This work focuses on developing two visual servoing algorithms that enable the Coralbot to stabilise itself relative to an area of reef. For this purpose, we present a fast coral reef detector based on supervised machine learning. We extract texture feature descriptors using a bank of Gabor Wavelet filters. We use a database of 621 images of coral reef located in Belize. The Decision Trees algorithm shows a fast execution time among the machine learning algorithms. We use the coral detections to estimate point features and moment features. We use these features through an Image-based approach and a Moment-based approach. We code the coral detector and the visual servoing algorithms in C ++ for obtaining a fast response of the system. We test the system performance through an underwater simulator, UWSim, which is supported by the Robot Operating System, ROS. We obtain promising results using point features instead of moment features.
|Title of host publication||Quantitative Monitoring of the Underwater Environment|
|Number of pages||10|
|Publication status||Published - 2016|
|Name||Ocean Engineering & Oceanography|