JoWUA

Abstract

Visually impaired people have a significant difficulty in moving, maneuvering, detecting hazards, and taking up visual information on their own. While a large number of assistive technologies have been created, they are mostly devoted to either object identification or text reading, and are not equipped with context-aware decision making. In this paper, VocalEyes, a real-time assistive vision system and a context-aware approach that convert the visual information into intelligent audio information for visually impaired users is introduced. The Plan designed to combine YOLOv8 object detection, distance and direction estimation, priority alert, intelligent silence, optical character recognition (OCR), and offline speech synthesis, all in one architecture. A prioritized decision module sorts the detected objects based on the relevance of the hazard, thus communicating the more important ones in a timely manner, while reducing unnecessary notifications. Furthermore, OCR only operates when there are no high priority hazards and reduces the distraction to the user during navigation. Experimental results show high accuracy (91%), precision (89%), recall (71%), F1-score (79%) and mAP@0.5 (78%) with low response time (less than one second) and a high frame rate (15–25 FPS) in real time. The smart silence feature also optimizes the usability, as it minimizes repetitive announcements and cognitive overloads. The outcomes highlight that VocalEyes is an effective and fully offline assistive solution, which is cost effective, modular and supports the independence of users, the safety of their navigation in the environment and the awareness of their surroundings.