Um método para identificação de superfície aquática turva para navegação autônoma

• Ano de defesa: 2016
• Autor(a) principal: Colet, Mateus Eugênio
• Orientador(a): Manssour, Isabel Harb
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: Faculdade de Informática
• País: Brasil
• Palavras-chave em Português: REDES NEURAIS (COMPUTAÇÃO); SISTEMAS EMBARCADOS (COMPUTADORES); ROBÓTICA; INFORMÁTICA
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/7181

Navigation in aquatic environments is a broad topic that in recent years has received considerable attention from the community working with mobile robotics. The use of aquatic surface vehicles for inspection, mitigation and natural aquatic disasters management, boosted the search for autonomous navigation in this area. in order to perform an autonomous navigation in outdoor environments, it is necessary to identify parts of the surface that can be navigable, and this is one of the fundamental problems in this type of application. In this context, the objective of this research is to propose a method for water surfaces identification based on the blurred optical property, found in these types of environments. More specifically, computer vision was used in conjunction with neural networks to build a classifier, which has the task of distinguishing and identifying navigable aquatic surface. ln order to achieve this objective, a study on the use of several features based on color and texture of these turbid surfaces for the extraction of various attributes to generate the classifier, such as: mean, variance, entropy and energy, varying in different color channels (RGB, HSV, YUV). In order to compress all of this information it was used statistical method of principal component analysis, whose results were used as input of the artificial neural network, thus constructing the classifier. The classifier has the fundamental task of generating the navigation map that is interpreted by a state machine for decision making. All the method developed was applied and embarked in aquatic vehicle prototype at the same time the results and assessments were validated using the vehicle in real environments and different scenarios.