| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 116.7 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Em missões com o objectivo de captar imagens subaquáticas, a necessidade de uma maior eficiência na obtenção de resultados ´e um problema recorrente. Este projecto tem como objectivo contribuir para a o robustecimento da câmara de plâncton do projecto MarinEye e solucionar os problemas de armazenamento causados pela necessidade de uma elevada resolução nas imagens captadas. Tendo isso em conta, foi desenvolvido um sistema de rápido processamento de imagens responsável pela selecção de informação relevante para recolha. Foram estudados métodos de melhoramento de imagem, extracçãao de características, edge detection e detecção de objectos. Foram também abordados métodos de detecção de objectos e melhoramento de imagem com redes neuronais. Nesta tese foi ainda estudado o estado da arte do campo da visão subaquática no que diz respeito a detecção de objectos e melhoramento de imagem. Esta tese implementa normalização para o melhoramento de imagens e uma cadeia de processos para detecção de objectos que contém canny edge detection, transformações morfológicas, detecção de contornos e detecção de áreas de modo a avaliar conteúdo de potencial interesse para pós processamento. A solução foi implementada em tempo real recorrendo a ROS e atinge 12 frames por segundo no seu estado actual de implementação. Foram realizados testes em ambiente semi-controlado e no rio Douro que validaram a solução desenvolvida. A solução apresentada melhora o contraste de imagens subaquáticas, detecta potenciais objectos de interesse, processa a informação em tempo real e ´e capaz de comunicar com a câmara de plâncton do MarinEye, contribuindo para o Technology Readiness Level deste sistema.
In missions with the objective of gathering underwater images the need for better efficiency when it comes to results is a recurrent problem. This project’s main objective is to contribute to MarinEye’s plankton camera improvement and solve the storage problem caused by the high quality of image needed for this project. Having that in mind it was developed a fast image processing system responsible for selecting relevant information to gather. There was a research regarding methods of image enhancement, feature extraction, edge detection and object detection. There were also studied methods of object detection and image enhancement with neural networks. This thesis explores the state of the art of the field of underwater vision regarding image enhancement and object detection. This thesis implements normalization for image enhancement and a series of processes for object detection like canny edge detection, morphological transformations, contour detection and blob detection in order to evaluate potential areas of interest for post-processing. This solution was implemented in real time with ROS and goes up to 12 frames per second in the current state of implementation. There were conducted tests in a semi-controlled environment and in Douro river which validated the solution presented in this thesis. The solution improves the contrast of underwater images, detects potential areas or objects of interest, processes that information in real time and it is capable of communicating with MarinEye’s plankton camera contributing for the Technology Readiness Level of the system.
In missions with the objective of gathering underwater images the need for better efficiency when it comes to results is a recurrent problem. This project’s main objective is to contribute to MarinEye’s plankton camera improvement and solve the storage problem caused by the high quality of image needed for this project. Having that in mind it was developed a fast image processing system responsible for selecting relevant information to gather. There was a research regarding methods of image enhancement, feature extraction, edge detection and object detection. There were also studied methods of object detection and image enhancement with neural networks. This thesis explores the state of the art of the field of underwater vision regarding image enhancement and object detection. This thesis implements normalization for image enhancement and a series of processes for object detection like canny edge detection, morphological transformations, contour detection and blob detection in order to evaluate potential areas of interest for post-processing. This solution was implemented in real time with ROS and goes up to 12 frames per second in the current state of implementation. There were conducted tests in a semi-controlled environment and in Douro river which validated the solution presented in this thesis. The solution improves the contrast of underwater images, detects potential areas or objects of interest, processes that information in real time and it is capable of communicating with MarinEye’s plankton camera contributing for the Technology Readiness Level of the system.
Description
Keywords
Underwater image enchancement Plankton Object detection