A Non-Invasive Attendance System Based on Face Identification (NASFI)

Project Title

A Non-Invasive Attendance System Based on Face Identification (NASFI)

Project Area of Specialization Artificial IntelligenceProject Summary

This project, titled as, “A Non-Invasive Attendance System Based on Face Identification (NASFI)”, is about the development of an automated attendance system of our university that is based on facial identification in uncontrolled environment like a running class.It is a general experience that manual attendance systems, that are either totally manual or through university portal are prone to human error and are also time-consuming. Therefore it has become imperative to develop an automated non-invasive attendance system that is easy to sustain and also accurate enough so that the teachers and staff are relieved with the cumbersome task of taking and maintaining attendance. The proposed NASFI provides both, i.e. easy maintenance, as it will be fully automatic with no manual input required, and accuracy, so that it is acceptable for the sensitive nature of reliability of attendance data.

Project Objectives

The main purpose of this project is to develop a smart attendance system which does not require any effort and avoids discomfort which the teachers normally face while going through manual attendance or the attendance using portal.Specific objectives of this project are as follows:

1. To design a fully automated attendance system that is capable of keeping record of attendance with 100% accuracy
2. To design and implement a Non-Invasive and easy to maintain attendance system that can mark attendance during a running class without disturbing the environment and execution in anunobservable manner.
3. To evaluate state of the art face recognition algorithm and implement them on locally customize class images for attendance gathering.
4. To identify each and every student in the class separately.
5. To customize the proposed system according to the requirement of the University.
6. To prepare a clear strategy for marketing the system in local colleges and universities in collaboration with the ORIC office.
7. To design anintelligent algorithm for marking attendance that can tolerate even partially recognized facesduring the face recognition process.

Figure 1 shows the algorithmic block diagram of the proposed system. A camera is used to capture multiple images of the whole class during its normal execution. These images contain multiple faces as they capture all participants of the class as a cohort. These images are passed through a face detection algorithm which detects and separates face of each and every student present in the class. The detected faces are then pre-processed so that they can be straightened, aligned and uniformly illuminated. The faces are then selected out of these multiple images based on their quality. Once the pre-processing completes, the faces are passed through a model which compares the pre-processed faces with the faces in local database which contains faces of each student enrolled for a particular class. As soon as a face is matched with the one in local database, that particular student is marked ‘Present’, thus completing the attendance. In addition, the system will be intelligent enough to tolerate even partially recognized faces in face identification module by profiling rank-2, and rank 3 faces also.

                           

Figure 1 Algorithmic Block of Proposed System

Project Implementation Method

The development of proposed NASFI includes development of following algorithmic modules:

1. Detection of students faces
2. Identification of students faces 
3. Reliable Class attendance and record maintenance

The following table show the implementation steps which will be followed during the course of the project.

Sr. no.	Activities to be performed
1.	Literature Review(Completed) Face detection in images from uncontrolled environment Face recognition in uncontrolled environment Gap Analysis and improvement required Shortlisting of algorithm
2.	Implementation of Algorithms(Completed) Pre processing of face images Face detection and separation of individual faces
3.	Testing of Algorithms(Completed) Component level testing Integrated level testing
4.	Post Processing Algorithms Ranking of Candidates Logic for reliable attendance
5.	Graphic User Interface and System integration System integration and interfacing with GUI Database Management
6.	Testing and Evaluation Component level testing Integrated testing

Sr. no.

1.

2.

3.

4.

5.

6.

Benefits of the Project

Being a university student, it was not that difficult to point out the most common problem every professor faces every day and that is to first log in to a portal and then call names of every student which is, no doubt, a hectic and time-consuming activity to do during the class timings. Hence, to relieve them of this difficult job, it was decided to develop an attendance system which uses facial identification and is non-invasive that does not disturb the normal execution of the class.

Summarizing the above discussion, he main limitations of the current manual attendance system are mentioned below:

• It is time-consuming
• It is difficult to maintain
• It is prone to human error

The above mentioned factors were the main source of motivation behind selecting this project. A more reliable and smart attendance system which avoids the chances of any type of discomfort caused by existing attendance systems is the need of hour.

