Date of Award

12-2002

Level of Access

Open-Access Thesis

Degree Name

Master of Science (MS)

Department

Spatial Information Science and Engineering

Advisor

Anthony Stefanidis

Second Committee Member

Peggy Agouris

Third Committee Member

M. Kate Beard-Tisdale

Abstract

Within a CIS environment, the proper use of information requires the identification of the uncertainty associated with it. As such, there has been a substantial amount of research dedicated to describing and quantifying spatial data uncertainty. Recent advances in sensor technology and image analysis techniques are making image-derived geospatial data increasingly popular. Along with development in sensor and image analysis technologies have come departures from conventional point-by-point measurements. Current advancements support the transition from traditional point measures to novel techniques that allow the extraction of complex objects as single entities (e.g., road outlines, buildings). As the methods of data extraction advance, so too must the methods of estimating the uncertainty associated with the data. Not only will object uncertainties be modeled, but the connections between these uncertainties will also be estimated. The current methods for determining spatial accuracy for lines and areas typically involve defining a zone of uncertainty around the measured line, within which the "actual" line exists with some probability. Yet within the research community, the proper shape of this 'uncertainty band' is a topic with much dissent. Less contemplated is the manner in which such areas of uncertainty interact and influence one another. The development of positional error models, from the epsilon band and error band to the rigorous G-band, has focused on statistical models for estimating independent line features. Yet these models are not suited to model the interactions between uncertainty fields of adjacent features. At some point, these distributed areas of uncertainty around the features will intersect and overlap one another. In such instances, a feature's uncertainty zone is defined not only by its measurement, but also by the uncertainty associated with neighboring features. It is therefore useful to understand and model the interactions between adjacent uncertainty fields. This thesis presents an analysis of estimation and modeling techniques of spatial uncertainty, focusing on the interactions among fields of positional uncertainty for image-derived linear features. Such interactions are assumed to occur between linear features derived from varying methods and sources, allowing the application of an independent error model. A synthetic uncertainty map is derived for a set of linear and aerial features, containing distributed fields of uncertainty for individual features. These uncertainty fields are shown to be advantageous for communication and user understanding, as well as being conducive to a variety of image processing techniques. Such image techniques can combine overlapping uncertainty fields to model the interaction between them. Deformable contour models are used to extract sets of continuous uncertainty boundaries for linear features, and are subsequently applied to extract a boundary of influence shared by two uncertainty fields. These methods are then applied to a complex scene of uncertainties, modeling the interactions of multiple objects within the scene. The resulting boundary uncertainty representations are unique from the previous independent error models which do not take neighboring influences into account. By modeling the boundary of interaction among the uncertainties of neighboring features, a more integrated approach to error modeling and analysis can be developed for complex spatial scenes and datasets.

Share