영상 검색 기반 측위 기술에 따른 위치 추정 성능 분석

초록

Recent advances in deep learning have led to robust performance in image retrieval-based visual localization, even in diverse environmental conditions. Consequently, state-of-the-art techniques in visual localization now rely heavily on image retrieval methods. However, the effectiveness of image retrieval-based visual localization varies significantly depending on the specific locations where the images were retrieved. In particular, accuracy tends to decrease noticeably when retrieved images are collinear and have high co-visibility. This study aims to analyze the performance of location estimation using image retrieval-based visual localization and explore alternative solutions to mitigate these challenges.