Fusing Satellite Imagery and Planimetric Maps for Cross-View Localization
Researchers have developed a novel method to improve cross-view localization by fusing satellite imagery with planimetric maps. This approach addresses the limitations of using single modalities, as satellite images offer fine detail while maps provide annotated objects and remain useful when ground features are obscured. The proposed fusion module enhances standard encoders through cross-modal conditioning and a patch-level fusion rule, leading to a 30.13% reduction in mean localization error and achieving state-of-the-art results. AI