Face Recognition System

Table of Contents

1. Setup
   • Create Environment
   • Test Environment
   • Development Environment
   • Production Environment
2. Technology
   • Face Detection
   • Face Recognition
   • Face Tracking
   • Triton Inference Server
   • Vector Store
   • Redis Streams

Setup

Create Environment

1. Create a new conda environment:

   conda create -n face_recognition python=3.9

2. Activate the environment:

   conda activate face_recognition

3. Install the required packages:

   pip install -r requirements.txt

Test Environment

1. Test Detection

   pytest --capture=no tests/test_detect_image.py

2. Test Tracking

   pytest --capture=no tests/test_tracking.py

3. Test Recognition

   pytest --capture=no tests/test_recognize_image.py

Development Environment

1. Create a Folder for Each Person

   • Create a folder with the folder name being the name of the person:

     datasets/
     ├── backup
     ├── data
     ├── face_features
     └── new_persons
         ├── name-person1
         └── name-person2
     

2. Add the Person's Photos to the Folder

   • Add the person's photo in the corresponding folder:

     datasets/
     ├── backup
     ├── data
     ├── face_features
     └── new_persons
         ├── name-person1
         │   └── image1.jpg
         │   └── image2.jpg
         └── name-person2
             └── image1.jpg
             └── image2.jpg
     

3. Run Triton Server, Redis, and Qdrant

   docker compose up

4. Train New Persons

   python train.py

5. Run Recognition App

   python recognize.py

6. Run Tracking App

   python tracking.py

Production Environment

1. Build the Docker Image

   docker compose --profile=prod build

2. Run the Application

   docker compose --profile=prod up

Technology

Face Detection

• SCRFD (Single-Shot Scale-Aware Face Detector): Designed for real-time face detection across various scales, effective in detecting faces at different resolutions within the same image. SCRFD models require input image shapes divisible by 32, such as 640x640 or 1024x768.

Face Recognition

• ArcFace: A state-of-the-art face recognition algorithm that focuses on learning highly discriminative features for face verification and identification. It is robust to variations in lighting, pose, and facial expressions.

  
  ArcFace

Face Tracking

• ByteTrack: A simple, fast, and strong multi-object tracker.

  

Triton Inference Server

• Triton Inference Server: An open-source software platform designed for deploying and managing AI models at scale. It supports various deep learning and machine learning frameworks, including TensorFlow, PyTorch, ONNX, and OpenVINO. Triton can be deployed on various hardware platforms, including CPUs, GPUs, and FPGAs.

  

Vector Store

• Qdrant: A vector similarity search engine and database offering a production-ready service with a user-friendly API for storing, searching, and managing vectors along with additional payload data. Qdrant supports extended filtering and is suitable for various applications such as neural-network or semantic-based matching and faceted search. It is written in Rust for speed and reliability under high loads.

  

Redis Streams

• Redis Streams: Used for streaming frames and bounding box tracking.