Publicação
Computer vision component to environment scanning
| Resumo: | Computer vision is usually used as the perception channel of robotic platforms. These platforms must be able of visually scanning the environment to detect specific targets and obstacles. Part of detecting obstacles is knowing their relative distance to robot. In this work different ways of detecting the distance of an object are analyzed and implemented. Extracting this depth perception from a scene involves three different steps: finding features in an image, finding those same features in another image and calculate the features’ distance. For capturing the images two approaches were considered: single cameras, where we capture an image, move the camera and capture another, or stereo cameras, where images are taken from both cameras at the same time. Starting by SUSAN, then SIFT and SURF, these three feature extraction algorithms will be presented as well as their matching procedure. An important part of computer vision systems is the camera. For that reason, the procedure of calibrating a camera will be explained. Epipolar geometry and the fundamental matrix are two important concepts regarding 3D reconstruction which will also be analyzed and explained. In the final part of the work all concepts and ideas were implemented and, for each approach, tests were made and results analyzed. For controlled environments the relative distance of the objects is correctly extracted but with more complex environment such results are harder to obtain. |
|---|---|
| Autores principais: | Soares, Pedro Emanuel Pereira |
| Assunto: | Computer Vision 3D reconstruction Visão por computador Reconstrução 3D |
| Ano: | 2012 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Computer vision is usually used as the perception channel of robotic platforms. These platforms must be able of visually scanning the environment to detect specific targets and obstacles. Part of detecting obstacles is knowing their relative distance to robot. In this work different ways of detecting the distance of an object are analyzed and implemented. Extracting this depth perception from a scene involves three different steps: finding features in an image, finding those same features in another image and calculate the features’ distance. For capturing the images two approaches were considered: single cameras, where we capture an image, move the camera and capture another, or stereo cameras, where images are taken from both cameras at the same time. Starting by SUSAN, then SIFT and SURF, these three feature extraction algorithms will be presented as well as their matching procedure. An important part of computer vision systems is the camera. For that reason, the procedure of calibrating a camera will be explained. Epipolar geometry and the fundamental matrix are two important concepts regarding 3D reconstruction which will also be analyzed and explained. In the final part of the work all concepts and ideas were implemented and, for each approach, tests were made and results analyzed. For controlled environments the relative distance of the objects is correctly extracted but with more complex environment such results are harder to obtain. |
|---|