Attendance System using Real-Time Face Recognition

Overview

This project automates attendance management by leveraging real-time face recognition technology. Traditional attendance systems, whether manual or digital, often suffer from inefficiencies and inaccuracies. By integrating advanced facial recognition methods, this system offers a streamlined, contactless, and secure attendance process, ideal for educational institutions, workplaces, and other organizations.

Features

• Real-Time Face Detection: Utilizes Multi-task Cascaded Convolutional Networks (MTCNN) for accurate face detection and alignment.
• Face Recognition: Employs FaceNet embeddings generated in PyTorch to ensure high-precision face matching.
• Database Integration: Stores facial embeddings and user details in a MySQL database for secure and efficient management.
• Attendance Logging: Automatically marks attendance upon successful face recognition.
• User-Friendly Interface: Displays real-time video feed with attendance status updates.

Workflow

1. Dataset Preparation:
   • Collect images of individuals and preprocess them.
   • Generate facial embeddings using a pre-trained FaceNet model.
   • Store embeddings in a MySQL database with associated user details.
2. Real-Time Recognition:
   • Capture live video feed and detect faces.
   • Align faces and generate embeddings.
   • Compare embeddings with the database using cosine similarity.
   • Mark attendance for matched individuals.
3. Interactive Display:
   • Show live video feed with matched names or "No Match Found" status.

Tech Stack

• Programming Language: Python
• Libraries:
  • OpenCV: For video capture and image processing.
  • Facenet-PyTorch: For MTCNN (face detection) and FaceNet (embedding generation).
  • NumPy: Numerical operations.
  • MySQL-Connector: For database integration.
  • SciPy: For calculating cosine similarity.

Database:

• MySQL: Stores facial embeddings and user details.

Results

• The system successfully detects and recognizes faces in real time.
• Attendance is accurately logged in the database.
• The interface provides clear, interactive feedback on recognition status.

Challenges and Future Work

Challenges:

• Variations in lighting and image quality can affect accuracy.
• Faces with extreme angles or occlusions are harder to detect.

Future Improvements:

• Add data augmentation techniques for better performance.
• Optimize the system with a lightweight model for faster processing.
• Implement a web-based or mobile-friendly version for broader accessibility.
• A model that enables real-time user registration, automatically preprocessing images and generating embeddings for face recognition without requiring manual intervention.

Contributors

• Mangalya D. Phaye
• Ridima Garg