Uma abordagem distribuída para a implementação de sistemas de SLAM visual baseados em reconstrução geométrica

• Ano de defesa: 2018
• Autor(a) principal: Ramos Neto, Otacílio de Araújo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Engenharia Mecânica; Programa de Pós-Graduação em Engenharia Mecânica; UFPB
• Palavras-chave em Português: VSLAM; SLAM Visual; SLAM por métodos indiretos e reconstrução geométrica; VisualSLAM; SLAM byindirect methods and geometric reconstruction; Robótica móvel – Mapeamento – Localização; CNPQ::ENGENHARIAS::ENGENHARIA MECANICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/15211

Visual SLAM systems are important components for the navigation of autonomous vehicles as they simplify the problem of path planning, turning it into verification of all the navigable spaces to the desired location. Currently, there are several solutions to the SLAM problem, such as SLAM by EKF, SLAM by particle filter and the Graph SLAM. Graph SLAM using cameras as sensors is a Visual SLAM method intensively researched in recent years due to the availability of low-cost cameras and the significant amount of environmental information they make accessible. Most of the Visual SLAM systems employs geometric reconstruction based on the parallel processing paradigm developed in the PTAM. This type of system is not scalable for processors with more than two cores. The various PTAMbased SLAM methods were also not designed to scale to all modern processor cores or to execute as a distributed system. This work proposes a modification in the architecture of this type of SLAM method so that it can scale to all processor’s cores or the processors of a weakly coupled multiprocessor system. The modified system according to the technique proposed in this work showed a considerable decrease in the time needed to estimate the position of the camera and the position of the reference points. This improved performance resulted from the additional ability to scale to all processor’scores as well the weakly coupled multiprocessor system.