Скачать или смотреть Visualising reference image matches

In areas where GPS signals are weak or jammed, alternative methods of obtaining absolute position updates such as vision based approaches become important. This research explores lightweight algorithms which can provide good matches with available satellite maps of suitable resolution. The study evaluates the accuracies, robustness, and update rates of various state-of-the-art feature extraction and matching models such as SIFT+SuperGlue, SuperPoint+SuperGlue, Siamese Networks as well as non-deep learning approaches. This study suggests a suitable pipeline for onboard vision based navigation for UAVs while using satellite maps for absolute position updates.