TR_LOC (Transformer based UAV Geo-Localization via Image matching)

Abstract

In this research, we aspire to utilize the acquisition method as it becomes more flexible than previous and produce high-resolution images for navigating and localizing un manned aerial vehicles (UAVs). Unmanned Aerial Vehicles (UAVs), also known as aerial drones, are pervasive in enormous fields, including photography, aerial mapping, agri culture, surveillance, search and rescue, and parcel delivery. However, drone and UAV localization with cross-view matching technology is a challenging research topic. We fa cilitate navigation without the Global Positioning System (GPS), which is favourable in GPS-denied environments. Moreover, it reinforces existing GPS-based systems. In this hypothetical model, a high-resolution satellite map is initially made in accordance with anticipated drone flying by using google map API. To acquire an attention-weighted matrix, the SSC (self-self-cross) and SCC (self-cross-cross) attention combinations are applied on drone and satellite images. we also developed adaptive clustering-based meth ods on a region that receives higher attention for improving the precision of matching anchors and deleting outlier points. Our methods successfully identify relevant matches in unstructured, regular patterns, texture-free, and virgin environments, which typically provide an advantage in existing art matching algorithms. We evaluate the suggested model’s qualitative and quantitative on data that acquired during various drone flights in different regions.. .