|dc.description.abstract||Remotely Operated Underwater Vehicles (ROVs) have represented a revolution by replacing divers. But now robotics can increase the efficiency of underwater operations in a new step of technology. This is mainly due to its low efficiency and high operational cost. As the new technologies gradually reforms, new equipment, and software have been applied and mounted on ROV that makes it more adaptive to the harsh underwater environment. Hence, instead of just putting it aside, it is now been used frequently and plays a decisive role in the current industries.
There is large potential for underwater operation by using the ROV. Not so long time ago, those operations were conducted by on-board control by humans. It is now possible, both theoretically and technically, to develop a system that enables the ROV's operations in autonomous modes. Obvisouly, if such modes are implemented, the ROV is able to propagate to the desired destination without human intervention. Such processes include systems like obstacle avoidance, object detection, and object tracking.
This thesis aims at developing an auto-system that would be able to detect a desired structure of interest (SOI) and tracks it in subsea while it ensures no collision between obstacles and the ROV during the subsea mission. This auto-system cooperating with a camera which is the most common sensor for ROV. However, the camera is replaced by a sonar sensor for safety and long distance considerations in this case, since the sonar explore longer detection range. Correspondingly, the ROV has longer time to react the existence of obstacles in subsea. \\
The challenges exist on benefiting the sonar as well. Firstly, the sonar sensor that mounts in the ROV only provides data onto 2D frame that generates a sonar map. Secondly, the sonar generated frame often contains big amount of noises and uncertainty. Thirdly, the SOI displays in the sonar frame has a certain gap compared with the material object, which increases difficulty in designing the system. \\
The system is divided into three parts in order to achieve it. The first part is to implement detection of the desired object of usage of sonar frames. The next part is designing a guidance system to track the SOI. The last part is to achieve obstacles avoidance system while the first two parts are conducting in the process. These parts are using computer vision methods and partial machine learning to gain high-level understating of images.
With trial and error in collecting appropriate computer vision methods with respect to functioning time and operational cost, Speeded up robust features (SURF) algorithm is used in sonar detection of desired SOI. Moreover, line of sight (SOI) tracking is utilized to implement a real-time tracking of the desired SOI. The method uses in Obstacle avoidance is depending on a shadow detection that provides proper information about the location of obstacles in the sonar map. To avoid collision between obstacles and the ROV, enhanced vector field histogram method is employed.
The completed design system has been tested by developing a simulation environment for the ROV through labVIEW which is a software for test, measurement and control applications. Through LabVIEW the functionality of the designed system includes detection, tracking, and obstacle avoidance is able to be examined. The result is expected.||en