Performance Evaluation of Deep Learning-Based Feature Detection and Matching Algorithms in GNSS/Visual-Inertial Navigation Systems Derin Ö?grenme Tabanli Öznitelik Tespit ve Eşleme Algoritmalarinin GNSS/Görsel-Ataletsel Seyrüsefer Sistemleri Üzerindeki Başarim Analizi
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/siu71813.2026.11636358
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Deep Learning, GVINS, Sensor Fusion, SuperVINS
- Hacettepe Üniversitesi Adresli: Evet
Özet
This study presents a comparative performance analysis of GNSS-aided visual-inertial navigation systems, namely GVINS with the conventional VINS-Mono visual frontend and GVINS integrated with the deep learning-based SuperVINS visual frontend. Experimental results obtained using the proposed software framework show that the high distinctiveness and trackability of deep learning-based features improve the Root Mean Square Error (RMSE) and standard deviation (STD) by 3% and 7%, respectively, in a sports field scenario. In the urban driving scenario, the conventional approach fails in GNSS-challenged areas and in the presence of dynamic objects, whereas the deep learning-based approach maintains a continuous positioning solution and reduces RMSE and STD by 14% and 28%, respectively. The results demonstrate that deep learning-based visual features can enhance system robustness and positioning performance, particularly in challenging urban navigation environments.