An Ai-based Human Scene Understanding Mechanism For Image Captioning In Blind Navigation And Assistance

Abstract

Visual Impairment Greatly Affects The Navigation And Interaction With The Surrounding Environment. Conventional Assistive Tools Are Generally Unintelligent, While Existing Digital Tools Are Generally Task-specific, Computationally Inefficient, And Require Connectivity, Which Makes Them Less Effective In Real-time Applications. In This Paper, An Integrated AI-based Assistive Vision System Is Proposed, Which Can Improve The Situational Awareness Of The Visually Impaired Through The Use Of Multimodal Perception. The Proposed System Incorporates Object Detection, Face Recognition, And Currency Recognition In An Integrated Manner. Real-time Images Are Captured Using A Camera Module, Which Are Fed Into The Optimized Deep Learning Models For Image Processing. The YOLO Algorithm Is Used For Efficient Obstacle Detection, Allowing The Identification Of Moving Objects In The Surrounding Environment. Face Recognition Is Achieved Using The Grassmann Approach, Allowing The Robustness Of The Algorithm To Pose And Illumination Variations. A Convolutional Neural Network Is Used For Reliable Currency Classification, Even In Cases Of Partial Occlusion. The Processed Information Is Converted Into Context-aware Audio Feedback, Enabling Effective Navigation And Interaction Without Visual Dependency. Offline Functionality Ensures Operational Reliability In Diverse Conditions. The Integrated Architecture Enhances Accuracy, Reduces System Complexity, And Improves User Independence, Representing A Significant Advancement In Intelligent Assistive Technology.

Keywords