Publikationsserver der RWTH Aachen University
Integration heterogener raumbezogener Objekte aus fragmentierten Geodatenbeständen
Abstract
dc:descriptionThe task of integrating heterogeneous spatial objects from fragmented geospatial datasets, which can be found in different forms in practice, is systematically analyzed and solved in this thesis. Starting from a use case analysis of geometrical integration, multiple process models for the solution of the integration problem are derived. As three main steps of geometrical integration we identify the correspondence problem for homologous spatial objects from multiple maps, the identification of implicit spatial relations and the computation of a transform that warps heterogeneous source datasets into one consistent target dataset. We define a simple model of a map and formulate geometrical integration as a transformation problem between one idealized true map only known partially and multiple realizations of the map perturbed by random noise and unknown systematic effects. In the statistical model the connection between the true map and its realizations is interpreted as a multivariate spatial random process and further investigated by geostatistical and classical estimation methods. As a main result a model of the distance-dependent relative accuracy (neighborhood accuracy) in spatial datasets is obtained. On the basis of this statistical hypothesis two models for the estimation of the true map, namely intrinsic kriging and collocation, are suggested. As an alternative to the stochastical model we suggest the deterministic model of the topology of the euclidean plane where the connection between the maps is treated as a homeomorphism. The consideration of constraints in form of a bijection for homologous points and nonlinear functions for geometrical constraints leads to the formal definition of the homogenization of maps. An extension of the collocation model allows us to estimate the unknown parameters (coordinates) in the system of the true map and to simultaneously consider geometrical constraints, the linear trend, the nonlinear signal and the random noise. Due to the high density of its design matrix the suggested model is unsuitable for practical applications with mass data. The hybrid approach of Benning appears to be an optimal compromise between the extended collocation model and other alternatives. It has the advantage that statistical least squares methods can be combined with (efficient) deterministic interpolation methods and furthermore leads to a sparse design matrix. By the application of a multi-level nested dissection algorithm a massive reduction of the run times is achieved for the estimation of the unknown coordinates in the hybrid model. The hardest problem in automating geometrical integration is the correspondence problem for the geometries of the participating datasets. As state-of-the-art approach the iterated closest point set algorithm (ICP) is discussed and extended by a robust estimator and various nonlinear transforms. As a fundamental alternative to the ICP algorithm we present a softassign deterministic annealing approach (RPM-TPS) for matching datasets perturbed by nonlinear deformations and random noise. A flexible rule-based procedure is suggested for the creation of a virtual cross-layer topology which enables us to efficiently identify geometrical constraints in multi-layer spatial datasets. Finally we discuss various aspects of geometrical integration in a GIS environment. Conventional and novel techniques for the visualization and interpretation of adjustment results are presented.
Degree
thesis:*- Grantor dc:publisher
- Publikationsserver der RWTH Aachen University
- Year dc:date
- 2005
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kampshoff, Stefan
- Contributors dc:contributor
-
- Benning, Wilhelm
Subjects
dc:subject × 7Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- ger