Face Recognition Attendance System

A Python-based attendance system that uses facial recognition to automatically mark attendance of students.

Features

• Face detection and recognition using MTCNN and FaceNet
• Real-time attendance marking
• Dataset collection for new students
• Automatic CSV generation for attendance records
• Support for both CPU and GPU processing

Requirements

opencv-python
numpy
pandas
pytorch
facenet-pytorch
retinaface

Project Structure

• collecting_dataset.py - Captures and saves face images for new students
• mtcnn_training.py - Generates facial embeddings from collected images
• mark_attendance.py - Real-time attendance marking system
• attendance.csv - Stores attendance records

Setup and Usage

1. Install dependencies:

pip install opencv-python numpy pandas torch facenet-pytorch retinaface

2. Add new students:

   • Run collecting_dataset.py
   • Enter student name when prompted
   • The script will capture 100 face images automatically
   • Press 'q' to quit early

3. Generate embeddings:

   • Run mtcnn_training.py
   • This will create student_embeddings.pkl with facial features

4. Mark attendance:

   • Run mark_attendance.py
   • System will recognize faces and mark attendance
   • Attendance is saved in attendance.csv
   • Press 'q' to exit

Technical Details

• Face detection: MTCNN (Multi-task Cascaded Convolutional Networks)
• Face recognition: InceptionResnetV1 (pretrained on VGGFace2)
• Image processing: OpenCV
• Data storage: Pickle for embeddings, CSV for attendance

File Descriptions

• collecting_dataset.py: Uses RetinaFace for face detection and captures 100 images per student
• mtcnn_training.py: Processes collected images to generate facial embeddings
• mark_attendance.py: Real-time face recognition and attendance marking system

Output

The system generates attendance.csv with the following columns:

• Name
• Time (YYYY-MM-DD HH:MM:SS)

Notes

• The system uses webcam index 1 by default (can be changed in mark_attendance.py)
• Face recognition threshold is set to 0.9 (can be adjusted for stricter/looser matching)
• Images are stored in the dataset folder, organized by student name