“Demata 2.0”: An On-Device AI Assistive Technology for the Visually Impaired Integrating YOLOv10 and OCR

Abadi, Reza Febri
Pratama, Toni Yudha
Asmiati, Neti
Devi, Ade Anggraini Kartika
Yuwono, Joko
Dwi Setia Permana
Bahrudin, Febrian Alwan

Advance Sustainable Science, Engineering and Technology (ASSET)
Vol. 7 No. 4 (2025): August-October

Accessibility to printed materials and independent recognition of the environment remain key challenges for students with visual impairments. To address this issue, this study introduces Demata 2.0, a fully offline on device multimodal AI system. The system integrates Google ML Kit for Optical Character Recognition (OCR) and the YOLOv10 model via TensorFlow Lite for object detection. A mathematical distance algorithm in the RGB color space enables color identification. Evaluation showed that object detection achieved a mean average precision of 31.83%, with an average processing speed of 2–3 FPS. For OCR, the system recorded a Character Error Rate (CER) of 4.81% and a Word Error Rate (WER) of 10.71% on printed documents. The RGB algorithm also determined the closest possible color effectively. Overall, Demata 2.0 advances assistive technology by providing an efficient and practical blueprint for AI integration.