Estratégias para aumentar a robustez de estimação de posição geográfica em VANTs através de imagens

• Ano de defesa: 2016
• Autor(a) principal: Brayan Rene Acevedo Jaimes
• 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; Engenharia elétrica; Detecção de sinais; Aeronave não tripulada
• Link de acesso: http://hdl.handle.net/1843/BUOS-ATJMTK

In this thesis, different approaches to improve the geographical position estimation process through UAV (Unmanned Aerial Vehicles) images are proposed. In first place, two new template matching approaches with low processing time were developed aiming to make a more robust autonomous navigation of the aircrafts without the need to useGPS (Global Position System) signal. The first uses an adaptive Canny edge detector and the second one uses thresholding. With these techniques, it is possible to solve the edge overestimation and the noise inclusion that affect the image comparison and, consequently, the position estimation. In second place, two approaches to correct projective distortion, scale adjustment, rotation and translation in UAV images were developed when the camera position is not perpendicular to earth. The first of them is a technique that uses the (previous) knowledge of UAV tilt angles provided by the aircraft inertial sensors to obtain the homographic matrix and correct the image. These angles compose the rotationparameters of the homographic matrix that is also composed by other concatenated matrices that representing the camera intrinsic parameters and the image translation. The second approach presents a robust correction of projective and spectral distortions in images captured by UAVs. This technique is based in the keypoints matching extractedbetween the UAV image and the georeferenced one. It also uses the SURF and MSAC algorithms in order to estimate the parameters that compose the homographic matrix and, thus, the image is corrected.The evaluation of the proposed approaches considered different land types (forest, urban and highway) in the tests application. Also, the evaluation with images obtained from different sensors with distortion of perspective, scale, rotation, and translation is considered. The evaluation metrics were the mean distance error in the position estimation and the processing time. Now, for the perspective distortion correction, metrics like the keypoints number extracted on each image, the estimated matching number between images, efficiency, recall, precision and processing time were consideiii red. The obtained results throughout the different tests applied in the techniques are promissory, have low processing time and indicate that they can be used in real flight conditions.