HUMAN - COMPUTER INTERACTION SYSTEM FOR PEOPLE WITH DISABILITIES

ABSTRACT Ensuring inclusive digital access remains a significant challenge for visually impaired users despite advances in assistive technologies and legislative frameworks. This study investigated the interaction barriers encountered by visually impaired individuals when using mainstream computing systems. A novel, AI-powered web assistant was proposed to enhance accessibility and user autonomy. Guided by a user-centered design methodology and Agile development cycle, requirements were elicited through iterative consultations with visually impaired participants. The resulting system adopted a three-tier architecture: an accessible presentation layer featuring voice and keyboard interfaces, an application layer for service orchestration, and a data/external service layer leveraging AI vision and speech APIs. Core functionalities included hands-free, voice-activated commands; real-time object detection using TensorFlow.js; AI-generated alternative text; and customizable accessibility preferences. Performance evaluation with ten visually impaired users yielded an average object-detection accuracy of 92% and voice-recognition accuracy of 87% in quiet conditions. Task success rates ranged from 78% for error recovery to 95% for homepage navigation, with overall satisfaction averaging 4.2 out of 5. Participants expressed a strong preference for voice-first interaction and reported improved situational awareness and ease of navigation. These results demonstrate that integrating multimodal AI services within an inclusive, user-driven framework can substantially improve HCI for visually impaired individuals.