A photo showing the skyline of a large city against a grey sky, with water in the foreground: Which city is that? In Geolocation Estimation, man and machine compete against each other and estimate where the photo was taken. The result: Seattle. In most cases, the computer wins - thanks to artificial intelligence.
The procedure for locating the place where photos were taken was developed by the Visual Analytics research group (headed by Prof. Dr. Ralph Ewerth) of the TIB - Leibniz Information Centre for Technology and Natural Sciences. A machine learning procedure based on so-called neural networks uses context-related information of the scene depicted in addition to geographical features to estimate the location of the images. Thus, the neural network acquires the ability to learn specific geographic features for different scenarios. "In the case of city impressions, for example, these are buildings or architectural details; in nature photographs, plants and animals are taken into account," says Prof. Dr. Ewerth, describing the process.
In estimating the location of the shot, the machine is demonstrably better than man. And although the Hanoverian scientists have fed their system with less training data, it already provides better and more accurate data than corresponding systems from Google researchers.
TIB Director and L3S member Prof. Dr. Sören Auer aims to develop this innovative geolocation method into a TIB web service in the medium term. "Such a service would make it easier to find images in memory institutions such as libraries and archives worldwide," says Auer.
You can even compete against the computer in the browser-based demonstration: https://labs.tib.eu/geoestimation
As a hands-on exhibit in the Deutsches Museum Bonn and on the MS Wissenschaft
In spring 2020, the geolocation tool will be on display as a hands-on exhibit to visitors at an exhibition on artificial intelligence (AI) at the Deutsches Museum in Bonn. Already in 2019 the exhibit was part of the exhibition "Artificial Intelligence" on the exhibition ship MS Wissenschaft in many German and Austrian cities.
Prof. Dr. Ralph Ewerth
L3S member Ralph Ewerth is Professor at the Institute of Distributed Systems - Knowledge-based Systems Department of Leibniz University Hannover and heads the Visual Analytics research group of the TIB. He researches visual analysis, search and presentation methods.