The main advantages/benefits of the proposed NASFI are

• It will reduce the amount of time wasted on manual attendance/attendance using portal
• It will not be that much difficult to maintain as compared to other approaches
• It will not be prone to human error
• As a student one of the main objectives of FYP is learning and hands on experience in artificial intelligence. In this project we will implement state of the art deep learning algorithm which will give us an invaluable experience of exciting field of artificial intelligence. While implementing deep learning algorithms, we will have invaluablehands-on experience on this most exciting and popular field.
• As mention in the earlier section this product has also a great business potential also. This project will be applicable in schools, colleges, universities, etc. as its solves a major issue in everyday running of these institutions. After successful completion of the project we will customize the project according to the needs of our university and install a prototype that will work in parallel to the existing system. After thorough testing and evaluation this system may replace the existing manual attendance system.

Technical Details of Final Deliverable

The proposed NASFI will be a software/hardware integrated system in which face detection, face recognition, pre-processing and post-processing algorithms will be implemented on GPU, and the intelligent attendance system with database management will be implemented on CPU. The system will also be integrated with camera in real time.The following paragraphs outline some of the technical detail regarding these algorithmic and system modules.

Face Detection algorithmsare used to detect the face of each and every student present in the class. The two most frequently and reliable used face detection algorithms are:

• Adaboost based Viola and Jones algorithm (available in OpenCV library)
• MTCNN (Multi-Task Cascaded Convolutional Neural Network)

The Viola Jones Algorithm was one of the widely used algorithm for face detection before the advancement in deep learning techniques.It is based on naïve classifies combined together through AdaBoost based ensemble. These classifiers also use some basic masks to detect the face region in an otherwise larger image .

MTCNN or Multi-Task Cascaded Convolutional Neural Networks is a neural network which detects faces and facial landmarks on images. It was published in 2016 by Zhang et al. MTCNN is one of the most popular and most accurate face detection tools today. It consists of 3 neural networks connected in a cascade . 

We implemented and analyzed the results of both the above algorithms and based on their comparison, we decided to use MTCNN for face detection as it outperforms the Viola Jones algorithm in terms of accuracy.

Face Recognition algorithms are used to compare the template faces with the reference faces in the local database. Following algorithms are some of the most popular and frequently used algorithms for face recognition:

• Feature Extraction Based Algorithms like Histograms of Oriented Gradients  and Local Binary Patterns 
• Subspace Based Algorithms like Fast-PCA, LDA and Laplacian Faces Based Algorithms 
• Deep Learning based algorithms like Sphere face ,Cosface and Arcface 

After comparing the efficiency and accuracy of each above mentioned algorithm, deep learning based algorithms are opted as they report the highest accuracy for recognizing and identifying faces. Even in uncontrolled environment, some recent studies have shown that deep learning give better results as compared to traditional approaches.

The selected Face Detection and Face Recognition algorithmsare both based on deep neural networks, which are computationally intensive therefore for real time implementation these algorithms will be implemented on efficient GPUs.

The final output will be a software/hardware integrated system in which GPU will be used for implementing deep neural networks, while the database management system and overall attendance logic along with camera interface will be implemented on CPU.The attendance logic will ensure 100% accuracy of attendance 

Final Deliverable of the Project HW/SW integrated systemCore Industry EducationOther Industries Security Core Technology Artificial Intelligence(AI)Other Technologies Big DataSustainable Development Goals Quality EducationRequired Resources

Elapsed time in (days or weeks or month or quarter) since start of the project	Milestone	Deliverable
Month 1	Literature Review	•Face detection in images from uncontrolled environment •Face recognition in uncontrolled environment •Gap Analysis and improvement required •Shortlisting of algorithm
Month 2	Implementation of Algorithms	•Pre processing of face images •Face detection and separation of individual faces
Month 3	Testing of Algorithm	•Component level testing •Integrated level testing
Month 4	Post Processing Algorithms	•Ranking of Candidates •Logic for reliable attendance
Month 5	Graphical User Interface and System integration	•System integration and interfacing with GUI •Database Management
Month 6	Testing and Evaluation	•Component level testing •Integrated testing