nikh-iam / InsightFace-FaceRecognition

Face Recognition and Detection using InsightFace

InsightFace-FaceRecognition

This project implements a real-time Face Recognition System using InsightFace. The system efficiently detects and recognizes faces using GPU acceleration, FAISS for embedding lookup, and adaptive thresholding for improved accuracy.

Features

• Real-time Face Detection & Recognition using InsightFace
• FAISS for fast embedding lookup
• Face Tracking (SORT/DeepSORT) to reduce redundant face processing
• Multi-threading for parallel frame processing
• Adaptive Thresholding for improved recognition accuracy
• CLAHE & Auto Brightness Correction to handle low-light conditions

Installation

1. Clone the Repository

git clone https://github.com/nikh-iam/InsightFace-FaceRecognition.git
cd InsightFace-FaceRecognition

2. Install Dependencies

Ensure you have Python 3.8+ installed, then run:

pip install -r requirements.txt

3. Install InsightFace Model

Download and place the buffalo_s model inside ~/.insightface/models/ or configure it accordingly.

Usage

Run the Face Recognition System

python main.py

Press q to exit the application.

Extract Face Embeddings

python extract_embeddings.py

This script extracts face embeddings from images in the faces_db/ directory and stores them in face_embeddings/.

Project Structure

├── face_recognition.py       # Real-time face recognition script
├── extract_embeddings.py     # Script to extract and store face embeddings
├── face_embeddings/          # Directory for stored face embeddings
├── faces_db/                 # Directory containing face images
├── requirements.txt          # List of dependencies
├── README.md                 # Documentation

How It Works

1. Face Detection: Detects faces in video frames using InsightFace.
2. Face Recognition: Matches detected faces with stored embeddings.
3. Face Tracking: SORT/DeepSORT reduces redundant processing.
4. Performance Optimization:
   • Multi-threading for real-time processing
   • FAISS for fast embedding searches
   • TensorRT for GPU acceleration

Future Improvements

• Support for multiple cameras
• Web-based interface for monitoring
• Automatic dataset updates with real-time learning

License

This project is licensed under the MIT License.