Real-Time Object Detection for Visually Impaired Individuals: A Comparative Study of YOLOv8 and YOLOv26

Abstract

Visual impairment significantly limits independent mobility and environmental awareness in daily life. Existing assistive technologies often provide limited contextual understanding and rely on passive feedback, while many deep learning-based detection systems continuously announce all detected objects, increasing cognitive load in real-world use. To address these challenges, this study presents VisionAssist, an iOS-based mobile application that integrates real-time object detection, voice command interaction, and directional auditory feedback to support active, user-centered environmental perception. The system is implemented using Swift and SwiftUI and deployed on-device via Apple’s Core ML framework, ensuring energy efficiency and data privacy. VisionAssist enables users to verbally request specific objects and provides spatial auditory cues indicating object direction and confidence-based reliability. A comparative evaluation between YOLOv8 and YOLOv26 shows that YOLOv8 achieves higher detection accuracy and better usability due to more stable recognition, while YOLOv26 offers lower computational and energy costs but reduced consistency. Overall, YOLOv8 is preferred for interaction quality and trustworthiness, whereas YOLOv26 is more suitable for efficiency-focused use cases. Additionally, confidence-based filtering reduces unnecessary or ambiguous feedback while maintaining situational awareness. The results demonstrate that VisionAssist is a practical and effective assistive solution for improving independent navigation and environmental understanding for visually impaired individuals.