Multimodal AI Assistant for Visually Impaired — A Literature Survey

Visual impairment continues to restrict independent participation in daily activities such as object recognition, text reading, and safe navigation. Over the past decade, advances in computer vision, natural language processing, and speech technology have given rise to a broad family of assistive systems intended to compensate for the loss of visual perception through audio channels. This paper surveys recent research on AI-driven assistive technologies for visually impaired individuals, organizing the literature into five thematic clusters: obstacle detection and navigation, object detection with audio feedback, vision-to-language captioning, optical character recognition combined with conversational interfaces, and wearable or spatial-audio navigation platforms. For each cluster, representative works are examined in terms of underlying methodology, reported contributions, and practical limitations. A comparative analysis is then presented to highlight recurring shortcomings across the surveyed body of work, in particular the fragmentation of assistive functionality into narrow, single-purpose systems, the computational burden imposed by large vision-language models, and the limited support for real-time, hands-free, bidirectional interaction. The survey concludes by identifying open research gaps and motivating the need for an integrated multimodal framework that unifies scene captioning, object and text recognition, and voice-driven interaction within a single accessible platform.