無人機（Unmanned Aerial Vehicle, UAV）廣泛應用於橋梁檢測、基礎設施巡檢及近距離影像拍攝等任務，但在橋梁底部、橋墩周圍或其他衛星訊號易受遮蔽的區域，全球導航衛星系統（Global Navigation Satellite System, GNSS）可能出現訊號中斷或定位不穩定的情形，進而影響無人機位置與姿態估計。為因應上述問題，本研究提出一套基於可重複使用之三維特徵地圖的單張影像視覺定位方法，作為無人機於無 GNSS 或 GNSS 訊號不穩定環境下之輔助定位方法。研究流程主要分為三維特徵地圖建置與單張測試影像外方位參數解算兩個部分。在三維特徵地圖建置階段，首先對無人機拍攝之多視角橋梁影像進行特徵萃取、特徵匹配及特徵軌跡建立，再透過光束法平差（Bundle Adjustment, BA）解算建圖影像之外方位參數並生成稀疏點雲。接著，整理稀疏點雲、影像觀測關係及代表描述子，形成可供後續定位使用之三維特徵地圖。在單張影像定位階段，將無人機新取得之影像作為測試影像，萃取其特徵點與描述子，並與三維特徵地圖進行匹配，以建立二維影像特徵與三維地圖點之 2D--3D 對應關係，最後透過單張後方交會解算相機之外方位參數（Exterior Orientation Parameters, EOPs）。本研究首先比較 SIFT、SURF、SuperPoint、ALIKED 及 DISK 五種特徵萃取方法於橋梁影像與公開資料集中的特徵萃取及匹配表現，再選用 SuperPoint、ALIKED 及 DISK 建置三維特徵地圖並進行單張影像後方交會解算。實驗結果顯示，三種特徵萃取方法皆可完成三維特徵地圖建置及 6 張原始測試影像之外方位參數解算。其中，ALIKED 於光束法平差後之重投影均方根誤差（Root Mean Square Error, RMSE）為 0.84 px，且測試影像定位之三維位置 RMSE 與姿態 RMSE 分別為 0.114 m 與 0.256°，整體成果優於 SuperPoint 與 DISK；其平均單張定位時間約為 3.77 秒，亦為三種方法中最短。利用預先建置之三維特徵地圖，後續測試影像可直接與既有三維特徵地圖建立 2D--3D 對應並解算外方位參數，無須重新與所有建圖影像共同執行特徵匹配及光束法平差。為驗證三維特徵地圖重複使用之可行性，本研究另針對代表性測試影像進行亮度、旋轉與尺度變化測試。結果顯示，在一定程度的影像條件變化下，三維特徵地圖仍可與測試影像建立有效對應並完成外方位參數解算，其中 ALIKED 於多數測試條件下呈現較穩定之定位成果。綜合上述結果，本研究所建置之三維特徵地圖具有潛力，可應用於 GNSS 訊號受限環境下無人機單張影像視覺定位，並可作為 GNSS 訊號受限環境下無人機輔助定位與導航研究之基礎。

This study develops a single-image visual localization workflow for UAV-assisted bridge inspection in GNSS-denied or GNSS-degraded environments. A reusable three-dimensional (3D) feature map is first constructed from multi-view bridge images acquired by a UAV. A subsequent test image can then be matched with the pre-constructed map to establish 2D--3D correspondences, and the exterior orientation parameters of the camera can be estimated through single-image space resection. Five feature extraction methods, namely SIFT, SURF, SuperPoint, ALIKED, and DISK, were first compared in terms of feature extraction and image matching performance. Based on the comparison results, SuperPoint, ALIKED, and DISK were selected for feature track construction, bundle adjustment, 3D feature map construction, and single-image exterior orientation parameter estimation.The experimental results showed that DISK generated the largest sparse point cloud and the highest number of inliers; however, these numerical advantages did not result in the highest localization accuracy. ALIKED achieved the lowest reprojection RMSE, the highest check-point coordinate accuracy, the smallest differences in mapping-image exterior orientation parameters relative to the Metashape reference results, and the highest single-image localization accuracy. For the six original test images, ALIKED achieved a position RMSE of 0.114 m, an attitude RMSE of 0.256 degrees, and an average localization time of approximately 3.77 s. These results indicate that the geometric consistency of 2D--3D correspondences and the quality of map points are more important than feature or map-point quantity alone. The proposed workflow demonstrates the potential of reusable 3D feature maps for UAV single-image visual localization and provides a basis for future UAV auxiliary positioning and navigation applications in bridge inspection environments.

