Enhancing Image Matching Between Digital Twin and Real-World Through Cross-Domain Geo-Localization Methods

Digital twins, virtual replicas of real-world entities, have garnered significant attention for their ability to synchronize with reality and facilitate simulations. As digital twins generate datasets akin to those from their real-world counterparts, the emergence of diverse image datasets has fueled the demand for advanced image-based matching techniques. However, inherent disparities exist between the representations of digital twin datasets and their real-world counterparts, posing challenges for traditional image-based matching methods. In this paper, we explore and experimentally analyze the performance of existing cross-domain image geo-localization methods between digital twins and the real world. We applied visual feature matching and semantic feature matching techniques to datasets collected from the digital twin and real-world. Our experiments demonstrate that both methods achieve excellent performance, with the semantic feature matching technique showing superior results. This suggests that using semantic information may yield better performance in matching between digital twin and real-world domains, which have different appearances. The evaluation results confirm the potential of using semantic feature matching approaches for cross-domain matching between digital twins and the real world.