Share Email Print

Spie Press Book

Sensor and Data Fusion: A Tool for Information Assessment and Decision Making, Second Edition
Author(s): Lawrence A. Klein
Format Member Price Non-Member Price

Book Description

The information in the second edition of this volume has been substantially expanded and updated to incorporate recent approaches to sensor and data fusion, as well as additional application examples. A new chapter about data fusion issues associated with multiple-radar tracking systems has also been added. This chapter includes topics such as sensor registration requirements, Kalman filtering, and a discussion of interacting multiple models. As in the first edition, the book discusses the benefits of sensor fusion that accrue when sensors that operate with different phenomenologies or surveil separate volumes of space are used to gather signatures and data about objects or events in their field of view. Subject matter includes: (1) applications of multiple-sensor systems to vehicular traffic management, target classification and tracking, military and homeland defense, and battlefield assessment; (2) target, background, and atmospheric signature-generation phenomena and modeling; (3) the JDL data fusion model; (4) sensor fusion architectures; and (5) detailed descriptions of algorithms that combine multiple-sensor data from target identity and tracking data fusion architectures. Bayesian, Dempster-Shafer, artificial neural networks, fuzzy logic, voting logic, and passive data association techniques for unambiguous location of targets are among the data fusion techniques that are explored.

Book Details

Date Published: 27 September 2012
Pages: 512
ISBN: 9780819491336
Volume: PM222

Table of Contents
SHOW Table of Contents | HIDE Table of Contents

List of Figures
List of Tables

Chapter 1. Introduction

Chapter 2. Multiple-Sensor System Applications, Benefits, and Design Considerations
2.1 Data Fusion Applications to Multiple-Sensor Systems
2.2 Selection of Sensors
2.3 Benefits of Multiple-Sensor Systems
2.4 Influence of Wavelength on Atmospheric Attenuation
2.5 Fog Characterization
2.6 Effects of Operating Frequency on MMW Sensor Performance
2.7 Absorption of MMW Energy in Rain and Fog
2.8 Backscatter of MMW Energy from Rain
2.9 Effects of Operating Wavelength on IR Sensor Performance
2.10 Visibility Metrics
2.10.1 Visibility
2.10.2 Meteorological range
2.11 Attenuation of IR Energy by Rain
2.12 Extinction Coefficient Values (typical)
2.13 Summary of Attributes of Electromagnetic Sensors
2.14 Atmospheric and Sensor System Computer Simulation Models
2.14.1 LOWTRAN attenuation model
2.14.2 FASCODE and MODTRAN attenuation models
2.14.3 EOSAEL sensor performance modelv
2.15 Summary

Chapter 3. Sensor and Data Fusion Architectures and Algorithms
3.1 Definition of Data Fusion
3.2 Level 1 Processing
3.2.1 Detection, classification, and identification algorithms for data fusion Physical models Feature-based inference techniques Cognitive-based models
3.2.2 State estimation and tracking algorithms for data fusion Search direction Correlation and association of data and tracks
3.3 Level 2, 3, and 4 Processing
3.3.1 Situation refinement
3.3.2 Impact (threat) refinement Database management Interrelation of data fusion levels in an operational setting
3.3.3 Fusion process refinement
3.4 Level 5 Fusion: Human-Computer Interface
3.5 Duality of Data Fusion and Resource Management
3.6 Data Fusion Processor Functions
3.7 Definition of an Architecture
3.8 Data Fusion Architectures
3.8.1 Sensor-level fusion
3.8.2 Central-level fusion
3.8.3 Hybrid fusion
3.8.4 Pixel-level fusion
3.8.5 Feature-level fusion
3.8.6 Decision-level fusion
3.9 Sensor Footprint Registration and Size Considerations
3.10 Summary

Chapter 4. Classical Inference
4.1 Estimating the Statistics of a Population
4.2 Interpreting the Confidence Interval
4.3 Confidence Interval for a Population Mean
4.4 Significance Tests for Hypotheses
4.5 The z-test for the Population Mean
4.6 Tests with Fixed Significance Level
4.7 The t-test for a Population Mean
4.8 Caution in Use of Significance Tests
4.9 Inference as a Decision
4.10 Summary