[1] Abdulrahman Alarifi, AbdulMalik Al-Salman, Mansour Alsaleh, Ahmad Alnafessah, Suheer Al-Hadhrami, Mai Al-Ammar, and Hend Al-Khalifa. Ultra wideband indoor positioning technologies: Analysis and recent advances. Sensors (Basel, Switzerland), 16(5):707–, 2016.
[2] Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pajdla, and Josef Sivic. Netvlad: Cnn architecture for weakly supervised place recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5297–5307, 2016.
[3] Herbert Bay, Andreas Ess, Tinne Tuytelaars, and Luc Van Gool. Speeded-up robust features (surf). Computer vision and image understanding, 110(3):346–359, 2008.
[4] Samarth Brahmbhatt, Jinwei Gu, Kihwan Kim, James Hays, and Jan Kautz. Geometry-aware learning of maps for camera localization. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2616–2625, New York, 2018. IEEE.
[5] Federico Camposeco, Andrea Cohen, Marc Pollefeys, and Torsten Sattler. Hybrid scene compression for visual localization. CoRR, abs/1807.07512, 2018.
[6] Bhupesh Chand, Frezer Ayele, Ian Pineiro-Dakers, Reihaneh Samsami, and Byungik Chang. Uncrewed aerial system (uas) applications in bridge inspection: A comprehensive review of platforms, sensors, and operational effectiveness. Drones, 10(2), 2026.
[7] COLMAP Developers. COLMAP Datasets. https://colmap.github.io/datasets.html. Accessed: 2026-08-12.
[8] Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superpoint: Self-supervised interest point detection and description. In Proceedings of the IEEE conference on computer vision and pattern recognition workshops, pages 224–236, 2018.
[9] Mihai Dusmanu, Ignacio Rocco, Tomáš Pajdla, Marc Pollefeys, Josef Sivic, Akihiko Torii, and Torsten Sattler. D2-net: A trainable CNN for joint detection and description of local features. CoRR, abs/1905.03561, 2019.
[10] Electronic Code of Federal Regulations. 23 cfr part 650 subpart c – national bridge inspection standards. https://www.ecfr.gov/current/title-23/chapter-I/subchapter-G/part-650/subpart-C, 2026. Accessed: 2026-07-09.
[11] Federal Highway Administration. Bridge Inspector’s Reference Manual (BIRM) – 2022 NBIS Version. Federal Highway Administration, March 2023. Accessed: 2026-07-09.
[12] Federal Highway Administration. National bridge inspection standards. https://www.fhwa.dot.gov/bridge/nbis.cfm, 2026. Accessed: 2026-07-09.
[13] Mengdan Feng, Sixing Hu, Marcelo H Ang, Gim Hee Lee, A Howard, K Althoefer, F Arai, F Arrichiello, B Caputo, J Castellanos, K Hauser, Isler, J Kim, H Liu, P Oh, Santos, D Scaramuzza, A Ude, R Voyles, K Yamane, and A Okamura. 2d3d-matchnet: Learning to match keypoints across 2d image and 3d point cloud. In Proceedings - IEEE International Conference on Robotics and Automation, pages 4790–4796, New York, 2019. IEEE.
[14] Martin A. Fischler and Robert C. Bolles. Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography, 1981.
[15] G. Forlani, F. Diotri, U. Morra di Cella, and R. Roncella. UAV block georeferencing and control by on-board GNSS data. International archives of the photogrammetry, remote sensing and spatial information sciences., XLIII-B2-2020:9–16, 2020.
[16] Paul D. Groves. Principles of GNSS, inertial, and multisensor integrated navigation systems / Paul D. Groves. GNSS technology and application series. Artech House, Boston, 2nd ed. edition, 2013.
[17] B. Hofmann-Wellenhof, Herbert. Lichtenegger, and Elmar. Wasle. GNSS–global navigation satellite systems : GPS, GLONASS, Galileo, and more / Bernhard Hofmann-Wellenhof, Herbert Lichtenegger, Elmar Wasle. Springer, Wien ;, 2008.
