Published

Face Recognition System v1.0

March 2, 2026
https://aminilay.github.io/face-recognition-system-official-website/
Face Recognition System v1.0
Position / Role

Lead Architect

Deployment

Published

Timeline

Mar 2026

Primary Artifact

This is a real-time biometric identification desktop app that recognises faces from a live video feed using 128-D face embeddings. I built it for security, attendance tracking and computer vision exploration.

Overview

It's a modular, optimised Python application that uses deep learning to identify faces in real time. A dark-mode GUI sits on top of a multi-threaded backend engine built to process high-FPS video feeds with low latency.

How it works

  • Real-time detection: Histogram of Oriented Gradients (HOG) detection, multi-frame skipping and performance-focused threading keep video processing fast.
  • Dynamic confidence logic: a colour-coded overlay (🟒 High, 🟑 Medium, πŸ”΄ Low) shows confidence directly on the live feed.
  • Identity management: tools to add or remove known faces via webcam capture or file upload.
  • Persistent logging: SQLite keeps a filterable history of every identification event, so you can see exactly who was seen and when.
  • Desktop GUI: a responsive dark-mode interface built with Tkinter.
Under the hoodEach face is reduced to a 128-D embedding, which is what the system compares to identify people in the live feed.

Tech stack

Python 3.10+ OpenCV Dlib / Face_Recognition SQLite3 Tkinter Pandas NumPy

Case Study

Face embedding

128-D

Vector compared to identify each face

  1. Problem

    Security, attendance tracking and computer vision exploration all need a way to identify faces from a live video feed quickly and keep a record of who was seen. I wanted a desktop app that does this in real time with an easy-to-use interface.

  2. Approach

    • Identified faces using deep learning and 128-D face embeddings (Dlib / Face_Recognition)
    • Used HOG detection, multi-frame skipping and multi-threading with OpenCV to keep video processing fast
    • Overlaid colour-coded confidence levels (High / Medium / Low) on the live feed
    • Added tools to add or remove known faces via webcam capture or file upload
    • Logged every identification event to SQLite as a filterable history
    • Built a dark-mode desktop GUI with Tkinter