Sign Language Detection Model

This project implements Hand sign language detection (letters / numbers) using hand landmarks only (wrist + fingers), currently without face or body pose.

It uses:

• MediaPipe for hand landmark detection
• Classical Machine Learning (KNN)
• No deep learning
• No image-based training

The system works in real time using a webcam.

---

What this project does

• Detects a single hand from a live camera feed
• Extracts 21 hand landmarks
• Converts landmarks into a scale- and position-invariant feature vector
• Trains a classifier on labeled hand poses
• Predicts static signs (e.g. A, B, Y) live

This project supports static signs only.

---

Project structure

├── data_collection.py # Collect labeled hand sign samples
├── train_knn.py # Train KNN classifier on collected data
├── predict_live.py # Real-time sign prediction
├── sign_dataset.csv # Generated dataset (after collection)
├── knn_model.pkl # Trained model (after training)
└── README.md

## Requirements

- Python **3.11**
- Webcam
- Linux / macOS / Windows

Python libraries:
- mediapipe
- opencv-python
- numpy
- pandas
- scikit-learn
- joblib