AUTHOR=Son Hojun , Weiland James TITLE=Wearable System to Guide Crosswalk Navigation for People With Visual Impairment JOURNAL=Frontiers in Electronics VOLUME=2 YEAR=2022 URL=https://www.frontiersin.org/journals/electronics/articles/10.3389/felec.2021.790081 DOI=10.3389/felec.2021.790081 ISSN=2673-5857 ABSTRACT=

Independent travelling is a significant challenge for visually impaired people in urban settings. Traditional and widely used aids such as guide dogs and long canes provide basic guidance and obstacle avoidance but are not sufficient for complex situations such as street crossing. We propose a new wearable system that can safely guide a user with visual impairment at a signalized crosswalk. Safe street crossing is an important element of fully independent travelling for people who are blind or visually impaired (BVI), but street crossing is challenging for BVI because it involves several steps reliant on vision, including scene understanding, localization, object detection, path planning, and path following. Street crossing also requires timely completion. Prior solutions for guiding BVI in crosswalks have focused on either detection of crosswalks or classifying crosswalks signs. In this paper, we demonstrate a system that performs all the functions necessary to safely guide BVI at a signalized crosswalk. Our system utilizes prior maps, similar to how autonomous vehicles are guided. The hardware components are lightweight such that they can be wearable and mobile, and all are commercially available. The system operates in real-time. Computer vision algorithms (Orbslam2) localize the user in the map and orient them to the crosswalk. The state of the crosswalk signal (don’t walk or walk) is detected (using a convolutional neural network), the user is notified (via verbal instructions) when it is safe to cross, and the user is guided (via verbal instructions) along a path towards a destination on the prior map. The system continually updates user position relative to the path and corrects the user’s trajectory with simple verbal commands. We demonstrate the system functionality in three BVI participants. With brief training, all three were able to use the system to successfully navigate a crosswalk in a safe manner.