Indoor localization for visually impaired travelers using computer vision on a smartphone

Abstract

Wayfinding is a major challenge for visually impaired travelers, who generally lack access to visual cues such as landmarks and informational signs that many travelers rely on for navigation. Indoor wayfinding is particularly challenging since the most commonly used source of location information for wayfinding, GPS, is inaccurate indoors. We describe a computer vision approach to indoor localization that runs as a real-time app on a conventional smart-phone, which is intended to support a full-featured wayfinding app in the future that will include turn-by-turn directions. Our approach combines computer vision, existing informational signs such as Exit signs, inertial sensors and a 2D map to estimate and track the user's location in the environment. An important feature of our approach is that it requires no new physical infrastructure. While our approach requires the user to either hold the smartphone or wear it (e.g., on a lanyard) with the camera facing forward while walking, it has the advantage of not forcing the user to aim the camera towards specific signs, which would be challenging for people with low or no vision. We demonstrate the feasibility of our approach with five blind travelers navigating an indoor space, with localization accuracy of roughly 1 meter once the localization algorithm has converged.