition
Dr.R.S. Sabeenian, Professor, Head of ECE Department, Sona College of Technology, Salem, India.
E-mail: [email protected]
S. Aravind, UG Scholar, Department of ECE, Sona College of Technology, Salem, India.
E-mail: [email protected]
P. Arunkumar, UG Scholar, Department of ECE, Sona College of Technology, Salem, India.
E-mail: [email protected]
P. Harrish Joshua, UG Scholar, Department of ECE, Sona College of Technology, Salem, India.
E-mail: [email protected]
G. Eswarraj, UG Scholar, Department of ECE, Sona College of Technology, Salem, India. E-mail: [email protected]
Abstract--- Making Attendance as smarter using recent technology and recognition in deep learning algorithms to
build a system that will detect and recognize frontal faces of students in a classroom. ―A face is the front part of a
person’s head from the forehead to the chin, or the corresponding part of animal‖ (Oxford Dictionary). Face
identification is one of the biometric methods to make essential for this system. In human sociable, the front part of
face is the most useful data as it contains important data about a group or individual. All person has the skill to
identify individuals from their individual faces. The suggested solution is to develop a working model of a system
that will promote attendance system in a classroom by recognize the frontal faces of students from a picture taken in
a classroom. By making this framework, the problem of intermediary and students being marked present even
though they are not physically present can easily be solved. In recent years, research has been taken out and face
identification and recognition systems have been developed. Some of which are used on social media platforms,
banking apps, government offices, etc. Since face recognition plays a vital role in mobile devices to make it perform
and functions on particular operations. Using this idea as a base we make this for smart attendance system and it will
spend minimum time than manual attendance. This model will be a successful technique to conduct the attendance
and records of students.
Keywords--- Convolutional Neural Network (CNN), Deep Learning, MXNet, TensorFlow, Onnx Model, Database,
Training and Recognition.

I. Introduction
Every institution requires a robust and constant system to record the attendance of their students. and every
institution have their own methodology to do, many in numbers are taking attendance manually with a attendance
registers by calling their respective names during lecture periods and little have takeup biometrics system such as
RFID card reader [1], fingerprint and iris system to take the attendance in daily basis.The ordinary method of calling
the names of students manually is time consuming process. RFID card system is assigned to each student with their
corresponding identity but there is posibility of card loss or unapproved person may mistreat the card for fake
attendance. while in other biometrics such as voice recognition, iris[2] or fingerprint [3], they are not 100% accurate.
A facial recognition system is a technology capable of recognizing or validating a person from a digital image
or a continuous image frame from a video source. There are numerous methods in which facial recognition systems
works, they work by differentiating selected facial information from given information with faces within a database.
It is also termed as Biometric Artificial Intelligence based on use cases that can individually identify a student by
figure out patterns based on the student's facial information and shapes.
The main use of face recognition [4] for the direction of attendance marking is the quick and smartest way of
attendance management system. Face recognition [5] is more proper and faster technique among other approach and
scale down chance of proxy attendance.
Face recognition contribute static identification that is a student which is to be determined does not to need to
take any action for its integrity. Face recognition contain following two steps, first step concern the detection of

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1079
Received: 17 Mar 2020/Accepted: 18 Apr 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

faces and second step contain of recognition of those identified face images with the existing database that contain
existing images [6].Hence it contains number of face detection and recognition methods introduced. Face
recognition entirety either in form of actualization based which covers the information of whole frontal face or
element based which covers the mathematical feature like eyes, nose, eye brows, and cheeks to perceive the face.
Our system accepts face recognition progress to slow down the defect of existing system with the help of deep
learning, it have need a good quality camera to capture the images of students, the detection process is done by
Scatter diagram of oriented angle and recognizing perform through deep learning. The frontend side (client) which
consist of GUI which is established on electron JS and backend side consist of logic and python (server), an IPC
(Inter Personal Communication) bridge is developed to communicate these two loads. The images that obtained by
the camera is transfer to system for further analysis, the input image is then correlated with a set of reference images
of each of the student and record the attendance of the students.

II. Proposed Method

Fig. 2.1: Model of Proposed Method

2.1 Model
The usage of Open Neural Network[7] Exchange, it is accessible to give the ability of optimization to the Data
Scientists. It implement the advance of ONNX compatible runtimes and libraries to every tool which prefer the
exported ONNX models, and it ensure in the development of the performance on some of the use cases in the
reality. At a immense level, ONNX is described to allow framework interoporability. There are many essential
machine learning libraries in different languages — PyTorch, TensorFlow[8], MXNet, and The concept is that you
can train a model with one tool stack and then arrange it using another for assumption and prediction. To establish
this interoperability you must consign your model in the model.onnx format which is serialized representation of the
model in a protobuf file.
2.2 Face Detection
In order to detect 0e a face, we would first need to detect a face from an image. The Ultra-light face detector [9]
was matchless in terms of speed and introduce a relatively good accuracy.

