@PhDThesis{PenhaNeto:2021:NaAuVA,
author = "Penha Neto, Gerson da",
title = "Navega{\c{c}}{\~a}o aut{\^o}noma de VANT por fus{\~a}o de
dados com rede neural artificial otimizada implementada em FPGA",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2021",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2021-01-26",
keywords = "navega{\c{c}}{\~a}o aut{\^o}noma, fus{\~a}o de dados,
ve{\'{\i}}culo a{\'e}reo n{\~a}o tripulado, rede neural
autoconfigurada, autonomous navigation, data fusion, unmanned
aerial vehicle, auto-configured neural network, field programmable
gate array.",
abstract = "O uso de VANT aumentou nos {\'u}ltimos anos e devido o aumento do
emprego de VANTs, cresceu tamb{\'e}m o interesse de empresas,
organiza{\c{c}}{\~o}es governamentais e da comunidade
cient{\'{\i}}fica, no desenvolvimento de sistemas aut{\^o}nomos
a{\'e}reos, principalmente de pequeno porte. A
navega{\c{c}}{\~a}o necessita de sensores, como um Global
Navigation Satellite System (GNSS) e um Inertial Navigation System
(INS). Contudo, existem problemas associados ao uso destes
sensores. Os principais problemas s{\~a}o interfer{\^e}ncias,
que podem ser provocadas intencionalmente ou por a{\c{c}}{\~a}o
natural, o que pode inviabilizar a navega{\c{c}}{\~a}o
aut{\^o}noma. Uma solu{\c{c}}{\~a}o quando os sensores falham
ou n{\~a}o est{\~a}o dispon{\'{\i}}veis {\'e} a
navega{\c{c}}{\~a}o por imagens. Contudo, ainda existem
problemas associados a navega{\c{c}}{\~a}o aut{\^o}noma, por
processamento de imagens. Esses problemas adv{\'e}m da sua
natureza t{\'e}cnica, que {\'e} sens{\'{\i}}vel a
interfer{\^e}ncias de fatores ambientais, como nuvens, chuva,
fuma{\c{c}}a, luminosidade, sensores imageadores de baixa
qualidade ou qualquer outro fator que atrapalhe a coleta de
imagens durante a navega{\c{c}}{\~a}o. Por isso, para mitigar
poss{\'{\i}}veis falhas das alternativas, que utilizam apenas
imagens, a solu{\c{c}}{\~a}o sugerida na literatura {\'e}
aplicar t{\'e}cnicas de fus{\~a}o de dados. Com a fus{\~a}o de
dados, {\'e} poss{\'{\i}}vel obter maior confiabilidade a
navega{\c{c}}{\~a}o aut{\^o}noma de VANTs e desta forma,
garantir uma melhor estimativa da posi{\c{c}}{\~a}o do VANT
durante a navega{\c{c}}{\~a}o. Uma das t{\'e}cnicas que se
destaca, como fusor de dados, {\'e} o Filtro de Kalman (FK).
Contudo, o FK possui desvantagens e dentre elas destaca-se a
complexidade exigida na
constru{\c{c}}{\~a}o/implementa{\c{c}}{\~a}o do FK. O objetivo
desta Tese {\'e} investigar, analisar e qualificar a
estima{\c{c}}{\~a}o da posi{\c{c}}{\~a}o de um VANT, a partir
da fus{\~a}o dos dados, dos sensores embarcados, utilizando uma
MLP autoconfigurada, como alternativa ao FK. Al{\'e}m de fornecer
uma alternativa ao FK, tamb{\'e}m {\'e} um objetivo a
constru{\c{c}}{\~a}o de um dispositivo de processamento em alto
desempenho, um hardware dedicado, para implementar a t{\'e}cnica
investigada num dispositivo pequeno e poss{\'{\i}}vel de
embarcar num VANT de pequeno porte. ABSTRACT: The use of UAVs has
increased in recent years and due to the increase in the use of
UAVs, the interest of companies, governmental organizations and
the scientific community in the development of autonomous aerial
systems, especially small ones, has also grown. Navigation
requires sensors, such as a Global Navigation Satellite System
(GNSS) and an Inertial Navigation System (INS). However, there are
problems associated with the use of these sensors. The main
problems are interferences, which can be caused intentionally or
by action natural, which can make autonomous navigation
unfeasible. One solution when sensors fail or are not available is
image navigation. However, there are still problems associated
with autonomous navigation, due to image processing. These
problems are due to their technical nature, which is sensitive to
interference from environmental factors, such as clouds, rain,
smoke, light, low quality imaging sensors or any other factor that
interferes with image collection during navigation. Therefore, to
mitigate possible failures of the alternatives, which use only
images, the solution suggested in the literature is to apply data
fusion techniques. With the merger of data, it is possible to
obtain greater reliability in the autonomous navigation of UAVs
and in this way, guarantee a better estimate of the UAV position
during navigation. One of the techniques that stands out, as a
data fuser, is the Kalman Filter (FK). However, FK has
disadvantages and among them stands out the complexity required in
the construction / implementation of FK. The objective of this
Thesis is to investigate, analyze and qualify the estimation of
the position of a UAV, from the fusion of data, of the embedded
sensors, using a self-configured MLP, as an alternative to the FK.
In addition to providing an alternative to FK, it is also an
objective to build a high performance processing device, dedicated
hardware, to implement the investigated technique in a small
device and possible to embark on a small UAV.",
committee = "Stephany, Stephan (presidente) and Campos Velho, Haroldo Fraga de
(orientador) and Shiguemori, Elcio Hideiti (orientador) and
Guimar{\~a}es, Lamartine Nogueira Frutuoso and Ramos, Alexandre
Carlos Brand{\~a}o and Braga, Antonio de P{\'a}dua",
englishtitle = "Autonomous UAV navigation by data fusion with optimized artificial
neural network implemented in FPGA.",
language = "pt",
pages = "183",
ibi = "8JMKD3MGP3W34R/4443NUP",
url = "http://urlib.net/ibi/8JMKD3MGP3W34R/4443NUP",
targetfile = "publicacao.pdf",
urlaccessdate = "23 maio 2024"
}