Machine Vision-Powered Assistive System For Visually Impaired

This paper presents a novel machine vision-powered assistive system to empower visually impaired individuals with real-time obstacle detection and avoidance capabilities. Using the YOLOv8 model for robust obstacle detection and MiDas for depth estimation, the proposed solution provides precise context-aware auditory guidance. The system operates on a laptop equipped with an NVIDIA GeForce RTX GPU, utilizing the laptop’s built-in camera for real-time video processing and a custom-designed user interface for delivering adaptive voice instructions to move Left, Right, Straight, and Stop accordingly. To address challenges in detecting critical structural elements, such as ascending and descending stairs and guardrails, YOLOv8 was fine-tuned on a custom dataset, resulting in enhanced detection accuracy. Among the six hyperparameter configurations tested, the best performance was achieved with a very high precision. These optimized parameters ensure superior reliability in detecting essential navigation features, bridging the gap between disability and independence. The integration of state-of-the-art vision algorithms with depth perception and audio feedback establishes a comprehensive and robust assistive tool that significantly enhances mobility and safety for visually impaired individuals, demonstrating the potential of the system for real-world applications.