Sistema de navegação inercial para veículos baseado em aprendizado de máquina e correspondência de mapas

• Ano de defesa: 2020
• Autor(a) principal: Gomes, Lucas de Carvalho
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Rio de Janeiro; Brasil; Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia; Programa de Pós-Graduação em Engenharia Elétrica; UFRJ
• Palavras-chave em Português: Navegação inercial; Sistemas de posicionamento por satélite; Aprendizado de máquina; Correspondência de mapas; Engenharia Elétrica
• Link de acesso: http://hdl.handle.net/11422/23216

Nowadays, Global Navigation Satellite Systems (GNSS) are employed in various contexts of the daily life. They yield real-time location information for any vehicle or person bearing a receiver, through the communications by electromagnetic waves between itself and satellites. The performance of this positioning technique is subject to several environmental and technical factors. Furthermore, certain scenarios, such as canyons (urban or geographic), forests and tunnels, are particularly challenging, since the coverage in them is absent or scarcely reliable, producing rogue positioning information or even no information at all. Due to these factors, systems which depend on this information frequently deploy other sensing devices. However, reducing the amount of these devices may be beneficial, since it reduces costs and the energy consumption. Aiming to improve the reliability and the availability of systems based on satellite positioning, keeping, simultaneously, the cost-effectiveness, this work proposes a dead reckoning system, which employs the last known position and data from auxiliary sensors to estimate the current location. The sensors deployed here are available in commercial vehicles. The estimates are calculated by Machine Learning models and improved by a Map Matching procedure. The performance of this procedure was evaluated through computer simulations fed with real data of locations and sensors, keeping track of the system capability of reproducing trajectories in an urban scenario.