Detalhes bibliográficos
| Ano de defesa: |
2018 |
| Autor(a) principal: |
SANTOS JÚNIOR, José Guedes dos
 |
| Orientador(a): |
OLIVEIRA JUNIOR, Wilson Rosa de |
| Banca de defesa: |
OLIVEIRA JUNIOR, Wilson Rosa de,
MIRANDA, Péricles Barbosa Cunha de,
TEIXEIRA, João Marcelo Xavier Natário |
| Tipo de documento: |
Dissertação
|
| Tipo de acesso: |
Acesso aberto |
| Idioma: |
por |
| Instituição de defesa: |
Universidade Federal Rural de Pernambuco
|
| Programa de Pós-Graduação: |
Programa de Pós-Graduação em Informática Aplicada
|
| Departamento: |
Departamento de Estatística e Informática
|
| País: |
Brasil
|
| Palavras-chave em Português: |
|
| Área do conhecimento CNPq: |
|
| Link de acesso: |
http://www.tede2.ufrpe.br:8080/tede2/handle/tede2/7865
|
Resumo: |
The term Augmented Reality is used to specify the systems that have a technology of inserting virtual objects in real scenes, allowing an increase in the amount of information present in the former real environment. In the final result of an Augmented Reality scene footage, the degree of naturalness of this insertion is related not only to the rendering quality of the virtual objects, but also to the accuracy with which the pose of the real objects in relation to the camera is known during the footage, that is, it also depends on the quality of the tracking of these objects. Artificial markers can facilitate and increase the quality of object tracking, however in some situations it is not always possible or desirable to manually insert markers in the scene to be tracked. The solution adopted has been to use features present naturally in the objects belonging to the scene, this type of tracking is called markerless tracking. Some markerless tracking techniques use prior knowledge of the objects to be tracked, this is done by obtaining virtual models of those objects in advance. There are several model-based tracking methods, some of which use search and optimization algorithms such as particle filter or particle swarm optimization to evaluate sets of candidate poses during tracking, these methods have shown very good results. When capturing a scene with a common digital camera, there is always an information loss, since – besides point sampling and quantization – the geometric representation of a real object in the camera image plane in each captured frame is always in 2D. However, by using RGB-D sensors it is possible to build 3D point clouds of a scene, allowing to obtain a more accurate representation of the points that belong to real world objects. This way, new 3D object tracking techniques that use features extracted from 3D points clouds, previously inaccessible in 2D images, have been developed, allowing more precise 6 degree of freedom markerless 3D tracking algorithms. In order to contribute with current research related to 6 degree of freedom markerless 3D generic object tracking algorithms, this work proposes the use of particle swarm optimization to handle multiple pose hypotheses during top-down model-based tracking from RGB-D images. GPU processing was utilized with the aim of improving execution time. A series of experiments were performed, which revealed an improvement in accuracy obtained by the proposed tracking method in comparison to other state of the art optimization-based techniques. |
