Abstract

In recent years, technological developments have advanced the capabilities of computer vision techniques, including those leveraged toward 3D reconstruction with Structure-from-Motion. This process allows for the generation of accurate 3D models from sets of 2D imagery. With accurate feature detectors and matchers, Structure-from-Motion remains a widely used technique. However, the reconstruction of 3D structure is heavily dependent on the spatial characteristics of the object of interest. Fast-moving, featureless, and dynamic objects suffer from this latter issue, where standard or advanced feature detectors cannot derive any information on the object's surface. With the goal of producing three-dimensional geometric reconstructions of condensed water vapor plumes, this research presents a robust methodology utilizing the space carving technique. Paired with an extensive data collection campaign, convolutional neural networks such as U-Net and Mask R-CNN are assessed and utilized to produce accurate segmentation masks of plumes. This research introduces a comprehensive methodology for transforming Structure-from-Motion models, reconstructed using a hierarchical global-to-local approach, to derive precise camera pose in a metric coordinate frame. Finally, a space carving implementation was utilized to produce scale volumetric reconstructions of condensed water vapor plumes from repeated observations in the visible region of the spectrum. This methodology, implemented on diverse data sets, demonstrates exceptional robustness against varying illumination and meteorological conditions, including challenges posed by the presence of condensed water vapor plumes emanating from mechanical draft cooling towers, a dynamic observation.

Library of Congress Subject Headings

Remote-sensing images--Data processing; Computer vision; Water vapor, Atmospheric; Plumes (Fluid dynamics); Estimation theory; Volumetric analysis

Publication Date

5-2024

Document Type

Dissertation

Student Type

Graduate

Degree Name

Imaging Science (Ph.D.)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science

College

College of Science

Advisor

Carl Salvaggio

Advisor/Committee Member

David Long

Advisor/Committee Member

Emmett Ientilucci

Campus

RIT – Main Campus

Plan Codes

IMGS-PHD

Download

Share

COinS