Chapter 5. Bayesian Inference
5.1 Bayes' Rule
5.2 Bayes' Rule in Terms of Odds Probability and Likelihood Ratio
5.3 Direct Application of Bayes' Rule to Cancer Screening Test Example
5.4 The Monty Hall Problem (Let's Make a Deal!)
5.4.1 Case-by-case analysis
5.4.2 Bayes solution
5.5 Comparison of Bayesian Inference with Classical Inference
5.6 Application of Bayesian Inference to Fusing Information from Multiple Sources
5.7 Combining Multiple-Sensor Information Using the Odds Probability Form of Bayes' Rule
5.8 Recursive Bayesian Updating
5.9 Posterior Calculation Using Multivalued Hypotheses and Recursive Updating
5.10 Enhancing Underground Mine Detection Using Two Sensors Whose Data Are Uncorrelated
5.11 Bayesian Inference Applied to Freeway Incident detection
5.11.1 Problem development
5.11.2 Numerical example
5.12 Fusion of Images and Video Sequence Data with Particle Filters
5.12.1 Particle filter
5.12.2 Application to multiple-sensor, multiple-target imagery
5.13 Summary

Chapter 6. Dempster-Shafer Evidential Theory
6.1 Overview of the Process
6.2 Implementation of the Method
6.3 Support, Plausibility, and Uncertainty Interval
6.4 Dempster's Rule for Combination of Multiple-Sensor Data
6.4.1 Dempster's rule with empty set elements
6.4.2 Dempster's rule with singleton propositions
6.5 Comparison of Dempster-Shafer with Bayesian Decision Theory
6.5.1 Dempster-Shafer-Bayesian equivalence example
6.5.2 Dempster-Shafer-Bayesian computation time comparisons
6.6 Developing Probability Mass Functions
6.6.1 Probability masses derived from known characteristics of sensor data IR sensor probability mass functions Metal detector probability mass functions Ground-penetrating radar probability mass functions Probability mass functions from sensor combinations
6.6.2 Probability masses derived from confusion matrices Formation of travel time hypotheses Confusion matrices Computing probability mass functions Combining probability masses for a selected hypothesis
6.7 Probabilistic Models for Transformation of Dempster-Shafer Belief Functions for Decision Making
6.7.1 Pignistic transferable belief model
6.7.2 Plausibility transformation function
6.7.3 Combat identification example Belief Plausibility Plausibility probability Pignistic probability
6.7.4 Modified Dempster-Shafer rule of combination
6.7.5 Plausible and paradoxical reasoning Proposed solution Resolution of the medical diagnosis dilemma
6.8 Summary

Chapter 7. Artificial Neural Networks
7.1 Applications of Artificial Neural Networks
7.2 Adaptive Linear Combiner
7.3 Linear Classifiers
7.4 Capacity of Linear Classifiers
7.5 Nonlinear Classifiers
7.5.1 Madaline
7.5.2 Feedforward network
7.6 Capacity of Nonlinear Classifiers
7.7 Generalization
7.7.1 Hamming distance firing rule
7.7.2 Training set size for valid generalization
7.8 Supervised and Unsupervised Learning
7.9 Supervised Learning Rules
7.9.1 u-LMS steepest descent algorithm
7.9.2 a-LMS error correction algorithm
7.9.3 Comparison of the u-LMS and a-LMS algorithms
7.9.4 Madaline I and II error correction rules
7.9.5 Perceptron rule
7.9.6 Backpropagation algorithm Training process Initial conditions Normalization of input and output vectors
7.9.7 Madaline III steepest descent rule
7.9.8 Dead zone algorithms
7.10 Other Artificial Neural Networks and Data Fusion Techniques
7.11 Summary

Chapter 8. Voting Logic Fusion
8.1 Sensor Target Reports
8.2 Sensor Detection Space
8.2.1 Venn diagram representation of detection space
8.2.2 Confidence levels
8.2.3 Detection modes
8.3 System Detection Probability
8.3.1 Derivation of system detection and false-alarm probability for nonnested confidence levels
8.3.2 Relation of confidence levels to detection and false-alarm probabilities
8.3.3 Evaluation of conditional probability
8.3.4 Establishing false-alarm probability
8.3.5 Calculating system detection probability
8.3.6 Summary of detection probability computation model
8.4 Application Example without Singleton-Sensor Detection Modes
8.4.1 Satisfying the false-alarm probability requirement
8.4.2 Satisfying the detection probability requirement
8.4.3 Observations
8.5 Hardware Implementation of Voting Logic Sensor Fusion
8.6 Application with Singleton-Sensor Detection Modes
8.7 Comparison of Voting Logic Fusion with Dempster-Shafer Evidential Theory
8.8 Summary

