face-recognitioniiti

Real-time Face Recognition System | DEMO

A Python-based real-time face recognition system that combines YOLO face detection with ArcFace embeddings for accurate face recognition. The system processes webcam feed in real-time, detecting faces and matching them against a stored database of known faces.

Features

• Real-time face detection using YOLO
• Face recognition using ArcFace embeddings
• Persistent storage of face embeddings
• Interactive interface for adding new faces to the database
• Configurable confidence thresholds for detection and recognition
• Live display of recognition results with bounding boxes and names

Prerequisites

• Python 3.8 or higher
• Webcam or camera device

Installation

1. Clone this repository:

git clone https://github.com/Warlord-K/InterIIT-facerecognition.git
cd InterIIT-facerecognition

1. Install required packages:

pip install -r requirements.txt

1. Download the Arcface model: Click this link and put it in the same folder as the script

Usage

1. Run the main script:

python face_recognition_system.py

1. Controls:

• Press 'q' to quit the application
• Press 'a' to add a new face to the database
  • When prompted, enter the name for the face
  • The current frame will be used to generate embeddings

Configuration

You can modify the following parameters in the FaceRecognitionSystem class initialization:

face_system = FaceRecognitionSystem(
    database_path="face_database.pkl",  # Path to store face embeddings
    confidence_threshold=0.5,           # Minimum confidence for YOLO detection
    similarity_threshold=0.6            # Maximum distance for face recognition match
)

System Architecture

The system consists of three main components:

1. Face Detection (YOLO)

   • Uses YOLO model optimized for face detection
   • Processes each frame to locate faces
   • Applies confidence threshold to filter detections

2. Face Recognition (ArcFace)

   • Generates embeddings for detected faces
   • Compares embeddings with stored database
   • Returns closest match based on similarity threshold

3. Database Management

   • Stores face embeddings with associated names
   • Supports adding new faces during runtime
   • Persists data between sessions using pickle

Performance Considerations

• Processing speed depends on:
  • Hardware capabilities (CPU/GPU)
  • Image resolution
  • Number of faces in the database
• Adjust confidence and similarity thresholds for optimal balance between accuracy and speed

Troubleshooting

Common issues and solutions:

1. No camera access:

   • Check if camera is properly connected
   • Verify camera permissions
   • Try changing camera index in cv2.VideoCapture(0)

2. Poor recognition accuracy:

   • Adjust similarity_threshold for stricter/looser matching
   • Ensure good lighting conditions
   • Add multiple embeddings per person under different conditions

3. Slow performance:

   • Reduce frame resolution
   • Increase confidence threshold to process fewer detections
   • Consider using GPU acceleration if available