What you mark is where it was. A user-centric approach for post-hoc georeferencing of urban terrestrial photos

Geotagging user-generated photos is in many cases not sufficient to properly describe their context. Besides not telling in which direction the photo is facing or what is seen in the photo, estimating the actual location during post-hoc georeferencing is cumbersome and imprecise. This paper presents an approach for georeferencing urban photos by marking at least three distinctive features in the photo and connecting them to their real-world locations using a map. Based on these connections, an algorithm calculates the camera viewshed, its orientation, and location. Algorithmic tests under ideal conditions showed high accuracy rates on a centimeter level with average run-times of ten minutes, as well as accuracy rates of 1.5 meters with average run-times of about one second. We developed a web application which helps users to create the connections using two different input types: points and lines. The user study showed accuracy rates of 9 meters in average. The approach is a user-friendly way of generating useful metadata and is relevant to address the issue of metadata scarcity in online photo sharing platforms.