UCLA Extension

Multitarget/Multisensor Data Fusion Techniques for Target Detection, Classification, and State Estimation

This course describes sensor and data fusion methods that improve the probability of correct target detection, classification, identification, and state estimation. These techniques combine information from collocated or dispersed sensors that utilize either similar or different technologies to generate target signatures or imagery. Topics include the effects of the atmosphere and countermeasures on millimeter-wave and infrared sensors to illustrate how the use of different phenomenology-based sensors enhances a data fusion system.

This course introduces the JDL data fusion model as well as several methods for describing sensor and data fusion architectures. Instruction discusses data fusion algorithm taxonomies and a general description of the algorithms and methods used for detection, classification, identification, and state estimation and tracking. This is followed by consideration of situation and threat assessment. Subsequent sections of this course more fully develop the classical inference, Bayesian, Dempster-Shafer, voting logic, artificial neural network, and fuzzy logic data fusion algorithms. Additional topics include radar tracking system design considerations, multiple sensor registration, track initiation in clutter, Kalman filtering, interacting multiple models, and data fusion maturity as it affects real-time tracking.

Examples demonstrate the advantages of multisensor data fusion in systems that use microwave and millimeter-wave detection and tracking radars, laser radars (imagery and range data), and forward-looking IR sensors (imagery data). You can also apply many of the data fusion techniques when combining information from almost any grouping of sensors as long as they can supply the input data required by the fusion algorithm.

Course Materials

Participants enhance their understanding of the:

  • Application of modern sensors to sensor and data fusion
  • Advantages of multisensor data fusion for object discrimination and state estimation
  • Multisensor data fusion principles, algorithms, and architectures that enable the assessment of new and existing systems
  • Taxonomies for target detection, classification, identification, and tracking algorithms
  • Skills needed to develop and apply data fusion algorithms to more complex situations
  • Practical applications

The course benefits:

  • Engineers, scientists, managers, designers, military operations personnel, and other users of multisensor data fusion for target detection, classification, identification, and tracking
  • Those interested in selecting appropriate sensors for specific applications and applying data fusion techniques to advanced dynamic systems, such as classification of airborne targets, ground-based targets, and underwater targets
  • Developers and users of real-time algorithms for intelligent machine development and multiple sensor technologies for non-cooperative target recognition

Recommended Prerequisite

There are no specific course prerequisites; however, a general background in electrical engineering, electro-optics, mathematics, or statistics is recommended for a better understanding of the concepts presented in the course.

Course Materials

Participants receive the text Sensor and Data Fusion: A Tool for Information Assessment and Decision Making, 2nd Edition, Lawrence A. Klein (SPIE, PM 222, 2012) and lecture notes on the first day of the course. These notes are for participants only and are not for sale or unauthorized distribution.

Coordinator and Lecturer

Lawrence A. Klein, PhD, a consultant specializing in developing multiple sensor concepts for tactical and reconnaissance military applications, millimeter-wave and infrared sensors for homeland defense, and sensor and data fusion concepts for intelligent transportation systems. While at Hughes Aircraft Company, Dr. Klein developed missile deployment strategies and sensors used in missile guidance. As a systems manager at Aerojet ElectroSystems, he was responsible for the conceptual design and execution of programs that integrated active and passive millimeter-wave and infrared multispectral sensors in satellites and smart “fire-and-forget” weapons. He was the program manager of three Manufacturing Methods and Techniques projects that lowered the cost of millimeter-wave integrated circuits. At Honeywell, he developed passive millimeter-wave midcourse missile guidance systems and millimeter-wave sensors for mine applications.

In addition to the course text, Dr. Klein is the author of Millimeter-Wave and Infrared Multisensor Design and Signal Processing (Artech House, 1997), which describes multisensor applications, design, and performance; and Sensor Technologies and Data Requirements for ITS (Artech House, 2001), which discusses sensor applications to traffic and transportation management systems. He collaborated with colleagues to prepare a review of data fusion methods that emphasize fusion of image features that aid object tracking using particle filtering, a Bayesian technique. It appears as “Sensor and Data Fusion: Taxonomy, Challenges, and Applications” in the Handbook on Soft Computing for Video Surveillance (Francis and Taylor, 2011). Dr. Klein received his PhD in electrical engineering from New York University in 1973. He is a past reviewer for the IEEE Transactions on Antennas and Propagation, IEEE Transactions on Geoscience and Remote Sensing, and IEEE Transactions on Aerospace and Electronic Systems.

