Urban photograph localization using the INSTREET application—accuracy and performance analysis

Abstract

The paper proposes a solution to the problem of geolocation of photographs by using an algorithm to compare their content against a geolocated database of street view images, and analyzing the performance of the algorithm. The algorithm makes it possible to pinpoint the location where a photograph was taken. In order to solve this problem, we propose an algorithm based on MPEG-7 features. The paper also describes the results of optimizing the performance of the algorithm and its accuracy. We show that the algorithm scales with the size of the reference database at least up to 130 km 2, which was the largest urban area we tested the algorithm on.