Multi-Level Bayesian Models for Environment Perception

✍ Scribed by Csaba Benedek

• Publisher: Springer
• Year: 2022
• Tongue: English
• Leaves: 208
• Edition: 1st ed. 2022
• Category: Library

✦ Synopsis

This book deals with selected problems of machine perception, using various 2D and 3D imaging sensors. It proposes several new original methods, and also provides a detailed state-of-the-art overview of existing techniques for automated, multi-level interpretation of the observed static or dynamic environment. To ensure a sound theoretical basis of the new models, the surveys and algorithmic developments are performed in well-established Bayesian frameworks. Low level scene understanding functions are formulated as various image segmentation problems, where the advantages of probabilistic inference techniques such as Markov Random Fields (MRF) or Mixed Markov Models are considered. For the object level scene analysis, the book mainly relies on the literature of Marked Point Process (MPP) approaches, which consider strong geometric and prior interaction constraints in object population modeling. In particular, key developments are introduced in the spatial hierarchical decomposition of the observed scenarios, and in the temporal extension of complex MRF and MPP models. Apart from utilizing conventional optical sensors, case studies are provided on passive radar (ISAR) and Lidar-based Bayesian environment perception tasks. It is shown, via several experiments, that the proposed contributions embedded into a strict mathematical toolkit can significantly improve the results in real world 2D/3D test images and videos, for applications in video surveillance, smart city monitoring, autonomous driving, remote sensing, and optical industrial inspection.

✦ Table of Contents

Acknowledgements
Contents
Acronyms and Notations
Abbreviations and Concepts
General Notations Used in the Book
Specific Notations Used in MRF/CXM Models
Specific Notations Used in MPP Models
1 Introduction
2 Fundamentals
2.1 Measurement Representation and Problem Formulations
2.2 Markovian Classification Models
2.2.1 Markov Random Fields, Gibbs Potentials, and Observation Processes
2.2.2 Bayesian Labeling Approach and the Potts Model
2.2.3 MRF-Based Image Segmentation
2.2.4 MRF Optimization
2.2.5 Mixed Markov Models
2.3 Object Population Extraction with Marked Point Processes
2.3.1 Definition of Marked Point Processes
2.3.2 MPP Energy Functions
2.3.3 MPP Optimization
2.4 Methodological Contributions of the Book
3 Bayesian Models for Dynamic Scene Analysis
3.1 Dynamic Scene Perception
3.2 Foreground Extraction in Video Sequences
3.2.1 Related Work in Video-Based Foreground Detection
3.2.2 MRF Model for Foreground Extraction
3.2.3 Probabilistic Model of the Background and Shadow Processes
3.2.4 Microstructural Features
3.2.5 Foreground Probabilities
3.2.6 Parameter Settings
3.2.7 MRF Optimization
3.2.8 Results
3.2.9 Summary and Applications of Foreground Segmentation
3.3 People Localization in Multi-camera Systems
3.3.1 A New Approach on Multi-view People Localization
3.3.2 Silhouette-Based Feature Extraction
3.3.3 3D Marked Point Process Model
3.3.4 Evaluation of Multi-camera People Localization
3.3.5 Applications and Alternative Ways of 3D Person Localization
3.4 Foreground Extraction in Lidar Point Cloud Sequences
3.4.1 Problem Formulation and Data Mapping
3.4.2 Background Model
3.4.3 DMRF Approach on Foreground Segmentation
3.4.4 Evaluation of DMRF-Based Foreground-Background Separation
3.4.5 Application of the DMFR Method for Person and Activity Recognition
3.5 Conclusions
4 Multi-layer Label Fusion Models
4.1 Markovian Fusion Models in Computer Vision
4.2 A Label Fusion Model for Object Motion Detection
4.2.1 2D Image Registration
4.2.2 Change Detection with 3D Approach
4.2.3 Feature Selection
4.2.4 Multi-layer Segmentation Model
4.2.5 L3Mrf Optimization
4.2.6 Experiments on Object Motion Detection
4.3 Long-Term Change Detection in Aerial Photos
4.3.1 Image Model and Feature Extraction
4.3.2 A Conditional Mixed Markov Image Segmentation Model
4.3.3 Experiments on Long-Term Change Detection
4.4 Parameter Settings in Multi-layer Segmentation Models
4.5 Conclusions
5 Multitemporal Data Analysis with Marked Point Processes
5.1 Introducing the Time Dimension in MPP Models
5.2 Object-Level Change Detection
5.2.1 Building Development Monitoring—Problem Definition
5.2.2 Feature Selection
5.2.3 Multitemporal MPP Configuration Model and Optimization
5.2.4 Experimental Study of the mMPP Model
5.3 A Point Process Model for Target Sequence Analysis
5.3.1 Application on Moving Target Analysis in ISAR Image Sequences
5.3.2 Problem Definition and Notations
5.3.3 Data Preprocessing in a Bottom-Up Approach
5.3.4 Multiframe Marked Point Process Model
5.3.5 Multiframe MPP Optimization
5.3.6 Experimental Results on Target Sequence Analysis
5.4 Parameter Settings in Dynamic MPP Models
5.5 Conclusions
6 Multi-level Object Population Analysis with an Embedded MPP Model
6.1 A Hierarchical MPP Approach
6.2 Problem Formulation and Notations
6.3 EMPP Energy Model
6.4 Multi-level MPP Optimization
6.5 Applications of the EMPP Model
6.5.1 Built-in Area Analysis in Aerial and Satellite Images
6.5.2 Traffic Monitoring-Based on Lidar Data
6.5.3 Automatic Optical Inspection of Printed Circuit Boards
6.6 Implementation Details
6.7 Quantitative Evaluation Framework
6.7.1 EMPP Benchmark Database
6.7.2 Quantitative Evaluation Methodology
6.8 Experimental Results
6.8.1 EMPP Versus an Ensemble of Single Layer MPPs
6.8.2 Application Level Comparison to Non-MPP-Based Techniques
6.8.3 Effects on Data Term Parameter Settings
6.8.4 Computational Time
6.8.5 Experiment Repeatability
6.9 Conclusion
7 Concluding Remarks
Appendix References
Index