Chapter 9. Fuzzy Logic and Fuzzy Neural Networks
9.1 Conditions under which Fuzzy Logic Provides an Appropriate Solution
9.2 Fuzzy Logic Application to an Automobile Antilock Braking System
9.3 Basic Elements of a Fuzzy System
9.3.1 Fuzzy sets
9.3.2 Membership functions
9.3.3 Effect of membership function widths on control
9.3.4 Production rules
9.4 Fuzzy Logic Processing
9.5 Fuzzy Centroid Calculation
9.6 Balancing an Inverted Pendulum with Fuzzy Logic Control
9.6.1 Conventional mathematical solution
9.6.2 Fuzzy logic solution
9.7 Fuzzy Logic Applied to Multi-Target Tracking
9.7.1 Conventional Kalman-filter approach
9.7.2 Fuzzy Kalman-filter approach
9.8 Scene Classification Using Bayesian Classifiers and Fuzzy Logic
9.9 Fusion of Fuzzy-Valued Information from Multiple Sources
9.10 Fuzzy Neural Networks
9.11 Summary

Chapter 10. Data Fusion Issues Associated With Multiple-Radar Tracking Systems
10.1 Measurements and Tracks
10.2 Radar Trackers
10.2.1 Tracker performance parameters
10.2.2 Radar tracker design issues
10.3 Sensor Registration
10.3.1 Sources of registration error
10.3.2 Effects of registration errors
10.3.3 Registration requirements
10.4 Coordinate Conversion
10.4.1 Stereographic coordinates
10.4.2 Conversion of radar measurements into system stereographic coordinates
10.5 General Principle of Estimation
10.6 Kalman Filtering
10.6.1 Application to radar tracking
10.6.2 State transition model
10.6.3 Measurement model Cartesian stereographic coordinates Spherical stereographic coordinates Object in straight-line motion
10.6.4 The discrete-time Kalman Filter algorithm
10.6.5 Relation of measurement-to-track correlation decision to the Kalman gain
10.6.6 Initialization and subsequent recursive operation of the filter
10.6.7 a-b filter
10.6.8 Kalman gain modification methods
10.6.9 Noise covariance values and filter tuning
10.6.10 Process noise model for tracking manned aircraft
10.6.11 Constant velocity target kinematic model process noise
10.6.12 Constant acceleration target kinematic model process noise
10.7 Extended Kalman Filter
10.8 Track Initiation in Clutter
10.8.1 Sequential probability ratio test
10.8.2 Track initiation recommendations
10.9 Interacting Multiple Models
10.9.1 Applications
10.9.2 IMM implementation
10.9.3 Two-model IMM example
10.10 Impact of Fusion Process Location and Data on Multiple-Radar State Estimation Architectures
10.10.1 Centralized measurement processing
10.10.2 Centralized track processing using single-radar tracking
10.10.3 Distributed measurement processing
10.10.4 Distributed track processing using single-radar tracking
10.11 Summary

Chapter 11. Passive Data Association Techniques for Unambiguous Location of Targets
11.1 Data Fusion Options
11.2 Received-Signal Fusion
11.2.1 Coherent processing technique
11.2.2 System design issues
11.3 Angle-Data Fusion
11.3.1 Solution space for emitter locations
11.3.2 Zero-one integer programming algorithm development
11.3.3 Relaxation algorithm development
11.4 Decentralized Fusion Architecture
11.4.1 Local optimization of direction angle track association
11.4.2 Global optimization of direction angle track association Closest approach distance metric Hinge angle metric
11.5 Passive Computation of Range Using Tracks from a Single-Sensor Site
11.6 Summary

Chapter 12. Retrospective Comments
12.1 Maturity of Data Fusion
12.2 Fusion Algorithm Selection
12.3 Prerequisites for Using Level 1 Object Refinement Algorithms

Appendix A. Planck Radiation Law and Radiative Transfer

Appendix B. Voting Fusion With Nested Confidence Levels

Appendix C. The Fundamental Matrix of a Fixed Continuous-Time System

© SPIE. Terms of Use
Back to Top