A Visual-Language Fusion Positioning Assistant System for Non-Communicative Individuals

This paper proposes a vision–language fusion–based, semantics-driven AI location assistance system for communication-impaired subjects such as dementia patients, individuals with developmental disabilities, and companion animals who cannot verbally describe their location. The proposed system integrates object detection from images, scene text recognition (OCR), Korean natural language processing, and geocoding combined with GPS data to estimate locations and provide them in human-readable expressions. Experiments using 13,000 Korean signboard and signage images demonstrate an average OCR F1-score of 0.925, a mean location error of 11.7 m, and a processing time of 0.78 s per image, confirming real-time applicability and significant improvements over conventional OCR+GPS-based approaches