Program

Defense Applications of Multisensor Systems and Data Fusion

  • Need for smart sensors
  • Defining detection, classification, and identification

Sensor Systems

  • Multiple sensor phenomenology
  • Military data fusion architectures
  • Benefits of multiple sensor systems
  • Atmospheric and obscurant effects on IR and MMW sensors
  • Influence of sensor application on choice of MMW frequency and IR wavelength band

Sensor and Data Fusion—What is It?

  • Levels 0, 1, 2, 3, 4, and 5 fusion processing
  • Duality of data fusion and resource management
  • Data fusion architectures

Taxonomy of Detection, Classification, and Identification Data Fusion Algorithms

  • Physical models
  • Feature-based inference: parametric (classical inference, Bayesian, Dempster-Shafer, and others); information-theoretic models (templates, artificial neural networks, clusters, voting, figures of merit, pattern recognition, and others)
  • Cognitive-based (knowledge-based expert systems, fuzzy set theory, and others)

Taxonomy of State Estimation and Tracking Data Fusion Algorithms

  • Search direction
  • Correlation and association of data and tracks:
  • Data alignment
  • Data and track association (prediction gates, correlation metrics, data, and track-to-track association)
  • Position, kinematic, and attribute estimation

Fusion Levels 2, 3, 4, and 5

  • Situation Assessment (Level 2)
  • Threat Assessment (Level 3)
  • Refinement of the data fusion process (Level 4)
  • Multiple-level data fusion architectures—examples

Classical Inference and Decision Theory

  • Confidence interval
  • Sample size for a desired margin of error
  • Decision theory (choosing between 2 hypotheses) and significance tests
  • Statistical significance
  • z-test for a population mean
  • t-test for a population mean
  • 1- and 2-sided tests
  • Type 1 and Type 2 errors
  • Power of a test

Bayesian Inference

  • Conditional probabilities
  • Bayes’s rule with illustrative examples
  • Comparison of Bayesian and classical inference
  • Bayesian inference fusion process with multiple sensor data
  • Recursive updating of posterior probabilities
  • Multispectral sensor example
  • Mine detection example
  • Freeway incident detection example

Dempster-Shafer Evidential Reasoning

  • Dempster-Shafer fusion process
  • Probability mass
  • Uncertainty interval
  • Dempster’s rule
  • Comparison with Bayesian inference
  • Probability mass function origination examples
    – Using known characteristics of data gathered by the sensors
    – Using confusion matrices
  • Modifications of D-S theory with examples
    – Pignistic transferable belief, plausibility transform, plausible and paradoxical reasoning, and other modifications of the original theory

Artificial Neural Networks

  • Linear classifiers
    – Adaptive linear combiner
    – Capacity of linear classifiers
  • Nonlinear classifiers
    – Madaline
    – Feedforward networks
    – Capacity of nonlinear classifiers
  • Generalization
  • Learning rules
    – Linear mean square algorithms
    – Perceptron rule
    – Back propagation algorithm

Fuzzy Logic

  • Fuzzy logic definitions and applications
  • Elements of fuzzy systems
    – Fuzzy sets
    – Membership functions
    – Production rules
  • Fuzzy logic processing and examples
    – Inverted pendulum
    – Influence of fuzzy set widths and slopes
    – Fuzzy Kalman filter and scene classifier
  • Fuzzy neural network examples

Voting Logic

  • Definition
  • Detection modes and confidence levels
  • System detection and false alarm probabilities
  • Relation of detection and false alarm probabilities to confidence levels
  • False alarm probability selection
  • 3-sensor examples

Radar Tracking System Design Considerations

  • Measurements versus tracks
  • Radar characteristics
  • Tracker functional block diagram
  • Measures of quality for tracking
  • Critical factors that determine performance
  • State space and coordinate conversion

Multiple Sensor Registration

  • Multiple sensor data fusion
    – Tracks or measurements?
    – Centralized or distributed?
  • Sensor registration: a prerequisite for data fusion
  • Potential multisensor architectures for data processing
  • Track correlation
  • Track fusion

Track Initiation in Clutter

  • What is the initiation problem?
  • The Sequential Probability Ratio Test (SPRT)
    – Definition of the decision criteria
    – Example application of the SPRT
    – Properties of the SPRT
  • Recommendations

Introduction to Kalman Filtering

  • What does it do?
  • Kalman filter equations
  • Kalman gain
  • Determining process noise

Interacting Multiple Models (IMM)

  • When are these used?
  • IMM process
  • Components of a model

Multiple-Sensor Tracking Architectures and Data Fusion Maturity

  • Architecture selection
  • Selection of data fusion algorithms and techniques
  • Ancillary support functions for data fusion and data processing requirements
  • Key outputs of tracking and data fusion systems

For more information contact the Short Course Program Office:
shortcourses@uclaextension.edu | (310) 825-3344 | fax (310) 206-2815

image_print