symmetry-12-00307

Implementation of CCTV-Based Attendance Taking Support System Using Deep Face Recognition

Authors

  • Ngo Tung Son

  • Bui Ngoc Anh

  • Tran Quy Ban

  • Le Phuong Chi

  • Bui Dinh Chien

  • Duong Xuan Hoa

  • Le Van Thanh

  • Tran Quang Huy

  • Le Dinh Duy

  • Muhammad Hassan Raza Khan

Abstract

The implementation of a CCTV-Based Attendance Taking Support System (ATSS) utilizing deep face recognition (FR) technologies represents a significant advancement in the realm of security monitoring. While face recognition technology has made progress, it primarily excels in controlled settings, presenting challenges when confronted with varying lighting conditions, motion blur, and camera resolution discrepancies. The objective of this project was to design an effective system and conduct empirical comparisons of various machine learning libraries to develop an ATSS optimized for reliability and accuracy, deployed at FPT Polytechnic College involving a sample of 120 students from five distinct classes. The results demonstrated a highly accurate attendance recording system that is adaptable to diverse environmental conditions.

1. Introduction

1.1 Problem and Motivation

CCTV Functionality: The use of CCTV systems not only plays a crucial role in monitoring building security but also enhances the collection of data that can be utilized for various applications including attendance tracking.

Benefits of Using Face Recognition for Attendance:

  • Time Efficiency: Automates the attendance process, significantly reducing the time and effort required to manually check attendance.

  • Management Evidence: Provides verifiable evidence of attendance for management and administrative purposes.

  • Health Considerations: Minimizes the potential spread of infectious diseases, making it a safer alternative to physical checks.

Existing Systems and Their Limitations: Traditional methods such as fingerprint recognition involve issues such as variability due to environmental factors and high costs associated with camera installations and maintenance tasks. These factors necessitate a more efficient solution like face recognition.

1.2 Related Works

Deep Learning in Face Recognition: Numerous advancements exist in the field of deep learning, specifically concerning face recognition, with significant focus on algorithms for feature extraction and model enhancement. Notable systems include ArcFace, SphereFace, FaceNet, and Cosface that address various aspects of facial recognition accuracy and robustness.

Modules of Face Recognition (FR):

  1. Face Detector: A critical component that localizes faces within images or video streams ensuring accurate identification.

  2. Landmarks Extractor: Aligns detected faces based on key facial features for precise processing.

  3. Feature Descriptor: Generates a unique numerical representation encoding the identity-related features of the face for recognition purposes.

1.3 Challenges of Face Recognition in Attendance Systems

  • Accuracy Requirements: Attendance systems necessitate nearly 100% accuracy; any discrepancies can significantly impact assessments and lead to incorrect reporting.

  • Environmental Constraints: Since CCTV primarily operates under security protocols, the effects of the attendance system must be minimal to avoid disrupting existing workflows.

  • Real-World Performance: Algorithms need to display robust functionality even under less than ideal conditions commonly found in educational institutions.

  • Integration with Existing Systems: A successful deployment requires seamless integration with current academic information systems to leverage existing data.

1.4 Contribution of the Paper

This paper presents the development of a comprehensive algorithmic framework designed for effective attendance management utilizing CCTV technology. The proposed system encompasses a full design architecture that includes:

  • Job Master: Task manager that oversees data processes

  • Job Workers: Specialized components handling specific tasks of face recognition

  • Central Database: A reliable storage system for attendance data and related information

  • User Interface Applications: Interfaces for administrative and operational personnel to manage attendance records The focus is on performance optimization and reliability, ensuring actionable insights can be derived.

2. Proposed System

Attendance Taking System (ATSS)

  • The ATSS is devised to interact seamlessly with existing CCTV infrastructures without causing disruption to their primary functions.

  • The system architecture features various integral components including a media recorder and a user interface designed for ease of access.

  • Task Management: Tasks are scheduled and managed via a central Job Master to enhance the operational efficiency of the system.

2.1 Job Master

  • The Job Master is responsible for overseeing the data flow and scheduling tasks within the ATSS.

  • It integrates with the existing Academic Portal, ensuring synchronization of attendee lists and relevant scheduling details.

2.2 Job Workers

  • This component performs face recognition tasks in parallel to bolster overall system efficiency.

  • Processes managed by Job Workers include: Face Identification, Data Collection, and Feature Processing for optimal recognition performance.

2.3 Face Recognition Building Block

  • Data Sampling: Employs strategic selection of frames based on head pose detection to build a robust training dataset.

  • Region of Interest (ROI): Defines targeted zones within the video feed to optimize attendance checking effectiveness.

  • Frame Processing and Response Time: Strikes a balance between processing speed and accuracy, ensuring students are ideally positioned for recognition.

  • Summarization Algorithm: Integrates detection results to enhance identification accuracy and final output reliability.

3. Experiment and Results

3.1 Experimental Design

  • A comprehensive evaluation of the system was conducted with 120 students utilizing a training set comprised of 7490 labeled images for both training and testing phases.

  • The Multi-task Cascaded Convolutional Network (MTCNN) was employed for effective face detection, while the ArcFace model was adopted for accurate feature extraction.

3.2 System Accuracy

  • The ATSS achieved elevated accuracy metrics with improved models demonstrating marked enhancement in managing instances of unknown faces or recognition errors.

  • The summarization process further enhanced performance metrics by significantly reducing the occurrence of false-positive identifications.

3.3 System Processing Time

  • A master-slave architecture was integrated within the system to facilitate processing efficiency. The implications of the number of operational workers on processing time were investigated, highlighting concerns related to resource allocation during intense processing tasks.

3.4 Application of Results

  • A Mobile Data Collector application was developed that utilizes video input from users to gather data in real-time.

  • The application supports tracking and management via user-friendly web interfaces, allowing for streamlined operational workflows.

4. Conclusion

The implementation of the designed system significantly enhances the efficiency of attendance taking through advanced face recognition techniques despite the various challenges posed in real-world applications. The experimental results underscore the need for continued improvement efforts, especially with larger datasets and the performance of classifiers to better categorize known and unknown identities. Future research suggestions include the exploration of 3D face recognition data to further bolster system capabilities and effectiveness.