Dissertation/Thesis Abstract

Unsupervised building detection from irregularly spaced LiDAR and aerial imagery
by Shorter, Nicholas Sven, Ph.D., University of Central Florida, 2009, 168; 3383696
Abstract (Summary)

As more data sources containing 3-D information are becoming available, an increased interest in 3-D imaging has emerged. Among these is the 3-D reconstruction of buildings and other man-made structures. A necessary preprocessing step is the detection and isolation of individual buildings that subsequently can be reconstructed in 3-D using various methodologies. Applications for both building detection and reconstruction have commercial use for urban planning, network planning for mobile communication (cell phone tower placement), spatial analysis of air pollution and noise nuisances, microclimate investigations, geographical information systems, security services and change detection from areas affected by natural disasters. Building detection and reconstruction are also used in the military for automatic target recognition and in entertainment for virtual tourism.

Previously proposed building detection and reconstruction algorithms solely utilized aerial imagery. With the advent of Light Detection and Ranging (LiDAR) systems providing elevation data, current algorithms explore using captured LiDAR data as an additional feasible source of information. Additional sources of information can lead to automating techniques (alleviating their need for manual user intervention) as well as increasing their capabilities and accuracy. Several building detection approaches surveyed in the open literature have fundamental weaknesses that hinder their use; such as requiring multiple data sets from different sensors, mandating certain operations to be carried out manually, and limited functionality to only being able to detect certain types of buildings.

In this work, a building detection system is proposed and implemented which strives to overcome the limitations seen in existing techniques. The developed framework is flexible in that it can perform building detection from just LiDAR data (first or last return), or just nadir, color aerial imagery. If data from both LiDAR and aerial imagery are available, then the algorithm will use them both for improved accuracy. Additionally, the proposed approach does not employ severely limiting assumptions thus enabling the end user to apply the approach to a wider variety of different building types. The proposed approach is extensively tested using real data sets and it is also compared with other existing techniques. Experimental results are presented.

Indexing (document details)
Advisor: Kasparis, Takis
Commitee:
School: University of Central Florida
School Location: United States -- Florida
Source: DAI-B 70/11, Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Electrical engineering, Remote sensing, Computer science
Keywords: Aerial imagery, Air pollution, Building detection, Light detection and ranging, Natural disasters
Publication Number: 3383696
ISBN: 9781109480313
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest