Navegação autônoma de VANTs baseada em imagens orbitais e métodos de otimização

• Ano de defesa: 2017
• Autor(a) principal: Ramon Santos Correa
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Engenharia elétrica; Otimização combinatória; Aeronave não tripulada
• Link de acesso: http://hdl.handle.net/1843/BUOS-ATLMV2

The image-based navigation of an Unmanned Aerial Vehicle (UAV) canbe treated as an optimization problem. It's necessary to determine basically four UAV's parameters: latitudinal and longitudinal position, rotation and height. The choice of the most suitable values for these parameters determines the better matching between UAV's registered images and the georeferenced satellite images and, consequently, the smaller estimation error of its trajectory. Nowadays, the main navigation system used is GPS, however there are some factors that limit the use of this technology, so other alternatives are necessary to make navigation safe, reliable and autonomous. The most common approach is to use a computer vision module to aid the navigation of the UAV. When the UAV's rotation and height are unknown or very imprecise, an alternative to the commonly adopted technique, template matching, is the use of optimization methods. Widely used in the medical eld, but still little explored in the UAV's autonomous navigation, optimization methods can be used to estimate the UAV's position with a dissimilarity measure as an objective function of the methods to perform the matching between the images. In this work, deterministic and stochastic optimization methods were used, continuous dissimilarity measures based on intensity and discrete based on extraction of characteristics, and real images obtained from distinct sensors of dierent terrain types (urban, rural and coastal).The purpose of this work is to evaluate the use of optimization methodsin the UAV's image-based autonomous navigation in relation to precision,accuracy and processing time. The application of the optimization methods proved to be promising, mainly because it allows to estimate variables such as rotation and altitude and the possibility of parallelization of the algorithms. Moreover, it was possible to identify the optimization methods and the dissimilarity measures most appropriate to be used in practical applications.