Amharic Sign Language Transcription Dataset
Led the end-to-end creation and annotation of a specialized dataset for Amharic Sign Language recognition, addressing the scarcity of resources for Ethiopic digital accessibility. Key Contributions: Data Collection & Curation: Systematically collected and organized raw image data for distinct Amharic characters (e.g., 'ሀ', 'ለ'), ensuring diverse lighting and background conditions. Automated Annotation Pipeline: Engineered a custom Python-based labeling workflow using MediaPipe to automatically extract and record 21-point 3D hand landmarks for every image, creating a rich feature set for model training. Quality Assurance: Performed manual verification of landmark accuracy and class consistency to maintain a high-quality ground truth, resulting in a robust dataset that powers a real-time transcription engine. Impact: Enabling real-time communication tools for the deaf community in Ethiopia by bridging the gap between traditional sign language and digital text.