Recognize American Sign Language letters using deep learning!
This project trains a Convolutional Neural Network (CNN) to classify hand gestures from grayscale images into 29 classes — the English alphabet (A–Z), plus space and nothing. It supports training, webcam inference, and browser deployment via TensorFlow.js.
- 29-class classification: A–Z, space, nothing
- Trained on 87,000+ preprocessed hand gesture images
- Run predictions in real time using your webcam
- Exportable to TensorFlow.js for browser-based use
- Fully modular: dataset → training → model → deployment
Required packages:
tensorflow pandas numpy scikit-learn opencv-python matplotlib joblib
The dataset was sourced from Kaggle: ASL Alphabet Dataset and converted into a single .csv with pixel-normalized grayscale values for training.
A → 0, B → 1, ..., Z → 25, space → 26, nothing → 27
Tip Why It Matters
Good lighting Helps Mediapipe detect the hand more accurately.
Solid background Avoid noisy backgrounds for better hand segmentation.
Show one hand The code is set to detect only 1 hand.
Hold the gesture still Prediction is per frame; stability helps confidence.
Hand centered in camera Ensures the whole hand is captured.
- Dataset by grassknoted on Kaggle
- Help of A.I. tools for brainstorming