[18] H Jégou, M Douze, C Schmid, and P Pérez. Aggregating local descriptors into a compact image representation. In 2010 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 3304–3311, Los Alamitos, 2010. IEEE.
[19] Alex Kendall, Matthew Grimes, and Roberto Cipolla. Posenet: A convolutional network for real-time 6-dof camera relocalization. In Proceedings / IEEE International Conference on Computer Vision, pages 2938–2946, New York, 2015. IEEE.
[20] Jouko Kinnari, Francesco Verdoja, and Ville Kyrki. Gnss-denied geolocalization of uavs by visual matching of onboard camera images with orthophotos. CoRR, abs/2103.14381, 2021.
[21] Yunpeng Li, Noah Snavely, Daniel Huttenlocher, and Pascal Fua. Worldwide pose estimation using 3D point clouds. In European Conference on Computer Vision, 2012.
[22] Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Pollefeys. Lightglue: Local feature matching at light speed. In Proceedings / IEEE International Conference on Computer Vision, pages 17581–17592, Los Alamitos, 2023. IEEE.
[23] David G. Lowe. Distinctive image features from scale-invariant keypoints. International journal of computer vision, 60(2):91–110, 2004.
[24] Stephanie Lowry, Niko Sunderhauf, Paul Newman, John J. Leonard, David Cox, Peter Corke, and Michael J. Milford. Visual place recognition: A survey. IEEE transactions on robotics, 32(1):1–19, 2016.
[25] Guilherme Potje, Felipe Cadar, Andre Araujo, Renato Martins, and Erickson R. Nascimento. Xfeat: Accelerated features for lightweight image matching. In 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2682–2691, Los Alamitos, 2024. IEEE Computer Society, IEEE.
[26] Tong Qin, Peiliang Li, and Shaojie Shen. Vins-mono: A robust and versatile monocular visual-inertial state estimator. IEEE transactions on robotics, 34(4):1004–1020, 2018.
[27] Jorge Pena Queralta, Carmen Martinez Almansa, Fabrizio Schiano, Dario Floreano, and Tomi Westerlund. Uwb-based system for uav localization in gnss-denied environments: Characterization and dataset. In Proceedings of the ... IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 4521–4528, New York, 2020. IEEE.
[28] Jérôme Revaud, Philippe Weinzaepfel, César Roberto de Souza, Noé Pion, Gabriela Csurka, Yohann Cabon, and Martin Humenberger. R2D2: repeatable and reliable detector and descriptor. CoRR, abs/1906.06195, 2019.
[29] E. Rublee, V. Rabaud, K. Konolige, and G. Bradski. ORB: An efficient alternative to SIFT or SURF. In 2011 IEEE International Conference on Computer Vision (ICCV), pages 2564–2571, New York, 2011. IEEE.
[30] Paul-Edouard Sarlin, Cesar Cadena, Roland Siegwart, and Marcin Dymczyk. From coarse to fine: Robust hierarchical localization at large scale. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 12708–12717, New York, 2019. IEEE Comp Soc, IEEE.
[31] Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4937–4946, Los Alamitos, 2020. IEEE.
[32] Torsten Sattler, Bastian Leibe, and Leif Kobbelt. Fast image-based localization using direct 2d-to-3d matching. In 2011 International Conference on Computer Vision, pages 667–674, 2011.
[33] Torsten Sattler, Bastian Leibe, and Leif Kobbelt. Improving image-based localization by active correspondence search. In Computer Vision – ECCV 2012, volume 7572 of Lecture Notes in Computer Science, pages 752–765. Springer, Berlin, Heidelberg, 2012.
[34] Torsten Sattler, Bastian Leibe, and Leif Kobbelt. Efficient & effective prioritized matching for large-scale image-based localization. IEEE transactions on pattern analysis and machine intelligence, 39(9):1744–1756, 2017.
[35] Torsten Sattler, Will Maddern, Carl Toft, Akihiko Torii, Lars Hammarstrand, Erik Stenborg, Daniel Safari, Masatoshi Okutomi, Marc Pollefeys, Josef Sivic, Fredrik Kahl, and Tomas Pajdla. Benchmarking 6dof outdoor visual localization in changing conditions. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8601–8610, New York, 2018. IEEE.
