Inhalt

•  
   Abstract
•  
   Zusammenfassung
•  
   Acknowledgements
•  
   Contents
•  
   List of Figures
•  
   List of Tables
•  
  1 Introduction
•  
   1.1 Motivation
•  
   1.2 Background of Proposed Work
•  
  1.3 Objectives of This Thesis
•  
   1.3.1 Vision-Based Trajectory Controller and Mapping
•  
   1.3.2 Visual Detection of Already Cleaned Areas
•  
   1.4 Outline of This Thesis
•  
  2 Cleaning Robots
•  
   2.1 Cleaning Robots as Subdomain of Mobile Service Robots
•  
  2.2 Domestic Floor-Cleaning Robots
•  
   2.2.1 Commercial Products
•  
   2.2.2 Academic Research
•  
   2.3 Relevance for Our Work
•  
  3 Visual Navigation Based on Omnidirectional Images
•  
   3.1 Motivation
•  
  3.2 Introduction
•  
   3.2.1 Definition of Navigation
•  
   3.2.2 Basic Categorization of Navigation Methods
•  
   3.2.3 Prerequisites for Reliable and Robust Navigation
•  
   3.2.4 Omnidirectional Vision
•  
  3.3 Representing and Recognizing Places
•  
   3.3.1 Representations of Places Based on Omnidirectional Visual Information
•  
   3.3.2 Role of Position Information
•  
   3.3.3 Discussion of Place Representation and Recognition
•  
  3.4 Visual Compass
•  
   3.4.1 Definition of Visual Compass
•  
   3.4.2 Literature Review on Visual Compass Methods
•  
   3.4.3 Discussion of Visual Compass Methods
•  
  3.5 Local Visual Homing
•  
   3.5.1 Definition of Local Visual Homing
•  
   3.5.2 Literature Review on Local Visual Homing
•  
   3.5.3 Discussion of Local Visual Homing
•  
  3.6 Map-Based Navigation
•  
   3.6.1 Definition and Common Principles of Map-Based Navigation
•  
   3.6.2 Hierarchical Maps
•  
   3.6.3 Literature Review on Quantitative Maps
•  
   3.6.4 Literature Review on Qualitative Maps
•  
   3.6.5 Discussion of Map-Based Navigation
•  
   3.7 Overall Discussion and Outlook
•  
   3.7.1 Comparability of Navigation Strategies
•  
   3.7.2 Applicability Under Real-World Conditions
•  
   3.7.3 Towards Real-World Applications
•  
  4 Trajectory Controller Based on Partial Pose Estimation and Dense Topo-Metric Maps
•  
   4.1 Introduction
•  
   4.1.1 Mapping for Cleaning Robots
•  
   4.1.2 Introduction to the Proposed Trajectory Controller
•  
   4.2 Dense Topo-Metric Map
•  
  4.3 Trajectory Controller
•  
   4.3.1 First Lane
•  
   4.3.2 Lane Changes
•  
   4.3.3 Lane-Distance Estimator
•  
   4.3.4 Trajectory Controller
•  
  4.4 Experiments and Setup
•  
   4.4.1 Procedure and Used Parameters
•  
   4.4.2 Min-Warping With Compass Acceleration
•  
   4.4.3 Custom-Built Cleaning Robot
•  
   4.4.4 Systematic Error
•  
   4.4.5 Passive Visual Tracking System
•  
   4.4.6 Data Evaluation
•  
  4.5 Results
•  
   4.5.1 Qualitative Analysis
•  
   4.5.2 Inter-Lane Distances
•  
   4.5.3 Cleaning Performance
•  
  4.6 Discussion and Conclusions
•  
   4.6.1 Partial Pose Estimation
•  
   4.6.2 Dense Topo-Metric Maps
•  
   4.6.3 Influence of Image Disturbances
•  
   4.6.4 Implications for Further Cleaning Strategies
•  
   4.6.5 Conclusions
•  
   4.7 Future Working Directions
•  
  5 Holistic Loop-Closure Detection and Visual Compass
•  
   5.1 Introduction
•  
   5.2 Methods
•  
   5.2.1 Image Preprocessing Functions
•  
   5.2.2 Global Image Dissimilarity Functions
•  
   5.2.3 Accelerated Compass Method Operating in the Fourier Domain
•  
   5.2.4 Robustness Against Changes of the Illumination
•  
  5.3 Experiments and Evaluation
•  
   5.3.1 Experimental Procedure