To use the ultra light model, python is highly required and it need specific model to perform an action. we will
be using the pretrained ultra_light_640.onnx model, we have to resize the input image to 640x480. After pre-
processing the given image, we need to prepare the ONNX model and create an ONNX runtime.
2.3 Face Recognition
MobileFaceNet is a deep learning algorithm and attain accuracy upto 85 percent on labelled faces in the given
dataset [10 ], and a 90 percent accuracy on recognising faces in the given dataset. The network consume around a
million parameters taking only 24 ms to run and introduce results on gpu. On comparing this performance to
accuracies of 98.70 percent and 89.27 for ShuffleNet, which had more parameters and takes a little longer to execute
on the CPU.The developers have made it easy to replace the global average pooling layer in the CNN with a depth

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1080
Received: 17 Mar 2020/Accepted: 18 Apr 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

wise convolution layer [11] and it will improves performance on facial recognition. The remarkable achievement is
that MobileFaceNets achieves comparable accuracy with very small budget.

III. Result and Discussion

The working of the system we proposed involves three steps:
1. Creating a database.
2. Training the model.
3. Recognizing the real-time data.
Before starting, let us import all necessary libraries. The libraries includes,
1. Opencv: to process the image/video.
2. Numpy: to work with pixels in form of arrays.
3. Onnx: to work with onnx models.
4. Dlib: for facial mappings (landmarks).
5. Os: to change/create directory and also to open files in the directory.
6. Imutiles: to manipulate the image (preprocessing).
7. Tensorflow: to create layers and create/load models.
8. Pickle: to serialize or deserialize to python objects (like list, tuples, dictionary).
9. Xlsxwriter: to create/read/write to excel sheet.
3.1 Database
Firstly, a database is to be created, where the videos of students are uploaded. It is preferred to be in.mp4 format,
for the model to extract features from it. More clear and variated your video is, the more features it could generate.
And each folder is allotted for each student to have their corresponding video.

Fig. 3.1: Database

3.2 Training
Then next we have to train the model. The is trained with the images extracted from the video. There is a
preprocessing step [12], that is taken to extract features from the video and to generate variations or alignments from
the source video. It is done by using face_mutils feature of imutils and shape_predictor_68_face_landmarks.dat file.
Then these extracted features are dumped to a.pkl file, with names and images mapped properly using pickle.

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1081
Received: 17 Mar 2020/Accepted: 18 Apr 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

Fig. 3.2: Training

3.3 Recognition
To recognise a face, it is mandatory for it to be detected. We have tried many face detectors like Yoloface,
MTCNN, HOG, Ultra light face detector. And due to an accurate detection with less computational power, we have
chose ultralight face detector. It is capable of detecting probably 70-80 faces even in a single frame. And the
computational time is also less in compared with other detectors.

Fig. 3.3: Recognition

to use the ultra light model, the following python (python version 3.6) packages are required:
onnx==1.6.0, onnx-tf==1.3.0, onnxruntime==0.5.0, opencvpython==4.1.1.26,
tensorflow==1.13.1
As we will be using the pretrained ultra_light_640.onnx model, we have to resize the input image to 640x480. If
you are using the 320 model, please resize accordingly.
After pre-processing the image, we will have to prepare the ONNX model and create an ONNX inference
session.
After detection, the next main step is recognition. Face recognition can be done through many techniques like
Openface, Resnet, FaceNet, VGGFaceNet,MobileNetV2. Among this we have chosen MobileNetV2, due to its high-
accuracy real-time face verification [14] on mobile and embedded devices.

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1082
Received: 17 Mar 2020/Accepted: 18 Apr 2020
Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

To recognise a face, simply load the embeddings dataset with corresponding names(.pkl file).Then use Euclidean
distance and threshold to determine the difference between the target face with that of a person in the database. And
the one with the minimum difference will be predicted to be the person from dataset. The minimum difference must
be greater than that of the threshold to consider it as known person. When the difference is more, the person will be
subjected as unknown.

