在現今的無人機應用中，幾乎所有都必須依靠定位系統來得知無人機的位置。
但如果在無法使用全球定位系統的地區時，就必須使用其他方法來定位無人機。
在以前的研究中，使用無人機上搭載的相機所拍攝的圖片來進行定位，而使用圖像
來進行定位有兩件事情非常重要，第一個就是提取圖像中的特徵，第二個就是要將
相對應圖像中的特徵對齊。基於視覺的方法可以大致上分為三類，視覺里程計、
同時定位與建圖以及樣板匹配。在這篇論文中使用樣板映射來當作我們主要的方
法。 在這篇論文中我們以[1]提出的自編碼器架構，以及由[2]改進的損失函數作為
基礎。我們在模型中使用額外的潛在相似性損失函數以及不同的訓練技巧來解決時
間遞移所造成景貌不同及透視變形的問題，讓提取出來的圖片特徵更具代表性。在
圖像匹配的階段，我們使用交叉多面體-局部敏感哈希來執行快速的比對任務，並
且利用我們提出的一套定位策略並結合擴展卡爾曼濾波器讓系統可以進行準確的定
位。

Nowadays, the UAV relative application almost all rely on the positioning system to know the location of the UAV. However, if the GPS sensor cannot be used in GPS-denied areas, other methods must be used to locate the drone.
In the previous researches, some people used the images taken by the UAV’s camera for positioning. Using images to locate UAVs, two things are essential: extracting the features in the picture. The other is to align the features between corresponding images to do the positioning. Visual-based methods can be roughly divided into visual odometry, SLAM, and template mapping. In this thesis, we use template mapping as our primary method.
In this paper, we take the AutoEncoder architecture proposed by [1] and improved by [2] as our primary model. We are using additional latent similarity loss and different training techniques to solve the problems of temporal aspect and perspective distortion, making the extracted features of images more representative. In the image matching task, we use Cross-polytope LSH to perform a quick search. In addition, we have proposed a set of localization strategies combined with an Extended Kalman Filter to allow the system to achieve accurate positioning.

[1] Xianxu Hou, Linlin Shen, Ke Sun, and Guoping Qiu. Deep feature consistent variational autoencoder. In 2017 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 1133–1141. IEEE, 2017.
[2] Mollie Bianchi and Timothy D Barfoot. Uav localization using autoencoded satellite images. IEEE Robotics and Automation Letters, 6(2):1761–1768, 2021.
[3] Hunter Goforth and Simon Lucey. Gps-denied uav localization using pre-existing satellite imagery. In 2019 International Conference on Robotics and Automation (ICRA), pages 2974–2980. IEEE, 2019.
[4] Simon J Julier and Jeffrey K Uhlmann. New extension of the kalman filter to nonlinear systems. In Signal processing, sensor fusion, and target recognition VI, volume 3068, pages 182–193. International Society for Optics and Photonics, 1997.
[5] Andrea Cesetti, Emanuele Frontoni, Adriano Mancini, Andrea Ascani, Primo Zingaretti, and Sauro Longhi. A visual global positioning system for unmanned aerial vehicles used in photogrammetric applications. Journal of intelligent & robotic systems, 61(1):157–168, 2011.
[6] Pauline C Ng and Steven Henikoff. Sift: Predicting amino acid changes that affect protein function. Nucleic acids research, 31(13):3812–3814, 2003.
[7] Andrea Ascani, Emanuele Frontoni, Adriano Mancini, and Primo Zingaretti. Feature group matching for appearance-based localization. In 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 3933–3938. IEEE, 2008.
[8] 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.
[9] Dragos Costea and Marius Leordeanu. Aerial image geolocalization from recognition and matching of roads and intersections. arXiv preprint arXiv:1605.08323, 2016.
[10] Anastasiia Volkova and Peter W Gibbens. More robust features for adaptive visual navigation of uavs in mixed environments. Journal of intelligent & robotic systems, 90(1):171–187, 2018.
[11] H. Bai. “variational autoencoder for face image generation in pytorch,”. https: //github.com/bhpfelix/Variational-Autoencoder-PyTorch, 2020.
[12] Sebastian Thrun. Probabilistic robotics. Communications of the ACM, 45(3):52–57, 2002.
[13] Piotr Indyk and Rajeev Motwani. Approximate nearest neighbors: towards removing the curse of dimensionality. In Proceedings of the thirtieth annual ACM symposium on Theory of computing, pages 604–613, 1998.
[14] Moses S Charikar. Similarity estimation techniques from rounding algorithms. In Proceedings of the thiry-fourth annual ACM symposium on Theory of computing, pages 380–388, 2002.
[15] Alexandr Andoni, Piotr Indyk, Thijs Laarhoven, Ilya Razenshteyn, and Ludwig Schmidt. Practical and optimal lsh for angular distance. arXiv preprint arXiv:1509.02897, 2015.
[16] Kengo Terasawa and Yuzuru Tanaka. Spherical lsh for approximate nearest neighbor search on unit hypersphere. In Workshop on Algorithms and Data Structures, pages 27–38. Springer, 2007.