Real-Time
Description
This dataset contains RGB images of hand signs collected from multiple volunteers under varying lighting conditions and backgrounds using different smartphone cameras. Images were captured in both daylight and dim lighting at multiple angles, using default camera settings with automatic exposure. The dataset is divided into two sections: one for training a detection model (with annotated bounding boxes) and another for testing a recognition model. It is designed to support research in sign language detection and recognition, especially for Bangla hand signs.
Files
Steps to reproduce
1. Participant Image Collection Invite multiple volunteers to pose for hand sign images under different lighting conditions (daylight and dim light), backgrounds, and angles. 2. Camera Setup Use various smartphones with default settings and automatic exposure enabled. Avoid filters or enhancements. 3. Image Capturing Capture full-frame RGB images covering all target hand signs. Ensure diversity in lighting, angle, and background. 4. Data Organization Divide the dataset into two parts: • Training (Detection): For model training with annotated bounding boxes. • Testing (Recognition): For model evaluation. 5. Annotation Annotate hand regions using bounding boxes and save in a standard format (e.g., .csv, .json, or .xml). 6. Upload to Mendeley Data • Log in to data.mendeley.com • Create a new dataset, upload image files and annotations • Provide metadata (title, authors, description, license) • Publish and obtain DOI
Institutions
- Jahangirnagar University