Fig. 3.4: Result

With that, we have created a system that can perform real-time face recognition with CPU. Although it is only
running at around 13 FPS, it is comparably much faster than using complex CNNs.
However, there are still many things we could do to improve the performance (both the accuracy and speed) of
this system. Potentially, we can apply knowledge distillation to compress the current model and further reduce the
model size using low bit quantization. Moreover, we could improve the accuracy using other machine learning
classification methods[15] on the embeddings.

IV. Conclusion
The project is done to take a attendance in smarter way. MobileFaceNet is a deep learning algorithm and attain
accuracy upto 85 percent on labelled faces in the given dataset, and a 90 percent accuracy on recognising faces in
the given dataset. This would help the management by reducing time taken for making attendance manually and
replace RFID card system is assigned to each student with their corresponding identity.so posibility of card loss
won’t affect the attendance of the students and fake attendance will be minimized. Thus if the model is implemented
in real life and it will help the institude by managing attendance and to avoid unknown entry to the institution

References
[1] T.S. Lim, S.C. Sim, M.M. Mansor, ―RFID Based Attendance System‖, 2009 IEEE Symposium on
Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia.
[2] S, Kadry; K. Smaili, ―A Design and Implementation of A Wireless Iris Recognition Attendance
Management System‖, ISSN 1392 – 124X Information Technology and Control, 2007, Vol.36, No.3.
[3] M. K. P. Basheer, C. V. Raghu, ―Fingerprint attendance system for classroom needs,‖ in Proc. India
Conference (INDICON), 2012 Annual IEEE, pp. 433-438, 7-9 Dec. 2012.

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1083
Received: 17 Mar 2020/Accepted: 18 Apr 2020
 Jour of Adv Research in Dynamical & Control Systems, Vol. 12, 05-Special Issue, 2020

[4] S. Chintalapati; M.V. Raghunadh, "Automated attendance management system based on face recognition
algorithms," Computational Intelligence and Computing Research (ICCIC), 2013 IEEE International
Conference on, vol., no., pp.1,5, 26-28 Dec. 2013, doi: 10.1109/ICCIC.2013.6724266.
[5] Nirmalya Kar and Ashim Saha; Study of implementing automated attendance system using face recognition
technique unit", Computer Architecture (ISCA) 2017 ACM/IEEE 44th Annual International Symposium on,
pp. 1-12, 2017.
[6] Dalal, N. and Triggs, B.: Histograms of oriented gradients for human detection. In: IEEE Computer Society
Conference on Computer Vision and Pattern Recognition (CVPR). pp. 886–893. (2005).
[7] N. Rotem, J. Fix, S. Abdulrasool, S. Deng, R. Dzhabarov, J. Hegeman, et al., "Glow: Graph lowering
compiler techniques for neural networks", CoRR, vol. abs/1805.00907, 2018.
[8] N. P. Jouppi, C. Young, N. Patil, D. Patterson, G. Agrawal, R. Bajwa, S. Bates, S. Bhatia, N. Boden, A.
Borchers et al., "In-datacenter performance analysis of a tensor processing unit", Computer Architecture
(ISCA) 2017 ACM/IEEE 44th Annual International Symposium on, pp. 1-12, 2017.
[9] T.-Y. Lin, P. Dollar, R. Girshick, K. He, B. Hariharan, and ´S. Belongie. Feature pyramid networks for
object detection. In CVPR, 2017. 1, 2, 4.
[10] Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., et al.: Mobilenets: Efficient
convolutional neural networks for mobile vision applications. CoRR, abs/1704.04861 (2017).
[11] Schroff, F., Kalenichenko, D., Philbin, J.: Facenet: a unified embedding for face recognition and clustering.
In: CVPR (2015).
[12] Han, S., Mao, H., Dally, W. J.: Deep compression: Compressing deep neural network with pruning, trained
quantization and Huffman coding. CoRR, abs/1510.00149 (2015).
[13] T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollar. Focal ´loss for dense object detection. In ICCV,
2017. 1, 2, 4.
[14] G. B. Huang, M. Ramesh, T. Berg, and E. Learned-Miller. Labeled faces in the wild: A database for
studying face recognition in unconstrained environments. Technical report, 2007. 6.
[15] Sun, Y., Wang, X., Tang, X.: Deeply learned face representations are sparse, selective, and robust. In:
Computer Vision and Pattern Recognition, pp. 2892–2900 (2015).

DOI: 10.5373/JARDCS/V12SP5/20201860
ISSN 1943-023X 1084
Received: 17 Mar 2020/Accepted: 18 Apr 2020

View publication stats