| Title |
Geo-localization in OpenStreetMap-derived semantic map space using structural embeddings |
| Authors |
Kanapickas, Tomas ; Lukošius, Ugnius ; Kriščiūnas, Andrius |
| Full Text |
|
| Is Part of |
CEUR workshop proceedings: IVUS 2026: proceedings of the 31st international conference on information society and university studies (IVUS 2026), Kaunas, Lithuania, May 12, 2026 / edited by Audrius Lopata,Tomas Krilavičius, Ilona Veitaitė.. Aachen : CEUR-WS. 2026, vol. 4317, p. 1-8.. ISSN 1613-0073 |
| Keywords [eng] |
UAV localization ; geo-localization ; semantic maps ; OpenStreetMap ; embedding retrieval ; contrastive learning ; transformer encoder ; GIS-based localization |
| Abstract [eng] |
Unmanned aerial vehicle (UAV) localization is important in many applications, especially when Global Positioning System (GPS) signals are unreliable or unavailable. While many approaches rely on matching RGB imagery, this work investigates geo-localization using only structural semantic information derived from OpenStreetMap (OSM) data. The proposed method operates entirely within a rasterized semantic map containing roads and buildings, allowing isolation of structural information without visual effects such as lighting, vegetation, or seasonal changes. Localization is formulated as an embedding-based retrieval task. Semantic map tiles are encoded using a convolutional–transformer network. Two additional variants are evaluated: an object-wise representation and a graph-based approach using a GINE model. The models are tested on a database of 13,330 semantic tiles using cosine similarity. Results show that structural and topological information alone can support large-scale localization. Although strict Top-1 performance is limited at small spatial thresholds, Top-1% retrieval consistently identifies the correct geographic region within practical distances. The method is computationally efficient, requiring only milliseconds per query, making it suitable for real-time applications. |
| Published |
Aachen : CEUR-WS |
| Type |
Conference paper |
| Language |
English |
| Publication date |
2026 |
| CC license |
|
