Bangladeshi Sign Language (Bdsl) Words dataset
Description
In this data repository two zip files: one contains the primary data, and the other contains augmented data. This dataset consists of images related to Bangladeshi Sign Language, focusing on 10 common words used in daily life. These words are categorized into 10 classes: myself, you, they, think, friend, salam (hello), color, surprise, request, and promise. - Primary Data: Contains 1000 images in total, with each class having 100 images. - Augmented Data: Through augmentation techniques, the dataset has been expanded to 6000 images. Augmentation methods include: - Flipping (vertical and horizontal) - Rotation (0-45 degrees) - Adding 3-7% blur in the background - Hue saturation adjustments (hue shift 0-5, saturation 0-50) - Conversion to grayscale This dataset aims to facilitate the recognition and understanding of Bangladeshi Sign Language gestures through machine learning and computer vision applications.
Files
Steps to reproduce
1. Dataset Description: Content: The dataset consists of images related to Bangladeshi Sign Language, focusing on 10 common words used in daily life. Classes: Organized into 10 classes: myself, you, they, think, friend, salam (hello), color, surprise, request, and promise. Original Size: 1000 images (100 per class). Augmented Size: Expanded to 6000 images through various augmentation techniques. 2. Data Collection Methods: Image Acquisition: Images were sourced or captured to represent each sign gesture for the 10 selected words. Categorization: Each image was categorized into one of the 10 classes based on the sign gesture it represents. 3. Data Augmentation Techniques: Flipping: Both vertical and horizontal flipping of images. Rotation: Rotation applied at angles ranging from 0 to 45 degrees. Blur Addition: Background blur introduced at a rate of 37%. Hue Saturation Adjustment: Hue shift ranging from 0 to 5, and saturation adjustment ranging from 0 to 50. Grayscale Conversion: Images were converted to grayscale to diversify the dataset. 4. Deep Learning Technique: Model Architecture: Specify the deep learning architecture used (e.g., Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), etc.). Training: Describe the training process, including batch size, number of epochs, and optimizer used (Adam, SGD). Validation and Testing: Outline how the dataset was split for training, validation, and testing purposes. 5. Evaluation: Accuracy: Report the achieved accuracy, specifically noting it exceeded 95%. Metrics: Include any other relevant metrics used for evaluation (e.g., precision, recall). 6. Application in RealTime Recognition: Implementation: Explain how the trained model was deployed for realtime recognition. Integration: If applicable, mention any hardware or software used for integrating the recognition system. 7. Future Applications: Impact: Discuss the potential applications of this research, particularly in improving communication between normal and deaf & dumb individuals. Future Work: Outline any future research directions or improvements that could enhance the system's performance or usability. 8. Resources Used: Tools and Software: List any specific tools, libraries, or software used for data preprocessing, model development, and evaluation. References https://www.ijournalse.org/index.php/ESJ/article/view/2062