Three-dimensional underwater acoustic image interpretation for ROV navigation

This thesis proposes a practical processing system for 3D underwater acoustic camera images which can assist underwater remotely operated vehicle (ROV) navigation in a semi-unstructured environment, i.e. an environment which includes man-made structures in an otherwise unstructured natural environment. For this research it is assumed that the ROV is moving around or within a rig structure composed of connected tubular elements. In order to achieve the objective of assisting ROV navigation, a cylinder detection framework for 3D underwater acoustic images is suggested. Cylinders are extracted from the 3D acoustic camera image and are matched to an a priori known model of the rig structure and the estimated global position is visualised in a virtual environment display. Methods for processing 3D acoustic camera images, from pre-processing to model matching, have been investigated and the integration of those in a formal description of a cylinder detection framework based on the theory of multilevel, hierarchical systems has been suggested. Conventional methods from intensity and range image processing and image understanding have been analysed for use with 3D acoustic camera images, and adaptations for use with this kind of data have been made where applicable. The main contributions of this thesis include a novel 3D pre-processing filter, additions to line and line segment detection methods, the system description using concepts from the theory of multi-level, hierarchical systems, and extensive evaluation tests using synthetic and real data