Word-level Afan Oromo Sign Language Recognition Using Deep Learning Approach

S ign language is a primary communication mode for the hearing-impaired community, yet barriers persist due to limited sign language proficiency among the hearing population and a scarcity of effective translation tools. This work addresses the critical need for improved communication accessibility by developing a real-time Afan Oromo sign language recognition. A primary challenge lies in the absence of comprehensive research on Afan Oromo sign language recognition and translation. To bridge this gap, this study proposes a novel approach utilizing the YOLOv10 model, enhanced for sign language recognition and translation. By leveraging a diverse dataset of 70 common sign language words, we perform data pre-processing steps such as frame extraction, resizing, cropping, flipping, normalization and data splitting to optimize model performance. The core contribution of this research is the development of a robust sign language recognition model capable of accurately translating Afan Oromo signs into text. We achieved impressive results with a Total Average Precision of 94.12%, Recall of 95.01%, and mAP@50 of 90.03% on the YOLOv10 model. This would enable accessible translation tools to be developed for the Afaan Oromo sign language community which could contribute towards improving communication and in- clusivity in their use of signing as a mode of expression.