ARTIFICIAL INTELLIGENCE PROGRAMMING PROJECT
Report 1 – Project Introduction
– Hanoi, May 2022 –
Table of Contents
I. Project Introduction
1. Overview
1.1 Project Information
1.2 Project Team
2. Problem & Motivation
3. Literature review and introduction
4. Contribution
5. Reference
I. Project Introduction
1. Overview
1.1 Project Information
• Project name: Time attendance system by facial recognition
• Group name: Team 3
1.2 Project Team
Full Name Email Mobile Role
Nguyễn Đình Dũng [email protected] 0867960176 Leader
Nguyễn Tiến Tuấn Anh [email protected] 0349947282 Member
Phạm Trung Hiếu hieupthe150046@fpt,edu,vn 0966249201 Member
2. Problem & Motivation
Currently, viral diseases are more and more common and have a very high risk of spreading
into epidemics, typically diseases such as flu, chickenpox, respiratory infections... typically
the recent covid 19 pandemic. Therefore, limiting contact and touching objects in crowded
places is a top priority. The use of traditional timekeeping machines, forcing employees to
have contact with timekeeping devices. Fingerprint time attendance machines must use
fingerprints to authenticate identity. The card timekeeper must use the card to swipe, thereby
increasing the possibility of infection if the unit or business has an infectious disease.
Traditional timekeeping methods such as swiping cards, recording names... still have many
shortcomings. Therefore, employees can completely "cheat" timekeeping. There are many
cases by swiping the attendance card or asking for registration. Therefore, it not only affects
work efficiency but also causes businesses to "unjustly pay" for hours when employees do not
go to work.
Meanwhile, face time attendance software can overcome the above problems. Employees
only need to install the face time attendance software on their mobile devices and
automatically time attendance at the working location without having to gather in a time
attendance area. that both creates convenience, fast, and ensures a safe distance.
The scope of the project is small company with about 20 employees
Facial recognition speed: The authentication scanning process is quite fast allowing
employees to get in and out in less than 2 seconds.
Input: each employee will send 2-3 photos of his face and personal information to create a
data set for the server.
3. Literature review and introduction
3.1 Current grading methods:
a.Timekeeping machines use magnetic cards
Advantage:
– Suitable for all different operating models of each unit and enterprise
– Can store data in a much larger amount than other forms of time attendance
– Use the identification mechanism by magnetic card or touch card – a special type of card
that is fully integrated with information in it
– Speed and ability to time attendance is relatively fast, about 2s per time
– The screen displays the timekeeper's information clearly, completely and accurately
- Stable time attendance management software
– With the card attendance solution since the timekeeping process does not depend heavily on
external influences
Limit:
– Due to the feature of using magnetic cards, it is possible to forget or lose the card, so
employees need to pay attention to always carry the card with them.
– There may be cases where the card is lost or the card is damaged so there will be additional
costs in remaking those cards
– It is easy to happen the process of timekeeping for each other
b. Fingerprint time attendance
Advantage :
- Suitable for areas such as offices, buildings
– Using fingerprint identification of each person, so unlike magnetic cards, there is no cost to
make a card.
- Ability to assign permissions to users
– There is no timekeeping for each other of the employees
- Stable operation process
Limit:
It will work well in the right environments, but in contrast to the characteristics of places
where employees have to work with grease or chemicals, timekeeping will be difficult, time
consuming and information is not guaranteed.
3.2 There are several facial recognition methods available on the market today.
FastWork's Face ID online attendance software
FastWork is the first company in Vietnam to provide an employee attendance solution
through FaceID facial recognition. App FastWork combines timekeeping - payroll - human
resource management with job management software to help create a comprehensive business
management system.
FastWork's online time attendance software with Face ID allows use on two web and mobile
platforms with two popular operating systems, iOS and Android. Help employees easily time
attendance at any work location by taking photos of their faces via the FastWork app on
phones with Wifi and GPS connections within a permitted radius.
FastWork also offers an email menu feature. When needing to report leave, overtime, work...
employees just need to choose the appropriate form and send it to the approver right on the
phone. By phone, the manager also makes quick application approval.
The software helps managers quickly grasp the situation of the number of employees going to
work and absent at each branch through the real-time reporting system. Human resource
management has also been significantly improved, timekeeping for the right people, locations
and times is recorded through GPS digital maps and FaceID censorship system.
3.3 Project with the same theme : FaceTime – Deep Learning Based Face Recognition
Attendance System
Advantage :
This deep learning based solution does not depend on the GPU in runtime. Thus, it could
be applicable in many other systems as a main or a side component that could run on a
cheaper and low-capacity hardware, even as a general-purpose Internet of things (IoT)
device.
Limit:
To get a collated image, facial recognition technology requires customers to turn at least
35 degrees towards the camera and not use masks, hats, hats.... This requirement is very
difficult, because we Can't force customers to do as requested, this will make them feel
uncomfortable.
Therefore, it is difficult to guarantee the accuracy of information when using people
counting cameras.
Currently, there are many face recognition methods such as Eigface, Fisherface, but using
Neural network achieves the highest results ~ 93%
The first step of the face recognition process is face detection. Face detection presents the
well-studied field in the computer vision domain. As a result of decades of research,
nowadays there are numerous machine learning algorithms applicable for this task. In recent
years, CNNs achieved advanced results in image classification and object detection .
The first step of the face recognition process is face detection. Face detection presents the
well-studied field in the computer vision domain. As a result of decades of research,
nowadays there are numerous machine learning algorithms applicable for this task. In recent
years, CNNs achieved advanced results in image classification and object detection .
Due to its runtime performance, for this step, a state-of-the-art CNN cascade is used for a face
detection task, introduced by Haoxiang Li et al in . The cascade consists of 6 CNNs, 3 CNNs
for binary classification (face and non-face) and 3 CNNs for bounding box calibration. A
Torch , machine learning framework is used for developing this face detector used as the first
step of face recognition model.
The third step presents the embedding process using the proposed system in – FaceNet, as
mentioned in the Section II. This method uses deep CNN for learning mapping from face
images to Euclidean space where distances match to the face similarity measurements. This
results in generating 128-bytes embeddings per face. Training of the network consists of
triplets: the face image of a target person, the test face image of the target person and the face
image of another person. OpenFace library with pre-trained FaceNet network was used for
training this deep CNN.
The final step of developing the face recognition model for tracking employees’ attendance
consists of training the classifier based on the previously generated embedding from
employees’ dataset by the deep CNN. Due to the fact that this system is based on smaller
dataset, linear Support Vector Machine (SVM) was applied for this classification task.
4. Contribution
Of all the face recognition methods, the neural network is the most effective. Therefore, our
team will implement this structure in face recognition with accuracy 95.02% in overall . The
system include 3 part: Dataset preparation and augmentation, Developing Face Recognition
Model and Integration with the Management System.
5. Reference
[1] Arsenovic, M., Sladojevic, S., Anderla, A., & Stefanovic, D. (2017). FaceTime — Deep
learning based face recognition attendance system. 2017 IEEE 15th International Symposium
on Intelligent Systems and Informatics (SISY). doi:10.1109/sisy.2017.8080587
[2] Siswanto, A. R. S., Nugroho, A. S., & Galinium, M. (2014). Implementation of face
recognition algorithm for biometrics based time attendance system. 2014 International
Conference on ICT For Smart Society (ICISS). doi:10.1109/ictss.2014.7013165
