🤟 Sign Language & Gesture Recognition

A real-time hand gesture recognition system using MediaPipe and TensorFlow/Keras. The model detects hand landmarks via webcam and classifies gestures into letters (A–L) and numbers (1–7) in real time.

---

📁 Project Structure

├── hand_tracking_dataset.py   # Capture gesture samples using webcam
├── train_gesture_model.py     # Train the gesture classification model
├── real_time_detector.py      # Run real-time gesture detection
├── label_map.json             # Gesture-to-label mapping
├── gesture_data/              # Captured gesture samples (not pushed to GitHub)
└── gesture_model.h5           # Trained model (not pushed to GitHub)

---

🧠 How It Works

1. Data Collection — MediaPipe detects 21 hand landmarks from webcam feed. Landmarks are normalized relative to the wrist and saved as .npy files.
2. Model Training — A simple Dense Neural Network is trained on the collected landmarks to classify gestures.
3. Real-Time Detection — The trained model runs live on webcam input and displays the predicted gesture on screen.

---

🖐️ Supported Gestures

Letters	Numbers
A, B, C, D, E, F, G, H, I, K, L	1, 2, 3, 4, 5, 6, 7

---

⚙️ Installation

# Clone the repository
git clone https://github.com/your-username/gesture-recognition.git
cd gesture-recognition

# Install dependencies
pip install opencv-python mediapipe tensorflow scikit-learn numpy

---

🚀 Usage

Step 1: Collect Gesture Data

python hand_tracking_dataset.py

• Enter a gesture label when prompted (e.g. A, 1, etc.)
• Perform the gesture in front of your webcam
• 200 samples will be captured per gesture
• Press q to quit anytime

Step 2: Train the Model

python train_gesture_model.py

• Trains a neural network on your captured data
• Saves gesture_model.h5 and label_map.json

Step 3: Run Real-Time Detection

python real_time_detector.py

• Opens your webcam and detects gestures in real time
• Press q to quit

---

🏗️ Model Architecture

Input (63 features: 21 landmarks × 3 coordinates)
→ Dense(128, ReLU) → Dropout(0.2)
→ Dense(64, ReLU)  → Dropout(0.2)
→ Dense(num_classes, Softmax)

• Optimizer: Adam
• Loss: Sparse Categorical Crossentropy
• Epochs: 50 | Batch Size: 16

---

📋 Requirements

• Python 3.8+
• Webcam
• opencv-python
• mediapipe
• tensorflow
• scikit-learn
• numpy

---

🙏 Acknowledgements

• MediaPipe by Google for hand landmark detection
• TensorFlow/Keras for model training