DOI: https://doi.org/10.15368/theses.2019.51
Available at: https://digitalcommons.calpoly.edu/theses/2059
Date of Award
6-2019
Degree Name
MS in Computer Science
Department/Program
Computer Science
Advisor
Zoë Wood
Abstract
Almost two thirds of the Earth's surface is covered in ocean, and yet, only about 5% of it is mapped. There are an unknown amount of sunken ships, planes, and other artifacts hidden below the sea. Extensive search via boat and a sonar tow fish following a standard lawnmower pattern is used to identify sites of interest. Then, if a site has been determined to potentially be historically significant, the most common next step is a survey by either a human dive team or remotely operated vehicle. These are time consuming, error prone, and potentially dangerous options, but autonomous underwater vehicles (AUVs) are a possible solution.
This thesis introduces a system for automatically generating paths for AUVs to survey and map shipwrecks. Most AUVs include software to set a lawnmower path for a given region of ocean, and individualized paths can be set via specifying GPS encoded nodes for the AUV to pass through. This thesis presents an algorithm for generating an individualized path that permits the AUV, equipped with a camera to "see" all sides of a region of interest (i.e. a shipwreck). This allows the region of interest to be completely documented. Photogrammetry can then be used to reconstruct a three-dimensional model, but a path is needed to do so. Paths are generated by a probabilistic roadmap algorithm that uses a rapidly-exploring random tree to quickly cover the volume of exploration space and generate small maps with good coverage. The roadmap is constructed out of nodes, each having its own weight. The weight of a given node is calculated using an objective function which measures an approximate view coverage by casting rays from the virtual view and intersecting them with the region of interest. In addition, the weight of a node is increased if this node allows the AUV to see a new side of the region of interest. In each iteration of the algorithm, a node to expand off of is selected based off its location in space or its high weight, a new node with a given amount of freedom is generated, and then added to the roadmap. The algorithm has degrees of freedom in position, pitch, and yaw as well as the objective function to encourage the path to see all sides of the region of interest. Once all sides of the region of interest have been viewed, a path is determined to be complete.
The algorithm was tested in a virtual world where the virtual camera acted as the AUV. All of the images collected from our automatically generated path were used to create 3D models and point clouds using photogrammetry. To measure the effectiveness of our paths versus the pre-packaged lawnmower paths, the 3D models and point clouds created from our algorithm were compared to those generated from running a standard lawnmower pattern. The paths generated by our algorithm captured images that could be used in a 3D reconstruction which were more detailed and showed better coverage of the region of interest than those from the lawnmower pattern.