Project Overview
A real-time multiple face detection and recognition system built in Python using
OpenCV and the face_recognition library
(built on top of dlib's deep learning models). The system can detect and identify
multiple faces simultaneously from a live webcam feed or video stream.
How It Works
- Uses HOG (Histogram of Oriented Gradients) or CNN-based face detection to locate faces in each video frame
- Extracts 128-dimensional facial encodings using a deep neural network
- Compares live encodings against a pre-built database of known faces using Euclidean distance
- Draws labeled bounding boxes around each identified person in real time
- Handles multiple simultaneous faces with smooth frame-by-frame processing
Technical Highlights
- Built on dlib's state-of-the-art face recognition neural network achieving 99.38% accuracy on the LFW benchmark
- OpenCV handles video capture and image processing pipeline
- Modular code structure: easily extend with new faces by adding images to the known faces directory
What I Learned
This project gave me hands-on experience with computer vision fundamentals, facial landmark detection, and deep learning embeddings. Understanding how face encodings work — mapping a face to a point in 128-dimensional space — was a fascinating introduction to the practical applications of deep learning.