Computer Vision Based Autonomous Panel Intervention for a Remotely Operation Vessel
MetadataShow full item record
- Institutt for marin teknikk 
The Remotely Operated Underwater Vehicle (ROV) has been operating for many years. The use of ROV is increasing with the increased activities in sub-sea missions which are moving towards deep sea. Therefore, the need to improve the autonomy of ROV is be- coming more and more urgent. The purpose of this thesis is to develop a system that can allow this ROV to adjust its pose towards intervention objects and to locate the intervention targets.The common used sensor for ROV intervention is the camera. And vast amount of research and tools already accessible from the field of computer vision. Therefore, it is advantageous to develop a system that utilizes these cameras.This paper compares and analyzes previous research projects on underwater autonomous intervention, and analyze whether this thesis can use the methods of previous projects. The conclusion is new strategies and methods are needed for this thesis, when dealing with a 3D subsea module installed in real underwater environment.A close-range visual guidance algorithm was proposed for ROV posture adjustment. This method utilize the line feature extracted from the camera image. The decision making of ROV motion is based on linear perspective theory. Besides, a label detection algorithm is developed as the starting point of valve intervention. These two algorithms are test with video record, the test result is in line with our expectations.Field trials can be conducted in the future, in order to test the cooperation of algorithms and control systems. There are many works left for the valve intervention, for example the label recognition, valve panel SLAM and manipulator path planning. Besides, due to the existence of marine snow, autonomous dust cleaning may be a interesting topic which is also needed for ROV autonomous intervention architecture.