Sign Language Detection with CNN

This repository contains a project aimed at recognizing sign language alphabets using a custom Convolutional Neural Network (CNN) model. The project includes the creation of a manual dataset, model training, and a small interactive user interface using OpenCV.

Features

• Custom Dataset: Manually created dataset with 1800 black-and-white images per alphabet (A-Z).
• Custom CNN Model: Designed and trained using Kaggle GPU resources.
• Interactive UI: Built using OpenCV, allowing users to make sign gestures and set HSV values for detection.
• Real-Time Prediction: The model processes the captured frame and predicts the corresponding alphabet.

Motivation

This project was developed as a hands-on learning experience while studying Convolutional Neural Networks (CNNs) in deep learning. It provided practical insights into data preparation, model building, and integration with OpenCV.

Requirements

• Python 3.x
• OpenCV
• TensorFlow/Keras
• Kaggle (for training)

Installation

1. Clone the repository:

 git clone https://github.com/yourusername/Sign-Language-Recognition-CNN.git
 cd Sign-Language-Recognition-CNN

2. Install the required libraries:

     pip install -r requirements.txt

Dataset

• The dataset was manually created with 1800 black-and-white images per alphabet. Each image captures hand gestures representing alphabets A-Z.

Model Architecture

• The CNN model was designed from scratch, optimized for this specific task. It includes layers for convolution, pooling, and dense connections to accurately classify the images.

Training

• The model was trained on Kaggle using GPU acceleration. Training details:

• Batch size: 32
• Optimizer: Adam
• Loss Function: Categorical Crossentropy

How to Use

1. Run the UI application

python recognise.py

2. Adjust HSV values to isolate hand gestures in real-time.
3. Perform a gesture, and the model will predict the corresponding alphabet.

Limitations

• This project was built as a learning exercise and is not intended for production use. It can serve as a foundation for more advanced sign language detection systems.

Future Work

• Extend the dataset to include numbers and common words.
• Implement transfer learning with pre-trained models for improved accuracy.
• Enhance UI for better user experience.

Acknowledgements

• Kaggle: For providing GPU resources.
• OpenCV: For enabling real-time image processing.
• Deep learning tutorials and resources that guided the project development.

License

• This project is licensed under the MIT License.

Contact

• For any queries or collaborations, feel free to contact me at [email protected]