Abstract

The recent advancements made in the field of computer vision, along with the ever increasing rate of computational power has opened up opportunities in the field of automated photogrammetry. Many researchers have focused on using these powerful computer vision algorithms to extract three-dimensional point clouds of scenes from multi-view imagery, with the ultimate goal of creating a photo-realistic scene model. However, geographically accurate three-dimensional scene models have the potential to be exploited for much more than just visualization. This work looks at utilizing automatically generated scene structure from near-nadir aerial imagery to identify and classify objects within the structure, through the analysis of spatial-spectral information. The limitation to this type of imagery is imposed due to the common availability of this type of aerial imagery. Popular third-party computer-vision algorithms are used to generate the scene structure. A voxel-based approach for surface estimation is developed using Manhattan-world assumptions. A surface estimation confidence metric is also presented. This approach provides the basis for further analysis of surface materials, incorporating spectral information. Two cases of spectral analysis are examined: when additional hyperspectral imagery of the reconstructed scene is available, and when only R,G,B spectral information can be obtained. A method for registering the surface estimation to hyperspectral imagery, through orthorectification, is developed. Atmospherically corrected hyperspectral imagery is used to assign reflectance values to estimated surface facets for physical simulation with DIRSIG. A spatial-spectral region growing-based segmentation algorithm is developed for the R,G,B limited case, in order to identify possible materials for user attribution. Finally, an analysis of the geographic accuracy of automatically generated three-dimensional structure is performed. An end-to-end, semi-automated, workflow is developed, described, and made available for use.

Library of Congress Subject Headings

Computer vision; Computer algorithms; Photogrammetry; Image analysis

Publication Date

3-31-2014

Document Type

Dissertation

Student Type

Graduate

Degree Name

Imaging Science (Ph.D.)

Department, Program, or Center

Chester F. Carlson Center for Imaging Science (COS)

Advisor

Carl Salvaggio

Advisor/Committee Member

David Messinger

Advisor/Committee Member

Nathan Cahill

Comments

Physical copy available from RIT's Wallace Library at TA1634 .N55 2014

Campus

RIT – Main Campus

Plan Codes

IMGS-PHD

Download

Share

COinS