Centre de documentation
scientifique

Enabling pattern-aware automated map generalization / Stefan Steiniger (2007) 

Public Titre : Enabling pattern-aware automated map generalization Type de document : Thèse/HDR Auteurs : Stefan Steiniger, Auteur ; Robert Weibel, Directeur de thèse ; Dirk Burghardt, Directeur de thèse Editeur : Zurich : Université de Zurich Année de publication : 2007 Importance : 180 p. Format : 21 x 30 cm Note générale : Bibliographie Dissertation zur Erlangung der naturwissenschaftlichen Doktorwürde (Dr. Sc. Nat.) vorgelegt der mathematisch-naturwissenschaftlichen Fakultät der Universität Zürich Langues : Anglais (eng) Descripteur : [Termes IGN] attribut géomètrique [Termes IGN] généralisation cartographique automatisée [Termes IGN] généralisation du bâti [Termes IGN] modèle conceptuel de données localisées [Termes IGN] procédure (document) [Termes IGN] système multi-agents [Vedettes matières IGN] Généralisation Résumé : (Auteur) In manual map generalization the cartographer's work is guided by a few principles such as selection of the essential content to meet the map's purpose, and preservation or accentuation of typical and unusual map elements. For instance in a topographic map for an urban area, urban building structures are considered to be typical elements. An example for an unusual element may be a group of ponds with regular spatial distribution and similar size that belong to a fish farm. The recognition and maintenance of such typical and unusual patterns is accomplished by a trained cartographer in an holistic manner. To automate this complex process it is necessary to transfer and decompose the cartographic knowledge and operations into a computer understandable form. The objective of this thesis is to develop and test an approach that enables the maintenance of object relations and patterns during the automated map generalization process. In response to the drawbacks of existing approaches of maintaining map object relations and patterns, we present several requirements for improved approaches. One of these requirements is that structural knowledge (i.e. knowledge about existing patterns) should be explicitly modeled and attached to the map data, and not hidden in the generalization algorithms. A so-called data enrichment strategy such as this should allow a flexible and pattern-aware control of the generalization process. As a consequence of the flexible control approach we establish the hypothesis that the quality of the generalization result and the efficiency of the generalization process can be improved when the data enrichment strategy is employed. The conceptual framework that we propose consists of five steps: The first step considers the identification of patterns and map object relations. In the second step the patterns are formalized using the relations. Subsequently the relations and patterns are extracted (step 3) and stored (step 4). Finally in step 5 the stored relations are utilized to enable pattern-aware decision making for generalization process control. Associated with these steps are the five research questions of this thesis: 1.) What types of relations exist in maps that can be used to describe patterns? 2.) How can we formalize these relations? 3.) How can we detect these relations? 4.) How can these relations be stored and the data be enriched? 5.) How can we exploit the enriched data for pattern preservation and process optimization? These research questions demand comprehensive answers that can not be elaborated thoroughly within the time frame of a PhD project. Hence, while the first research question is answered comprehensively in this thesis, we have chosen to answer the remaining questions with respect to two case studies that serve as a proof of concept of the 5-step framework. The first case study concentrates on the extraction and exploitation of urban structures such as inner city areas, urban areas, suburban areas, etc. In the second case study we aim to identify groups of islands. The contributions of this thesis to map generalization research are essentially associated with the research questions. In response to the first research question we established a comprehensive typology of so-called horizontal relations (and patterns) that we derive from an analysis of topographic maps, thematic maps, and the cartographic literature. With respect to the second question we show for both case studies how identification and formalization of patterns by use of horizontal relations can be accomplished. For the formalization of the island groups, which have been identified in a 'pencil and paper' experiment, we could utilize the Gestalt principles established by Max Wertheimer. To detect the urban structures (the third research question) we developed a supervised classification approach. For the recognition of large island groups formed by the perceptual principle of proximity, we developed an approach that utilizes a minimum spanning tree. The storage of relations, addressed by the fourth research question, has not been discussed in detail, but we use a graph structure and attribute values in the case studies. Finally we discussed for the islands example how relations can be exploited (the fifth research question). In order to evaluate the hypothesis, practical experiments have been conducted with expert generalization rules that account for the urban structure classification of buildings. We obtained an improvement in quality of the generalization result but could not clearly identify a gain in generalization efficiency. However, by accomplishing all five steps of the framework, we show its applicability and utility for the preservation of spatial patterns and relations during the map generalization process. Based on the results and open problems that we discovered in our research, we identify three areas of future map generalization research: 1.) the further formalization and detection of relations and patterns, 2.) the revision and development of constraints to control the preservation of patterns, and 3) research on human computer interaction methods and tools to define and confirm patterns, and control the entire map generalization process more flexibly. Note de contenu : Synthesis 1 Introduction 1.1 The Motivation for Pattern-Aware Map Generalization 1.1.1 Two Examples of Pattern-Aware Manual Map Generalization 1.1.2 Problem Definition 1.1.3 Patterns and Pattern-Aware Map Generalization 1.2 Objective, Methodology and Research Questions 1.3 Structure of the Thesis 2 Theoretical Background on Automated Map Generalization 2.1 Decomposing Manual Map Generalization for Automation 2.1.1 Cartographic Principles 2.1.2 Cartographic Knowledge Acquisition to Achieve a Decomposition 2.1.3 Cartographic Requirements 2.1.4 Cartographic Operations 2.1.5 Conceptual Map Generalization Models - Process Oriented Models - Hierarchical Modeling 2.2 Approaches to Automated Map Generalization 2.2.1 Interactive Systems and Rule-Based Systems 2.2.2 From Rules to Constraints 2.2.3 Constraint-based Automated Map Generalization using Workflow Systems, Multi Agent Systems and Optimization 3 State of the Art in Spatial Pattern Analysis and Emerging Research Challenges 3.1 Spatial Pattern Analysis in Related Disciplines 3.2 Spatial Pattern Analysis and Data Enrichment in Map Generalization 3.2.1 Topographic Maps - Analysis of Building Configurations - Analysis of Polygon Configurations - Analysis of Networks - Analysis of Lines 3.2.2 Thematic Maps 3.3 Research Challenges Addressed in this Thesis 4 Summary of Papers 4.1 Research Paper 1: Exploring Object Relations in Maps 4.1.1 Objectives 4.1.2 Methods and Results 4.1.3 Contributions 4.2 Research Paper 2: Identifying Urban Structures 4.2.1 Objectives 4.2.2 Methods and Results 4.2.3 Contributions 4.3 Research Paper 3: Use of Detected Urban Structures to Control Map Generalization 4.3.1 Objectives 4.3.2 Methods and Results 4.3.3 Contributions 4.4 Research Paper 4: Detecting Large Island Groups within an Archipelago 4.4.1 Objectives 4.4.2 Methods and Results 4.4.3 Contributions 5 Discussion 5.1 Revisiting the Research Questions 5.1.1 What types of relations exist in maps that can be used to describe patterns? 5.1.2 How can we formalize relations and patterns? 5.1.3 How can we detect relations and patterns? 5.1.4 How can relations be stored and the data be enriched? 5.1.5 How can we exploit the enriched data for pattern preservation and process optimization? 5.2 Evaluating the Hypothesis 6 Conclusions and Perspectives 6.1 Main Contributions 6.2 Summarized Research Needs and Outlook References Numéro de notice : 13567 Affiliation des auteurs : non IGN Thématique : GEOMATIQUE/INFORMATIQUE Nature : Thèse étrangère Note de thèse : Dissertation zur Erlangung der naturwissenschaftlichen Doktorwürde (Dr. Sc. Nat.) vorgelegt der mathematisch-naturwissenschaftlichen Fakultät der Universität Zürich : 2007 Organisme de stage : COGIT (IGN) & Ordnance Survey Great Britain nature-HAL : Thèse DOI : sans Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=45204

Documents numériques

peut être téléchargé 13567_these2007_steiniger.pdf Adobe Acrobat PDF