Method of Position Correction Exploiting Street View and Character Tag for Location-based AR games in Urban Environments

Abstract

In location-based AR games, accurately estimating a position of a mobile device is one of important factors. The accuracy of AR contents is determined by how accurately a position estimation system of a mobile device estimates a position. However, in urban environments, signals needed for estimating a position will be reflected, refracted, diffracted, and blocked by obstacles such as buildings (especially skyscrapers), billboards, and signs. As a result, a position error occurs in the position estimated by the position estimation system of the mobile device. In this paper, we propose a simple and reliable method of position correction exploiting commercial street view and character tags. We calculate a matching score by comparing character tags generated from an query image and street view panorama images. Then, according to the matching score, the position of the mobile device is corrected through retrieving a street view that is close to the actual position (ground truth) where the query image was taken. We corrected a position with a position error of 43.71m to a position with a position error of 4.09m using the proposed method. One of the advantages of the proposed method is that the position can be corrected regardless of day or night.