ibrahim-shedo/kenya-sing-language-translator-using-LSTM-model-deep-learning-

# 🇰🇪 Kenyan Sign Language Recognition (KSL) - Open Research Project This project is an open-source effort to build a real-time Sign Language Recognition system focused on **Kenyan Sign Language (KSL)**. It's designed to be an accessible base for researchers, students, or developers who wish to expand or adapt the model further. There are **two main code paths** included in this project: - **🧪 Tutorial Code:** Used during data collection and debugging. This version helps with **retaking actions** that didn’t perform well and is useful for experimentation or when dealing with noisy/incorrect samples. - **⚙️ Pure Final Code:** A clean, production-ready version with minimal clutter. Use this if you want a **stable baseline** to improve, modify, or integrate into other systems. --- ## 📌 Project Goals - Collect and process KSL signs using webcam and MediaPipe Holistic. - Train an LSTM deep learning model to recognize gestures in real time. - Construct **rule-based natural language sentences** from detected signs. - Open the project to encourage **collaboration and extension** of the KSL dataset. --- ## 📁 Project Structure ├── MP_Data/ # Collected and saved keypoint data for each action │ └── [action]/[sequence]/[frame].npy ├── model/ # Trained model file │ └── ksl_model.h5 ├── main.py # Main real-time detection and sentence construction script ├── utils.py # Helper functions: landmark drawing, keypoint extraction ├── README.md # This file yaml Copy Edit --- ## 🧪 Actions (Signs) Currently, the model is trained on **30 signs**, including: ['me', 'you', 'friend', 'name', 'mine', 'who', 'how', 'please', 'help-me', 'wait', 'now', 'home', 'where', 'give-me', 'thank-you', 'polite', 'hello', 'good', 'mother', 'father', 'uncle', 'cousing', 'brother', 'sister', 'doughter', 'parent', 'relative', 'yes', 'no', 'sorry'] yaml Copy Edit --- ## 🛠️ Technologies Used - **Python** - **OpenCV** – for video capture and rendering - **MediaPipe Holistic** – for hand, pose, an …