🖐️ Hand Gesture Classification

Real-time hand gesture and motion recognition using MediaPipe and TensorFlow Lite.

🚀 Demo

Gesture Classification

History Classification

🎯 Features

• 8 Hand Gestures: 👍 👎 ☝️ ✌️ ✊ ✋ 🤟 👌
• 4 Motion Types: Still • Clockwise • Anticlockwise • Move
• Dual Hand Detection: Recognize 2 hands simultaneously
• High Performance: TF Lite models, ~30ms inference per frame
• Real-time Video Processing: 1280×720 @ 30 FPS

📦 Installation

Windows

pip install -r requirements_win.txt

WSL/Linux

pip install -r requirements_wsl.txt

🚀 Quick Start

1️⃣ Collect Training Data

python collect_data.py

• Press S to start/stop recording
• Press Q to quit
• Choose mode: Gestures or Motion sequences

2️⃣ Run Real-time Classification

python app.py

• Point camera at your hand
• Press Q to exit

📂 Project Structure

├── app.py                  # Real-time gesture classifier
├── collect_data.py         # Training data collection tool
├── src/
│   ├── config.py          # Configuration & labels
│   ├── classifier.py      # TF Lite model wrapper
│   ├── processor.py       # Landmark normalization
│   └── visualizer.py      # Visualization utilities
├── models/                # Pre-trained TF Lite models
├── data/                  # Training datasets
└── notebooks/             # Training scripts

🔧 Key Components

Module	Purpose
MediaPipe	Hand landmark detection
TF Lite	Fast gesture/motion inference
OpenCV	Video capture & rendering
NumPy	Landmark processing

📊 Model Architecture

• Gesture Model: Classifies static hand poses
• Motion Model: Analyzes 16-frame sequences for movement direction

⚙️ Configuration

Edit src/config.py to customize:

• Max hands: MAX_HANDS = 2
• Input resolution: IMAGE_SHAPE = (1280, 720)
• Model paths and labels

📝 Notes

• Requires webcam input
• Optimal lighting recommended
• Models run on CPU (GPU optional)

👤 Author

Made with 💻 and ☕ by @udham2511

🙏 Inspiration

Inspired by @Kazuhito00 and their innovative hand gesture recognition projects.

📜 License

This project is open-source and available under the MIT License.