Efficient UAV localization using combined autoencoder and SIFT

In GNSS-denied environments, accurate Unmanned Aerial (UAV) localization faces significant challenges. This paper introduces a vision-based localization method combining autoencoder and SIFT algorithms, referred to as AE+SIFT. The method compresses high-resolution map images into low-dimensional vectors, which are stored in a database for efficient retrieval. During the localization process, UAV images are encoded and matched with the database, followed by SIFT and homography projection for precise positioning. The AE+SIFT approach enhances localization accuracy, achieving an average coordinate error of 3.94 meters relative to the ground truth. Notably, when UAV images are misaligned with reference images, our method outperforms the existing AE method in terms of accuracy.