Face Recognition Backend API & ReactJs for Front-End

Overview

This project provides a backend for a face recognition application, using FastAPI, Prisma ORM for PostgreSQL, and Milvus for vector database storage and similarity search. DeepFace is used for facial recognition and embedding extraction.

Features

• Employee registration with face embedding storage.
• Visitor registration with face embedding storage.
• Face recognition for employees and visitors.
• Integration with Prisma (PostgreSQL) for structured data storage.
• Integration with Milvus for vector similarity search.
• Preloading of DeepFace model for optimized performance.

Installation

Prerequisites

• Python 3.9+
• PostgreSQL
• Milvus
• Node.js (for Prisma migrations)

Environment Variables

Create a .env file with the following variables:

DATABASE_URL=postgresql://user:password@localhost:5432/database_name
MILVUS_HOST=localhost
MILVUS_PORT=19530

Dependencies

Install the required Python packages:

pip install fastapi uvicorn prisma pymilvus deepface Pillow numpy python-dotenv pytz

API Endpoints

1. Root Endpoint

GET /

Description: Test endpoint to verify the server is running.

Response:

{"message": "Face App Backend"}

2. Employee Registration

POST /register-employee/

Description: Register a new employee with personal details and face embedding.

Form Data:

• name: Employee name (string)
• age: Employee age (integer)
• gender: Employee gender (string)
• photo: Employee photo (file)
• designation: Job designation (string)
• contactNumber: Contact number (string)
• department: Department (string)
• description: Additional description (string)

Response:

{"message": "Employee registered successfully."}

3. Visitor Registration

POST /register-visitor/

Description: Register a visitor with personal details and face embedding.

Form Data:

• name: Visitor name (string)
• age: Visitor age (integer)
• gender: Visitor gender (string)
• photo: Visitor photo (file)
• contact: Contact number (string)
• purposeOfVisit: Purpose of the visit (string)
• description: Additional description (string)

Response:

{"message": "Visitor registered successfully."}

4. Recognize Employee

POST /recognize-employee/

Description: Recognize an employee from a photo.

Form Data:

• photo: Uploaded photo (file)

Response:

If a match is found:

{"name": "John Doe", "similarity": 0.85, "status": "success"}

If no match is found:

{"message": "No match found", "status": "failed"}

5. Recognize Visitor

POST /recognize-visitor/

Description: Recognize a visitor from a photo.

Form Data:

• photo: Uploaded photo (file)

Response:

If a match is found:

{"name": "Jane Doe", "similarity": 0.90, "status": "success"}

If no match is found:

{"message": "No match found", "status": "failed"}

Backend Architecture

1. FastAPI

• Framework for building the backend.
• Provides routing, middleware, and dependency injection.

2. Prisma ORM

• Manages structured data in PostgreSQL.
• Used to store:
  • Employee details
  • Visitor details

3. Milvus

• Vector database for storing face embeddings.
• Used for similarity searches.

4. DeepFace

• Extracts facial embeddings.
• Model used: VGG-Face.

Database Structure

PostgreSQL Tables

Employees Table

Column	Type	Description
id	UUID	Primary key
name	String	Employee name
age	Integer	Employee age
gender	String	Employee gender
photoBase64	Text	Base64-encoded photo
designation	String	Job designation
contactNumber	String	Contact number
department	String	Department
description	String	Additional description

Visitors Table

Column	Type	Description
id	UUID	Primary key
name	String	Visitor name
age	Integer	Visitor age
gender	String	Visitor gender
photoBase64	Text	Base64-encoded photo
contact	String	Contact number
purposeOfVisit	String	Purpose of the visit
description	String	Additional description

Key Functions

1. extract_face_embedding(photo_bytes: bytes)

   • Extracts face embedding from photo bytes using DeepFace.
   • Returns a 4096-dimensional embedding.

2. save_to_milvus(collection_name, name, embedding)

   • Saves normalized embeddings to Milvus.

3. search_in_milvus(collection_name, query_embedding, threshold=0.6)

   • Searches Milvus for a match to the query embedding.
   • Returns the best match if similarity exceeds the threshold.

4. save_to_prisma(name, age, gender, photo_base64, table, additional_data)

   • Saves user details to PostgreSQL using Prisma ORM.

Running the Application

Run Locally

uvicorn main:app --reload

Access the API

• Swagger UI: http://127.0.0.1:8000/docs

Future Enhancements

• Implement rate limiting for API endpoints.
• Add authentication and authorization.
• Use GPU acceleration for faster embedding extraction.
• Integrate more facial recognition models.

Screenshots

Backend : FastAPI

Milvus DB : using Attu for UI

Primsa - Using Postgres (ORM with Prisma Python Client)

FrontEnd using React JS

Landing Page

Main Dashboard

Check In interface : Dashboard For checkIn (Visitor / Employee) : Face recognition

Custom Error-Page for handling unauthorized requests

Face recognition Working

Self Check-In model for visitor/employee

Employee

Visitor

Working Model Representation

Scope for Improvements

1. Rate limiter for requests
2. Authentication
3. Session using redux and redux tools (setting up store and auth-sessions)
4. Form validations
5. Pydantics models for Proper models