Vision Transformer-Based Cross-Modal Change Detection Method for HD Maps HD Haritalar için Görü Dönüştürücü Tabanli Çapraz-Modlu De?gişim Tespiti Yöntemi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636763
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: change detection, crossmodal fusion, High definition maps, Vision Transformer
- Hacettepe Üniversitesi Adresli: Evet
Özet
In autonomous driving, a high-definition (HD) map is a critical component for lane-level localization and safe self-driving decisions. However, the dynamic nature of urban environments leads to inconsistencies between the real world and pre-built HD maps over time; therefore, changes must be detected reliably in the map maintenance process. In this study, we present a ViT-DeepLabV3+ based change detection framework that detects the change (change/no-change) between a perspective-view camera image and its corresponding egoview projected HD map view, and is robust to domain gap. In our approach, feature maps are generated using DeepLabV3+ based feature extractors specialized for each modality, and these representations are compared with a Vision Transformer to make the change decision. Experimental results show that the proposed method successfully detects changes and achieves better performance than previous studies.