Geo-localization of ground and aerial imagery by cross-view alignment to overhead reference

Abstract:

This talk will focus on cross-view geo-localization, specifically aligning ground-level RGB images with overhead reference data such as aerial or satellite imagery, geo-referenced 3D data, and maps. We will explore practical data choices and the challenges of bridging the domain gap between ground and aerial views. Different methods for overcoming this gap, including geometric transformations and feature alignment techniques, will be introduced. Additionally, we will discuss the potential of multi-modal data fusion and how combining RGB images with other data sources can improve geo-localization accuracy.