Abstract:This work presents a process pipeline that addresses the problem of reconstructing surfaces of underwater structures from stereo images and sonar scans collected with a micro-ROV on the islands of Malta and Gozo. Using a limited sensor load, sonar and small GoPro Hero2 cameras, the micro-ROV is able to explore water systems and gather data. As a preprocess to the reconstruction pipeline, a 3D evidence grid is created by mosaicing horizontal and vertical sonar scans. A volumetric representation is then constructed using a level set method. Fine-scale details from the scene are captured in stereo cameras, and are transformed into point clouds and projected into the volume. A raycasting technique is used to trim the volume in accordance with the projected point clouds, thus reintroducing fine details to the rough sonar-generated model. The resulting volume is surfaced, yielding a final mesh which can be viewed and interacted with for archaeological and educational purposes. Initial results from both steps of the reconstruction pipeline are presented and discussed.

Procedural Generation and Rendering of Large-Scale Open-World Environments Ian DunnOpen-world video games give players a large environment to explore along with increased freedom to navigate and manipulate that environment. These requirements pose several problems that must be addressed by a game's graphics engine. Often there are a large number of visible objects, such as all of the trees in a forest, as well as objects comprised of large amounts of geometry, such as terrain. An open-world graphics engine must be able to render large environments at varying levels of detail and smoothly transition between detail levels to provide a believable experience.Often this involves finding a way to both store and generate the large amounts of geometry that represent the environment.In this thesis we present a system for generating and rendering large exterior environments, with a focus on terrain and vegetation. We use a region-based procedural generation algorithm to create environments of varying types. This algorithm produces content that can be rendered at multiple levels of detail. The terrain is rendered volumetrically to support caves, overhangs, and cliffs, but is also rendered using heightmaps to allow for large view distances. Vegetation is implemented using procedurally generated meshes and impostors. The volumetric terrain is editable in real time, which limits our ability to pre-generate or cache large amounts of geometry, and also limits the number of assumptions we can make with regard to visibility.We support a view distance of at least 25 miles in each direction, though distant objects are rendered at low resolution. The heightmap terrain used to achieve this view distance consists of over 360,000 triangles. Our system runs at 180 frames per second on commodity desktop hardware.iv ACKNOWLEDGMENTS Thanks to:• Andrew Guenther, for uploading this template.• My advisor, Dr. Wood, for listening to me talk about terrain generation for five years, and for all her help in making this thesis happen.• The members of the graphics thesis group, for their insight and support: