Deep Learning-Based Facial Recognition System For Automated Attendance

Attendance of students in universities is an important task to evaluate the regularity of students. Therefore, maintaining and monitoring the attendance of students has been a major concern for most of the universities. Traditionally, attendance of students is marked manually that becomes computatio

Project Title

Deep Learning-Based Facial Recognition System For Automated Attendance

Project Area of Specialization

Artificial Intelligence

Project Summary

Attendance of students in universities is an important task to evaluate the regularity of students. Therefore, maintaining and monitoring the attendance of students has been a major concern for most of the universities. Traditionally, attendance of students is marked manually that becomes computationally expensive for the universities in term of time having hundreds of students. Since it has hard and tedious job to manually mark the attendance of hundreds of students on a piece of paper. To avoid manual attendance, biometric based attendance systems, for example, RFIDs, tags, fingerprint scanner, were introduced that automatically mark the attendance. However, these systems suffer from the following limitations.

1. Fraud and security risk: Lost, stolen, or exchanged cards can be used by people who should not have access.
2. Impractical: In case of frequent use, having to grab a card or tag, again and again, causes frustration and irritation.
3. Unhygienic: Objects like card scanners, door handles, and doorknobs are often susceptible to bacteria and viruses. Touching these can increase the spread of bacteria and viruses.
4. Costs and effort: In case of forgotten or lost cards, the old cards must be deactivated, forms for a new card must be completed and a new card must be paid for.

 To avoid problems, an alternative way is to use Queue based facial recognition-based attendance system. In this system, an image is captured by the camera and compare with the database of faces. The attendance of an students is marked if the match is found. Such facial recognition-based attendance system costs a lot of time, since the system requires all the students to stand in a queue for scanning their faces.

For this purpose, we propose an efficient and effective group-based attendance system by employing deep facial-recognition frameworks. The proposed framework will take an image of a group of people from the smart phone camera and automatically mark the attendance of students by recognizing their faces.

Project Objectives

The main objective of the project is to enhance the features of the Queue-Based Facial Recognition system. For this purpose, we need to add a new feature of Group based Recognition system. In this feature, we would use the deep learning technique. Deep learning is one of the most up-to date ways to enhance group-based recognition and improve the accuracy of facial recognition. Deep learning extracts unique facial embeddings from images of faces and uses a trained model to recognize photos from a database in other photos and videos. There will be no extra time effort in group base recognition system.

Project Implementation Method

The project implementation method of system involves following Artificial Intelligence (AI) Modules:

• Phase-1: Attendance Marking

• Phase-2: Face Detection and Recognition

• Phase-3: Updating Database

Module-1: During this phase, Face is recognize using camera and mark attendance in excel sheet using matching faces in database.

Module-2: We will develop an algorithm to identify videos during this phase. The algorithm will detect the face in the scene to identify people in videos. Later, the algorithm extracts feature and matches them with the database data. Finally, the algorithm recognizes the individual and display their information. However, the performance of the identification algorithm depends on the reliability and accuracy of detection methods. For detection, we will develop a deep learning framework. It is further divided into two parts

Part 1: Multi-scale face Detection and Refinement

In general, group-based face detection is a subset of the object detection task. Detecting a group of faces in a single image is a tough problem since face detectors rely on facial traits that are difficult to extract in natural environment. Distance from the camera generates perspective distortions, making it harder to distinguish between faces of different sizes. For example, the size of the face near the camera seems huge, whereas the size towards the rear appears little. To detect faces precisely, we must first tackle the scale problem, which is the object detectors.

To solve this problem, we will use deep learning developed models that will detect human faces in complex environment. Once the faces are detected, the next step is to extract the patches corresponding to faces. However, during this step, the extracted patches are noisy which need to be refined and improve before processed by face-recognition module. To recover the information and improve the quality of patches, we will use employ deep learning model-based filters.

Part 2: Face- Recognition:

During this step, we will employ deep learning methods to identify the face of students in patches obtained during part 1. One of the advantages of using deep learning models is that the models automatically learn rich hierarchical features and even perform better than humans in recognizing human faces in complex environment. We will use open source DeepFace, which is a lightweight face-recognition library in Python. This library includes most of state-of-the-art deep learning models, for example, VGG-Face, FaceNet, Dlib, ArcFace, Deep ID, Facebook DeepFace, and OpenFace.

Module-3: In this phase, we will design a database schema to store the information efficiently. We also provide a mechanism that will help display the data by storing preprocessed data.

Benefits of the Project

We summarize the benefits of system as follows:

1. The system will have wide range of applications in attendance management.
2. The system framework will provide support to security personnel to discover potential behaviors and ensure public safety in public places like airports, malls, train stations etc.
3. System can automatically collect meaningful information, for example, identification, count of people etc. from the scene which is otherwise impossible in the traditional setup.
4. System can be employed in the shopping mall to detect workers in and out timings.
5. System will provide a remote management system for various industries to improve the accountability of employees by tracking and understanding the individuals as well as groups of people.
6. A smart portable prototype that can be installed anywhere very easily.
7. A 24/7 smart uninterrupted monitoring system, where no labor is required to operate.
8. System is scalable, user-friendly, and easy to install without affecting existing infrastructure.

Technical Details of Final Deliverable

The final Deliverable is Verification and Validation of Whole system. In this task, complete system will be evaluated and validated with different real time scenarios. We will check the algorithms and their efficiencies with different performance metrics. The system will be tested using multiple inputs from fixed cameras connected with the web applications.

Final Deliverable of the Project

HW/SW integrated system

Core Industry

IT

Other Industries

Core Technology

Artificial Intelligence(AI)

Other Technologies

Sustainable Development Goals

Industry, Innovation and Infrastructure

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com