[36] Torsten Sattler, Qunjie Zhou, Marc Pollefeys, and Laura Leal-Taixe. Understanding the limitations of cnn-based absolute camera pose regression. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), volume 2019-June, pages 3297–3307, Los Alamitos, 2019. IEEE Comp Soc, IEEE.
[37] Johannes L. Schonberger, Marc Pollefeys, Andreas Geiger, and Torsten Sattler. Semantic visual localization. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6896–6906, New York, 2018. IEEE.
[38] Thomas Schöps, Johannes L. Schönberger, Silvano Galliani, Torsten Sattler, Konrad Schindler, Marc Pollefeys, and Andreas Geiger. A multi-view stereo benchmark with high-resolution images and multi-camera videos. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 3260–3269, 2017.
[39] Johannes L. Schönberger and Jan-Michael Frahm. Structure-from-motion revisited. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 4104–4113, New York, 2016. IEEE.
[40] Tixiao Shan, Brendan Englot, Drew Meyers, Wei Wang, Carlo Ratti, and Daniela Rus. Lio-sam: Tightly-coupled lidar inertial odometry via smoothing and mapping. In Proceedings of the ... IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5135–5142, New York, 2020. IEEE.
[41] J Sivic and A Zisserman. Video google: a text retrieval approach to object matching in videos. In Computer Vision, 9th IEEE International Conference On, pages 1470–1477 vol.2, Los Alamitos, 2003. IEEE.
[42] Jiaming Sun, Zehong Shen, Yuang Wang, Hujun Bao, and Xiaowei Zhou. Loftr: Detector-free local feature matching with transformers. CoRR, abs/2104.00680, 2021.
[43] Richard Szeliski. Computer Vision: Algorithms and Applications. Springer Nature, 2 edition, 2022.
[44] Akihiko Torii, Relja Arandjelovic, Josef Sivic, Masatoshi Okutomi, and Tomas Pajdla. 24/7 place recognition by view synthesis. In 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1808–1817, New York, 2015. IEEE.
[45] Bill Triggs, Philip F. McLauchlan, Richard I. Hartley, and Andrew W. Fitzgibbon. Bundle adjustment – a modern synthesis. In Vision Algorithms: Theory and Practice, volume 1883 of Lecture Notes in Computer Science, pages 298–372. Springer, Berlin, Heidelberg, 2000.
[46] Michal J. Tyszkiewicz, Pascal Fua, and Eduard Trulls. DISK: learning local features with policy gradient. CoRR, abs/2006.13566, 2020.
[47] Oliver J Woodman. An introduction to inertial navigation. Technical report, University of Cambridge, Computer Laboratory, 2007.
[48] Hailin Yu, Youji Feng, Weicai Ye, Mingxuan Jiang, Hujun Bao, and Guofeng Zhang. Improving feature-based visual localization by geometry-aided matching, 2023.
[49] Kevin Yu, Prajwal Shanthakumar, Jonah Orevillo, Eric Bianchi, Matthew Hebdon, and Pratap Tokekar. View planning and navigation algorithms for autonomous bridge inspection with uavs. CoRR, abs/1910.02786, 2019.
[50] Ji Zhang, Sanjiv Singh, et al. LOAM: Lidar odometry and mapping in real-time. In Robotics: Science and systems, volume 2, pages 1–9. Berkeley, CA, 2014.
[51] Xiaoming Zhao, Xingming Wu, Weihai Chen, Peter C. Y. Chen, Qingsong Xu, and Zhengguo Li. Aliked: A lighter keypoint and descriptor extraction network via deformable transformation. IEEE transactions on instrumentation and measurement, 72:1–16, 2023.
[52] 交通部. 公路橋梁檢測及補強規範, 2018. 中華民國107年10月.
[53] 国土交通省道路局国道・技術課 and 国土技術政策総合研究所道路構造物研究部. 道路構造物の定期点検の実施にかかる参考資料：「道路橋定期点検要領（令和6年3月）」運用の手引き. https://www.mlit.go.jp/road/sisaku/yobohozen/tenken/yobo7_23.pdf, April 2025. Accessed: 2026-07-09.