Research on Real-time Optimization of Intelligent Guide System Integrating Visual Perception and Edge Computing

Abstract

With the rapid advancement of artificial intelligence and Internet of Things (IoT) technologies, intelligent assistive devices are gradually transforming the daily lives of visually impaired individuals. However, existing navigation assistance systems face critical challenges in complex dynamic environments, including insufficient real-time performance and limited computational resources.This paper proposes an intelligent navigation system based on the fusion of visual perception and edge computing, which addresses these technical bottlenecks through three key innovations:(1) An information richness assessment model based on superpixel segmentation and optical flow analysis, enabling data-driven adaptive preprocessing;(2) A lightweight multi-task neural network, MT-LiteNet, optimized for edge platforms;(3) A dynamic offloading framework with end-edge-cloud collaboration.Experimental results demonstrate that the system achieves an average end-to-end latency of 87 ms (a 5.3× improvement over traditional cloud-based solutions) in typical urban environments, with an obstacle recognition accuracy of 92.4% and a 37% reduction in energy consumption.This research not only provides a safe and reliable navigation solution for the visually impaired but also offers a scalable technical framework applicable to other real-time edge computing scenarios, such as autonomous driving and industrial inspection.