Skip to content

README.md

Latest commit

 

History

History
65 lines (49 loc) · 2.56 KB

README.md

File metadata and controls

65 lines (49 loc) · 2.56 KB
Raw

Face Recognition By Training (Static YOLO)

This subsystem implements a robust, supervised learning pipeline for face recognition. It uses YOLOv8 for high-accuracy face detection and a custom-trained YOLOv8 classifier to identify specific individuals.

🛠️ Features

  • Deterministic Classification: High precision for pre-enrolled individuals.
  • Robust Detection: Powered by yolov8m-face.pt, capable of spotting faces in complex backgrounds.
  • Data Augmentation: Automatically enhances training data with rotations, hue tweaks, and scaling.
  • Hardware Triggers: Supports physical output (e.g., door locks) via communication.py and Raspberry Pi GPIO.

🔄 Workflow

1. Data Collection

Collect face crops for specific individuals. This will save images to the CollectedDataset/ directory.

python Register.py --persons <person1> <person2> --test-cases <case_name>

Example: python Register.py --persons alice bob --test-cases office_staff

2. Model Training

Train the classification model using the collected dataset. This process splits data into 80/10/10 (Train/Val/Test) and runs the fine-tuning.

python Register.py --train --test-cases <case_name>

3. Inference (Recognition)

Run the real-time recognition loop. If a person is tracked consistently for over 50 frames, it can trigger a physical signal.

python Register.py --find --test-cases <case_name>

📂 Script Definitions

  • Register.py: The primary CLI tool that wraps collection, training, and finding logic.
  • FaceDataCollector.py: Handles frame capture and face cropping during registration.
  • FaceTraining.py: Orchestrates the YOLOv8 classification training loop.
  • FaceTesting.py: The core inference engine that combines detection and classification.
  • communication.py: Manages serial/GPIO communication for external hardware triggers.
  • Augmentation.py: utilities for diversifying the training dataset.

📦 Models Used

  • Detection: yolov8m-face.pt (Medium model for reliable face localization).
  • Classification: yolov8n-cls.pt (Nano model for efficient real-time identification).

📁 Directory Structure

FaceRecognitionByTraining/
├── CollectedDataset/              # Raw face crops per person
├── TrainingDataSetForClassification/ # Organized Train/Val/Test sets
├── runs/                           # YOLO training artifacts and weights
├── Register.py                    # Main CLI entry point
└── ...

Return to Root README