•  
   5.3.2 Evaluation
•  
   5.4 Results of Standard Compass Method
•  
   5.4.1 Performance of Loop-Closure Detection
•  
   5.4.2 Robustness Against Perceptual Aliasing and Perceptual Variability
•  
   5.4.3 Compass Accuracy
•  
   5.4.4 Computational Aspects of Image Preprocessing Functions
•  
   5.4.5 Computational Aspects of Global Image Dissimilarity Functions
•  
   5.4.6 Discussion and Conclusions
•  
  5.5 Results of Accelerated Compass Method
•  
   5.5.1 Performance of Loop-Closure Detection
•  
   5.5.2 Compass Accuracy
•  
   5.5.3 Computational Aspects
•  
   5.5.4 Discussion and Conclusions
•  
  5.6 Overall Discussion and Conclusions
•  
   5.6.1 Comparison Between Standard and Accelerated Compass Methods
•  
   5.6.2 Applicability on a Real Cleaning Robot
•  
   5.7 Future Working Directions
•  
   5.7.1 Efficient Implementation
•  
   5.7.2 Improvement of the Accelerated Compass Variant
•  
   5.7.3 Subdivision of the Panoramic Image Into Rings
•  
   5.7.4 Testing Further Preprocessing and Image-Dissimilarity Functions
•  
  6 Signature-Based Loop-Closure Detection
•  
   6.1 Introduction
•  
   6.2 Methods
•  
   6.2.1 Signature Functions
•  
   6.2.2 Dissimilarity Functions
•  
   6.2.3 Dimensionality of Signatures
•  
   6.2.4 Robustness Against Changes of the Illumination
•  
   6.3 Database Experiments
•  
   6.3.1 Methods
•  
   6.3.2 Results
•  
   6.3.3 Discussion and Conclusions
•  
  6.4 Real-Robot Experiments
•  
   6.4.1 Methods
•  
   6.4.2 Results
•  
   6.4.3 Discussion and Conclusions
•  
   6.5 Overall Discussion and Conclusions
•  
   6.6 Future Working Directions
•  
   6.6.1 Increasing the Robustness Against Illumination Changes
•  
   6.6.2 Integration into a More Complex Control Scheme
•  
   6.6.3 Signature-Based Approaches to Topological Mapping
•  
   6.6.4 Biological Modeling
•  
  7 Overall Summary, Discussion, and Outlook
•  
   7.1 Overall Summary
•  
   7.2 Overall Discussion
•  
   7.2.1 Local Map Consistency vs. Global Map Consistency
•  
   7.2.2 Relation of the Proposed Work to Spatial Cognition
•  
   7.2.3 Towards a Prototype Product
•  
   7.2.4 Alternative Applications for the Proposed Methods
•  
   7.3 Future Working Directions
•  
   7.3.1 Increasing Robustness Against Dynamic Scene Changes
•  
   7.3.2 Towards a Full-Fledged Cleaning Robot Control Scheme
•  
   7.3.3 Hierarchical Mapping Based on the Decomposition Into Cleaning Segments
•  
   7.3.4 Omnidirectional Vision as Multi-Purpose Sensor
•  
   7.3.5 Summary and Appraisal of Future Work
•  
  Appendix
•  
   A Lane Controller With Partial Position Information
•  
  B Panoramic Image Databases
•  
   B.1 Example Images
•  
   B.2 Positions of Image Acquisition
•  
  C Holistic Loop-Closure Detection and Visual Compass
•  
   C.1 Derivation of the Distance Measures Proposed by MOELLER10TR1,MOELLER10TR2
•  
   C.2 Influence of Linearly Transformed Pixel Intensities on Preprocessed Images
•  
   C.3 Influence of Linearly Transformed Pixel Intensities on Image Dissimilarity Functions
•  
   C.4 Further Results
•  
  D Signature-Based Loop-Closure Detection
•  
   D.1 Ground Distance and Relation to `39`42`"613A``45`47`"603AL1-Norm
•  
   D.2 Influence of Linearly Transformed Pixel Intensities on Signature Functions
•  
   D.3 Further Results
•  
  References
•  
   Bibliography
•  
   Internet Sources
•  
   Datasheets and Technical Specifications
•  
   Supplemental Materials