🖐️ Hand Sign Recognition using MediaPipe and TensorFlow

A real-time hand sign recognition system that detects static hand gestures using MediaPipe Hands and classifies them using a TensorFlow Deep Neural Network (DNN). The project recognizes hand signs such as A, B, and C from a webcam feed.

---

📌 Features

• Real-time hand detection using MediaPipe
• Hand landmark extraction (21 landmarks, 63 features)
• Deep Neural Network for hand sign classification
• Live webcam prediction
• Confidence score visualization
• Easy to extend for additional hand signs

---

📂 Project Structure

WasteDetection/
│
├── Data/
│   ├── A/
│   ├── B/
│   └── C/
│
├── MP_Data/
│   ├── A/
│   ├── B/
│   └── C/
│
├── Logs/
│
├── function.py
├── datacollection.py
├── data.py
├── trainmodel.py
├── app.py
├── model.keras
├── requirements.txt
└── README.md

---

🛠️ Technologies Used

• Python 3.10+
• TensorFlow / Keras
• MediaPipe
• OpenCV
• NumPy
• Scikit-learn

---

⚙️ Installation

1. Clone the repository

git clone https://github.com/yourusername/Hand-Sign-Recognition.git

cd Hand-Sign-Recognition

---

2. Create a virtual environment

Windows

python -m venv .venv

Activate

.venv\Scripts\activate

---

3. Install dependencies

pip install -r requirements.txt

---

📊 Dataset

The dataset consists of static hand images.

Classes:

• A
• B
• C

Each class contains approximately 30 images.

The images are captured using the webcam and stored in the Data folder.

---

🚀 Project Workflow

Step 1: Collect Images

Run

python datacollection.py

This captures hand images for each gesture.

---

Step 2: Extract Hand Keypoints

Run

python data.py

This extracts

• 21 hand landmarks
• x, y, z coordinates

Total Features:

21 × 3 = 63

The extracted features are stored as .npy files inside MP_Data.

---

Step 3: Train the Model

Run

python trainmodel.py

This

• Loads all keypoints
• Splits the dataset
• Trains a Deep Neural Network
• Saves the trained model as

model.keras

---

Step 4: Run Real-Time Recognition

Run

python app.py

The webcam opens and predicts the detected hand sign in real time.

Press

Q

to quit.

---

🧠 Model Architecture

Input Layer
      │
      ▼
Dense (128)
      │
Dropout (0.3)
      │
Dense (64)
      │
Dropout (0.3)
      │
Dense (32)
      │
Output Layer
(Softmax)

Input Features:

63

Output Classes:

3

---

📷 Hand Landmark Detection

MediaPipe detects 21 hand landmarks.

Each landmark contains

• x
• y
• z

Total features:

63

These features are used for training instead of raw images.

---

📈 Future Improvements

• Add all 26 alphabet signs
• Add digit recognition (0–9)
• Improve dataset size
• Support two-hand gestures
• Build sentence recognition
• Add speech output
• Deploy as a web application using Streamlit
• Deploy using Flask or FastAPI

---

📦 Requirements

Example

tensorflow==2.15.0
mediapipe==0.10.14
opencv-python
numpy
scikit-learn
matplotlib

---

👩‍💻 Author

Ruchika Bambal

B.Tech Student

Interested in AI, Machine Learning, Computer Vision, and Web Development.

---

⭐ Acknowledgements

• Google MediaPipe
• TensorFlow
• OpenCV
• Scikit-learn

---

📜 License

This project is created for educational and learning purposes.

Feel free to use and modify it.