🤟 TradutorDeLibras

Real-time Brazilian Sign Language (LIBRAS) recognition and translation system

---

📌 About

TradutorDeLibras is an AI-powered system that detects and translates Brazilian Sign Language (LIBRAS) gestures into text in real time. Built as an applied research project at LAPISCO AI Research Lab (IFCE), it combines hand landmark detection with deep learning gesture classification to make communication more accessible for the deaf and hard-of-hearing community.

---

✨ Features

• 📸 Automated Data Collection — script-based webcam capture to build a custom LIBRAS dataset
• 🤚 Hand Landmark Extraction — MediaPipe identifies 21 key hand points per frame for precise gesture mapping
• 🧠 Deep Learning Classification — YOLOv11 model trained on LIBRAS gesture data for robust real-time recognition
• 🎥 Real-time Translation — live webcam feed with gesture detection and text output
• 🖼️ Static Image Testing — support for gesture recognition on individual images
• ⚡ FastAPI Backend — lightweight REST API to serve model inference

---

🏗️ Architecture

Camera Input / Static Image
         │
         ▼
  MediaPipe (Hand Landmark Detection — 21 keypoints)
         │
         ▼
  YOLOv11 (Gesture Classification)
         │
         ▼
  FastAPI (Inference API)
         │
         ▼
  Text Output (Translated Sign)

---

🛠️ Tech Stack

Layer	Technology
Hand Detection	MediaPipe
Gesture Classification	YOLOv11 (Ultralytics)
Image Processing	OpenCV
Backend / API	FastAPI (Python)
Numerical Processing	NumPy

---

📂 Project Structure

├── data/
├── images/
├── collect_imgs.py       # Automated webcam data collection
├── create_dataset.py     # Converts images to hand landmarks
├── training.py           # Model training pipeline
├── testing.py            # Real-time webcam testing
├── testing_imgs.py       # Static image testing
├── model.p               # Trained model
└── data.pickle           # Processed landmark dataset

---

⚙️ Getting Started

Prerequisites

pip install opencv-python mediapipe ultralytics fastapi uvicorn numpy

1. Data Collection (optional — skip if using existing dataset)

python collect_imgs.py

2. Create Dataset

python create_dataset.py

3. Train the Model

python training.py

4. Run Real-time Translation

python testing.py

5. Run the API

uvicorn main:app --reload

---

🤝 Contributing

Contributions are welcome! Feel free to open an Issue or submit a Pull Request.

---

📄 License

This project is licensed under the MIT License.

---

Developed by Derick Bessa @ LAPISCO AI Research Lab