Krzysztof Gromada
supervisor: Barbara Siemiątkowska
Unmanned Aerial Vehicles are the fastest-growing segment of the military aerospace market. Their popularity rapidly increases in civil, commercial usage, and national security agencies for search-and-rescue or environment protection missions. A wide range of UAV payloads (including electro-optical visible/infrared light (EO/IR) cameras, Synthetic Aperture Radars (SAR)) is systematically enriched and introduced to the market.
Due to higher system integrity and reliability demand in the market, new localization systems need to be implemented for GPS (and other radio-based localization systems) signal unavailability scenarios. These can occur when a GPS signal is jammed, damned by nearby objects/terrain, or disabled with the conflict declaration.
Presented work defines an approach to localization of the UAV based on Visible Light and/or Thermal and SAR Imagery analysis. The central element of the systems is a set of algorithms that conduct segmentation tasks on given images. For SAR imagery is used as well as deep neural networks For optical, near-infrared and thermal cameras segmentation Deep Neural Networks are used, while for the SAR imagery additionally a dedicated histogram-based segmentation method was developed. The resulting segmentation mask allows for global UAV positioning on digital maps using optimization algorithms and stereometrics.