Bi-directional Visual Geo-localization-based Cross-Domain Matching between Digital Twin and Real World

Abstract

Recently, the digital twin has been developing rapidly with the advancement of key enabling technologies such as artificial intelligence and the Internet of Things. To implement the digital twin, cross-domain image matching is required for bi-directional mapping between the real world and the virtual world. Cross-domain image matching enables the use of various image resources by matching the image across different domains. Researchers have contributed different mechanisms to address the cross-domain matching issues. In this study, we investigated and analyzed the performance of the state-of-the-art visual geo-localization approaches for cross-domain matching between the digital twin and the real world. We applied the visual geo-localization technologies to the datasets that we collected from the Berlin Smart City and Bing Map. In the experiment, the visual geo-localization technology, used for cross-view image-based matching between satellite (aerial) view and ground view, performed the best when applied to cross-domain matching between the digital twin and the real world. The evaluation result confirmed the possibility of utilizing the existing visual geo-localization approaches for the cross-domain matching between the digital twin and the real world, and vice versa.