Camera-Based Indoor Navigation in Known Environments with ORB for People with Visual Impairment

Abstract

Indoor navigation is a hard problem for those with visual impairment or blindness. This paper presents a real-time camera-based indoor navigation application for the blind and visually impaired in known environments. Contrary to other systems for similar purposes, our system executes all computation locally on mobile devices and does not need any infrastructure-based sensor and signal support. The use of the system is divided into two stages: an offline stage, in which a sighted person helps with landmark identification and route planning, and an online stage, in which a visually impaired person can navigate with voice prompts from the app. Predefined landmark images and navigation information are stored in a database in the offline stage. Navigation information consists of the landmark position, the turning angle, the turning direction, and the distance to the next landmark. In the online stage, navigation information is retrieved from the current landmark image by successfully matching video frames in real time. A similarity score is calculated by ORB using Hamming distances. Voice feedback is provided for users with a text-to-speech function. Our system minimizes the error caused by users’ manipulation and improves stability and accuracy by updating similarity scores continuously.