Recogneyes: a raspberry pi-based philippine banknote and coin value recognition system using vision-language model with audio and vibrotactile feedback for the visually impaired/
Sharia Dolce B. Barruga, Roldan Jr. C. Baruel, Federico III D. Chavez, Micko B. Fortes, Richell Mark B. Miguel, and Mercedez D. Pestelos.--
- Manila: Technological University of the Philippines, 2025.
- xv, 204pages: 29cm.
Bachelor's thesis
College Of Engineering.--
Includes bibliographic references and index.
Visual impairment affects millions globally, with many cases remaining untreated despite advancements in modern medicine. As a result, individuals often rely on assistive technologies; however, there is a limited number of tools specifically designed for currency recognition. This study presents RecognEYES, a wearable, Raspberry Pi–based recognition system for Philippine banknotes and coins, integrating a Vision-Language Model with audio and vibrotactile feedback. It utilizes Pi Camera v2 to capture images and process it through Raspberry Pi 4B. The data is sent to a trained GPT-4o — the VLM with the highest mAP upon benchmark. Output is presented via speakers and gloves for audio and vibrotactile feedback, respectively. The system was deployed at RM Massage Clinic and evaluated by 15 visually impaired working adults. Feedback collected via a Likert scale indicated mean scores of 3.76 for usability, 3.68 for comfort, and 4.31 for marketability, suggesting high potential for real-world application. Accuracy testing, validated by a Professional Electronics Engineer (PECE), demonstrated optimal detection distances of 15–20 cm for coins with accuracies of 88.10% and 90.48%, and 10–40 cm for banknotes with 100% accuracy. For optimal performance, it is recommended that the device is used in well-lit or outdoor environments.