| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.39 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A necessidade de informação externa, de referência, acerca da atitude e posição de um
dado robô constitui a motivação da presente dissertação, onde foram desenvolvidos
algoritmos para auxílio na navegação, ou simplesmente para aferição e validação da
resposta do sistema autónomo ao meio envolvente. O principal contributo passa pelo
desenvolvimento de um sistema de posicionamento externo multi-câmara, que determina a
localização e atitude de sistemas robóticos ou objetos, com base em marcadores óticos
ativos.
Através da caracterização dos sistemas robóticos e dos cenários de atuação, percebe-se a
necessidade de desenvolver um sistema de visão constituído por multi-câmaras para
endereçar as situações de oclusão, aumentar a cobertura espacial e potenciar a qualidade
dos resultados de posicionamento.
Assim é proposta uma arquitetura para um sistema global dirigido aos vários requisitos
identificados. A utilização de múltiplas câmaras e objetos de interesse munidos de um
conjunto de marcadores ativos, torna viável o seguimento em tempo real destes objetos, em
ambiente terrestre, aéreo ou subaquático.
Esta arquitetura do sistema global, é demonstrada através de um sistema composto por três
câmaras e uma plataforma de quatro Light-Emitting Diodes (LEDs) para validação de
múltiplos módulos de software, nomeadamente: identificação e validação de pontos de
interesse em imagens; cálculo da posição tridimensional dos marcadores através da
combinação de pares de câmaras stereo e geometria 'multi view'; seleção de resultados
mais precisos; cálculo da atitude do alvo.
Para validação do sistema implementado, foram realizados ensaios experimentais que
demonstram o correto funcionamento dos vários módulos do sistema, para diversas
configurações e condicionantes. Simultaneamente, instalaram-se dois dispositivos
comerciais (Faro e Pixhawk) para aferição e comparação de resultados. Os resultados experimentais mostraram uma clara vantagem do sistema de posicionamento
multi-câmara face ao stereo, quer em qualidade de informação de posicionamento obtida,
quer nos aspetos de cobertura espacial e oclusões.
This dissertation aims to contribute to development Groundtruth System of Multicamera Vision that is also to determinate the pose of robotic systems or objects based on active optical markers. The focus of this thesis is the absence of external information about the target’s attitude to correct navigation algorithms or simply to validate the answer of autonomous system to the surrounding environment. Through the description of the robotic systems in site, it was noticed there was a need to develop a vision system constituted by multi cameras, to address occlusion situations, increase the floor space and up-level the results accuracy. Highly motivated, it is proposed architecture for a global system facing the multiple identified requirements. By using multiple cameras and objects of interest with several active markers, it has become possible to follow objects in real time different environments such as land, air or underwater. Regarding about of global architecture system, it is presented a contribution through a system composed by three cameras and a platform with four Light-Emitting Diodes (LEDs) to validate the points of interest on images. Through this process, it is also possible to measure the tridimensional position of the markers using stereo cameras pairs with different combinations and multi view triangulation, in order to obtain a precise selection of results and targets attitude measurement. In order to validate the implemented system, it were performed several trials showing that the system multiple modules converge to the ideal results under various conditions. Simultaneously, two commercial devices were set up and used (Faro Focus and Pixhawk) to compare results of position and attitude.
This dissertation aims to contribute to development Groundtruth System of Multicamera Vision that is also to determinate the pose of robotic systems or objects based on active optical markers. The focus of this thesis is the absence of external information about the target’s attitude to correct navigation algorithms or simply to validate the answer of autonomous system to the surrounding environment. Through the description of the robotic systems in site, it was noticed there was a need to develop a vision system constituted by multi cameras, to address occlusion situations, increase the floor space and up-level the results accuracy. Highly motivated, it is proposed architecture for a global system facing the multiple identified requirements. By using multiple cameras and objects of interest with several active markers, it has become possible to follow objects in real time different environments such as land, air or underwater. Regarding about of global architecture system, it is presented a contribution through a system composed by three cameras and a platform with four Light-Emitting Diodes (LEDs) to validate the points of interest on images. Through this process, it is also possible to measure the tridimensional position of the markers using stereo cameras pairs with different combinations and multi view triangulation, in order to obtain a precise selection of results and targets attitude measurement. In order to validate the implemented system, it were performed several trials showing that the system multiple modules converge to the ideal results under various conditions. Simultaneously, two commercial devices were set up and used (Faro Focus and Pixhawk) to compare results of position and attitude.
Description
Keywords
Robótica Ground Truth Visão computacional Câmaras Marcadores ativos Triangulação Stereo Geometria Multi view Posicionamento 3D Atitude Correspondência de pontos Robotic Ground truth Computer Vision Cameras Active Markers Stereo Triangulation Multi View Geometry 3D Position Attitude points correspondence