Akademik Veri Yönetim Sistemi
5th International Conference on Informatics and Software Engineering, IISEC 2026, Ankara, Türkiye, 5 - 06 Şubat 2026, ss.689-693, (Tam Metin Bildiri)
Cross-view geo-localization aims to determine the geographic location of a query image by matching it against reference images captured from significantly different viewpoints, such as satellite and drone perspectives. This task is particularly challenging due to pronounced appearance variations arising from viewpoint changes, scale differences, and scene geometry distortions. In this paper, we propose an AI-based RGBD framework for cross-view geo-localization that enhances geometric awareness by integrating monocular depth estimation into satellite imagery. Depth maps are generated using a pretrained MiDaS model and fused with RGB channels to form a four-channel (RGBD) input representation. The proposed framework employs a two-stream architecture with modality-specific encoders and adopts a metric learning strategy based on cosine similarity and Circle Loss to learn a discriminative shared embedding space. Experimental evaluations conducted on the University-1652 benchmark demonstrate that incorporating depth information consistently improves retrieval performance compared to an RGB-only baseline, validating the effectiveness of depth-augmented representations for cross-view matching.