Comparação de métodos de filtragem e geração de modelos digitais de terreno a partir de imagens obtidas por veículo aéreo não-tripulado
Carregando...
Data
2017-12-07
Autores
Niemann, Rafaela Soares [UNESP]
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Os modelos digitais de elevação são importantes para a geração de informações sobre variáveis ambientais correlacionadas à topografia, principalmente como subsídio à análises geomorfométricas. O sensoriamento remoto pode contribuir com a geração de modelos digitais de elevação, principalmente através do uso de sensores de alta resolução espacial e tecnologias avançadas. Os VANTs – Veículos Aéreos Não Tripulados – tem sido cada vez mais explorados no âmbito da cartografia e topografia, com atuação cada vez mais importante dentro da ciência, devido à capacidade de transportarem diferentes sensores e ao seu baixo custo de operação. Câmeras fotográficas simples acopladas aos VANTs podem ser combinadas com tecnologias de visão computacional, trazendo novas oportunidades para explorar a geração de modelos digitais de elevação. Algoritmos de visão computacional, como o Structure-from-Motion (SfM), permitem a extração de pontos tridimensionais a partir de imagens sobrepostas obtidas por VANTs. Esses pontos compõem nuvens de pontos capazes de subsidiar a geração de modelos digitais de superfície (MDS), quando combinadas com algoritmos de interpolação de dados. Contudo, os modelos gerados desta maneira nos retornam informações relacionadas à superfície dos objetos presentes sobre o terreno, incluindo por exemplo construções e dosséis vegetais. A filtragem e classificação das nuvens de pontos se faz assim necessária para geração de modelos digitais que descrevam mais fielmente a superfície do terreno, sem estes elementos. Nesta dissertação, avaliamos dois métodos para a filtragem e interpolação de modelos digitais de terreno (MDT) a partir de nuvens de pontos geradas por imageamento ótico baseado em VANTs. A área de estudo escolhida foi a região da Serra do Cipó-MG, caracterizada por relevo acidentado e cobertura vegetal variada. O primeiro método consistiu na filtragem (classificação) direta da nuvem de pontos, e o segundo na filtragem do modelo digital de superfície em formato raster, ambos seguidos de interpolação. Os métodos avaliados se mostraram adequados, com coeficientes de determinação da ordem de R² = 0,98 em relação a dados de referência tomados por DGPS. A filtragem foi bastante eficiente para áreas íngremes e com vegetação baixa, e menos eficiente em áreas de vegetação arbórea densa. Os métodos avaliados no presente trabalho contribuirão para a melhoria da geração de MDTs com base na tecnologia emergente oferecida pelos VANTs, que poderão ser utilizados como subsídios a estudos ambientais diversos.
Digital elevation models are important for producing information on different environmental variables correlated to topography, especially for geomorphometric analyses. Remote sensing can contribute to the generation of digital elevation models, mainly through high spatial resolution sensors and advanced technologies. UAVs - Unmanned Aerial Vehicles - have been increasingly employed in the fields of cartography and topography, and have had an increasingly prominent role in science, as they can carry different sensors and have low-cost operation. Simple cameras attached to UAVs can be combined with computer vision technologies, bringing new opportunities to explore the production of digital elevation models. Computer vision algorithms such as Structure-from-Motion (SfM) allow the extraction of threedimensional points from superimposed images obtained by UAVs. These points make up points clouds capable of supporting the production of digital surface models (DSM) when combined with interpolation algorithms. However, models generated this way give us information related to the surface of objects present on the ground, including buildings and plant canopies. Point cloud filtering and classification is thus necessary for producing digital models that more accurately describe the bare terrain surface. In this dissertation, we evaluated two methods for filtering and interpolating digital terrain models (DTM) from point clouds generated by UAVbased optical imaging. The chosen study area was the Serra do Cipó region (Minas Gerais, Brazil), characterized by rugged relief and heterogeneous vegetation cover. The first method consisted of direct filtering (classification) of the point cloud, and the second method was based on filtering the digital surface model in raster format, both followed by interpolation. The evaluated methods were adequate, with determination coefficients of the order of R² = 0.98 in relation to reference data taken by DGPS. Filtering was quite efficient for steep areas with low vegetation, and less efficient in areas of dense arboreal vegetation. The methods evaluated in the present work will contribute to the improvement and generation of DTMs based on the emerging technology offered by UAVs, which can be used as subsidies to diverse environmental studies.
Digital elevation models are important for producing information on different environmental variables correlated to topography, especially for geomorphometric analyses. Remote sensing can contribute to the generation of digital elevation models, mainly through high spatial resolution sensors and advanced technologies. UAVs - Unmanned Aerial Vehicles - have been increasingly employed in the fields of cartography and topography, and have had an increasingly prominent role in science, as they can carry different sensors and have low-cost operation. Simple cameras attached to UAVs can be combined with computer vision technologies, bringing new opportunities to explore the production of digital elevation models. Computer vision algorithms such as Structure-from-Motion (SfM) allow the extraction of threedimensional points from superimposed images obtained by UAVs. These points make up points clouds capable of supporting the production of digital surface models (DSM) when combined with interpolation algorithms. However, models generated this way give us information related to the surface of objects present on the ground, including buildings and plant canopies. Point cloud filtering and classification is thus necessary for producing digital models that more accurately describe the bare terrain surface. In this dissertation, we evaluated two methods for filtering and interpolating digital terrain models (DTM) from point clouds generated by UAVbased optical imaging. The chosen study area was the Serra do Cipó region (Minas Gerais, Brazil), characterized by rugged relief and heterogeneous vegetation cover. The first method consisted of direct filtering (classification) of the point cloud, and the second method was based on filtering the digital surface model in raster format, both followed by interpolation. The evaluated methods were adequate, with determination coefficients of the order of R² = 0.98 in relation to reference data taken by DGPS. Filtering was quite efficient for steep areas with low vegetation, and less efficient in areas of dense arboreal vegetation. The methods evaluated in the present work will contribute to the improvement and generation of DTMs based on the emerging technology offered by UAVs, which can be used as subsidies to diverse environmental studies.
Descrição
Palavras-chave
Sensoriamento remoto, VANT, UAV, DTM, Modelos digitais de elevação, Nuvem de pontos, Filtragem, Digital elevation models, Point clouds