Automatic Attendance System using Deep Learning

1. Automatic Attendance
System Using Deep
Learning
Presented By: Sunil Aryal
Paper No: 438

2. 1 A brief overview
Introduction
2 Reason behind development of this idea.
Purpose
3 Technicality how entire project is working
Methodology
AGENDA
4
Conclusion and Result
Conclusion

3. Introduction
• Automatic attendance system is automated version of physical presence of
person without any manual interference.
• Real-time face recognition algorithms are used and integrated with existing
University management system.

4. Introduction
• Machine Learning: Use of algorithm in such way that it makes prediction on
basis of previous observations.
• With advancement in parallel computing, complex learning like facial
recognition (Computer Vision) are being used in today’s time with less cost error
and high accuracy.

5. In huge classes, there is high probability of error in entire
process by use of conventional method but our system is
highly accurate with accuracy of 98% with no or less
biased.
Accurate(Less Biased)
As system is automated, the exist no biasness in system
and system remains transparent.
Transparency
Traditional system of marking attendance require
manual interference but this system do not involve
human interaction with system.
Remove manual work
20% of total time of lecture gets wasted in taking
attendance but with this system we can mark attendance
in no time or real time attendance system.
Time Efficient
Why this project was built?
Purpose

6. Methodology
This system consist of 5 major steps.
2
3
4
1
5
Take picture using high
definition camera
Detect faces
Recognize Faces
Database Processing
Mark attendance

7. Detect Faces
• Faces are being detected by using CNN(Convolutional Neural Networks) and are optimized using
transfer learning to speed up learning process.
• Landmark detection is used to pre-process the detected faces.
• To detect multiple faces from single image MTCNN is used.

8. Detect Faces

9. Face Recognition
• Our network architecture for face recognition is based on ResNet-34 from the Deep
Residual Learning for Image Recognition paper by He et al., but with fewer layers and
the number of filters reduced by half with addition of feedback.
• The network itself was trained by Davis King on a dataset of ~3 million images. On the
Labelled Faces in the Wild (LFW) dataset the network compares to other state-of-the-
art methods, reaching 97% accuracy.

10. Face Recognition

11. Database Processing and Marking Attendance
• After recognition of faces with unique id from image captured are ingested to database with
existing university management system.

12. Results
97%
Deep Metric Learning
23%
Eigen Face
13%
Fisher Face
20%
LBPH

13. References
[1] Bae, Mi-Young, and Dae-Jea Cho. "Design and implementation of automatic attendance check system using BLE
beacon." International Journal of Multimedia and Ubiquitous Engineering 10, no. 10 (2015): 177-186.
[2] Jogiji, Aditya, and P. Ghate. "WSN Based Automatic Attendance Monitoring System." International Journal of
Computer & Mathematical Sciences (IJCMS) 6, no. 8 (2017).
[3] Ahmedi, Aziza, and Suvarna Nandyal. "An Automatic Attendance System Using Image processing." The International
Journal Of Engineering And Science 4, no. 11 (2015): 1-8.
[4] Krishnan, R. Ramya, R. Renuka, C. Swetha, and R. Ramakrishnan. "Effective Automatic Attendance Marking System
Using Face Recognition with RFID." IJSRST 2920 (2016): 158-162.
[5] Shengli, K., Jun, Z., Guang, S., Chunhong, W., Wenpei, Z. and Tao, L., 2015. The Design and Implementation of the
Attendance Management System based on Radio Frequency Identification Technology. In International Conference on
Electronic Science and Automation Control,(ESAC 2015) Google Scholar.

14. Questions?