ASL Recognition Model

This project provides an open-source implementation of an American Sign Language (ASL) Recognition Model. The model leverages machine learning and computer vision techniques to recognize ASL hand signs from images.

Features

Hand Landmark Detection: Utilizes MediaPipe to accurately detect 21 hand landmarks in images.
Feature Extraction: Calculates angles between all pairs of landmarks to form a 420-dimensional feature vector.
- Vector Calculation: Computes vectors between each pair of landmarks.
- Angle Computation: Uses the arccosine of normalized vector components to derive angles.
Model Input: The extracted angles serve as input features for the Random Forest model, which classifies the ASL sign.

Technical Stack

Python: Core programming language.
OpenCV: For image processing and manipulation.
MediaPipe: For detecting hand landmarks.
Scikit-learn: Provides the Random Forest model for classification.
Streamlit: Facilitates an interactive user interface for real-time recognition.

Supported Alphabets

The model currently works for the following ASL alphabets:

A, B, C, E, F, G, H, I, J, K, L, O, Q, R, S, W, Y

The model does not support or may not work correctly for:

D, M, N, P, T, U, V, X, Z

Usage

Upload an image of an ASL sign through the Streamlit interface.
The model processes the image and provides the top 5 predictions along with visualizations of detected hand landmarks.

Contribution

We welcome contributions to improve the model's accuracy and expand its alphabet coverage. Feel free to fork the repository, submit issues, or create pull requests.

License

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

Acknowledgments

Thanks to the contributors of MediaPipe and Scikit-learn for their powerful libraries that made this project possible.