This project utilizes computer vision and machine learning to detect and classify sign language gestures in real-time. The system leverages MediaPipe for hand landmark detection and scikit-learn for classification.

This application recognizes hand gestures corresponding to the 26 letters of the American Sign Language (ASL) alphabet using a webcam. The system extracts hand landmark features and employs a Random Forest classifier to predict the corresponding letter.

• 🎥 Real-time hand gesture recognition
• 🔠 Support for all 26 ASL alphabet letters
• 📈 99.6% classification accuracy
• 👀 Visual feedback with landmark visualization
• 🖥️ Easy-to-use interface

• Python 3.8 or higher
• Webcam
• Required Python packages (listed in requirements.txt)

1. Download the zip file

   cd sign-language-detector-python

2. Install dependencies:

   pip install -r requirements.txt

sign-language-detection/
├── collect_imgs.py           # Collect hand gesture images
├── create_dataset.py         # Process images and extract features
├── train_classifier.py       # Train the Random Forest model
├── inference_classifier.py   # Real-time sign language detection
├── requirements.txt          # Required dependencies
├── data/                     # Collected image data
│   ├── 0/                    # Images for specific gestures
│   ├── 1/
│   ├── 2/
│   └── ...
└── models/                   # Trained models
    └── random_forest_model.pkl  # Trained classifier

If you want to collect your own dataset (skip if using pre-existing data):

python collect_imgs.py

Follow the on-screen prompts to capture images for each gesture. Press 'q' to stop collecting images for a gesture.

Process the collected images and extract hand landmark features:

python create_dataset.py

This will generate a structured dataset for training.

Train the Random Forest classifier:

python train_classifier.py

The trained model will be saved in the models/ directory.

Run real-time sign language detection:

python inference_classifier.py

Make hand gestures in front of your webcam and see the detected letters displayed in real time.

• Hand Detection: MediaPipe extracts 21 key landmarks from each hand.
• Feature Extraction: Each landmark provides x and y coordinates, yielding 42 features per gesture.
• Model: Random Forest classifier with optimized hyperparameters.
• Preprocessing: Landmark coordinates are normalized for position-invariant features.

• ✅ 99.6% accuracy on the test dataset
• 🚀 Real-time inference at ~30 FPS (depends on hardware)
• 🖐️ Robust to variations in hand size and positioning

All dependencies are listed in requirements.txt:

opencv-python==4.7.0.68
mediapipe==0.9.0.1
scikit-learn==1.2.0
numpy>=1.20.0
matplotlib>=3.5.0

• Hand landmark detection powered by MediaPipe
• Classification powered by scikit-learn