Sign Language Recognition (Hand Tracking + ML)

What is this?

This is a small project I built to recognize hand signs (like letters) using the webcam.

The idea is simple:

• detect the hand
• extract its shape (landmarks)
• train a model
• predict the sign in real time

---

What I used

• Python
• OpenCV → for camera and image processing
• MediaPipe → to detect hand landmarks
• Scikit-learn → Random Forest model

---

How it works

1. Collect data I capture images of different hand signs using the webcam.

2. Feature extraction MediaPipe gives me hand landmarks (basically coordinates of fingers and joints).

3. Dataset I store these coordinates with labels (A, B, etc.).

4. Training I train a Random Forest classifier on this data.

5. Prediction The webcam runs in real time and predicts the hand sign.

---

Files

• collect_img.py → collect images from webcam
• create_dataset.py → convert images to landmark data
• train_classifier.py → train the model
• inference_classifier.py → run real-time prediction
• data/ → saved images
• data.pickle → processed dataset
• model.p → trained model

---

Notes

• Accuracy depends a lot on the dataset (lighting, angles, etc.)
• Works well for simple signs, but not full words