Towards Intelligent Assistive System for Visually Impaired People: Outdoor Navigation System

Abstract

Visually impaired people face numerous challenges in their everyday life, especially in outdoor environments. The most widely adopted mobility aids among the visually impaired community are white cane and guide dogs. With the advent of information and communication technologies (ICTs), they are now making use of smartphones and related services, e.g. GPS to assist them in navigation and everyday activities. Nevertheless, they still experience navigation difficulties in unfamiliar outdoor environments. Deep learning-based approaches have shown tremendous performance in various domains, e.g. computer vision, robotics, etc. which makes them an excellent choice for solving the navigation difficulties faced by visually impaired people. In this paper, we propose a learning-based outdoor navigation system to empower visually impaired people in their independent mobility. The approach builds upon two navigation strategies; namely macro- and micro-navigation. Further, we present our unique pedestrian-centric dataset. To evaluate the system performance, both navigation strategies were put to test. The results show the usability of the navigation system.