Final Long Program

Monday - Tuesday - Wednesday - Thursday

Monday July 10

Session MoA1: Plenary talk of Prof. David Schum

Information Fusion and Inference Networks: Evidential Foundations

Prof. David Schum, George Mason University, USA

Abstract: Devices frequently employed in the fusing of information in many situations come in the form of complex inference networks. The construction and analysis of inference networks have a surprisingly long history, dating back to 1913 in the work of an American legal evidence scholar named John H. Wigmore. Methods for probabilistic analyses of complex inference networks have a more recent history and now form an area of vigorous research. Many of the current strategies for analyzing inference networks rest on extensions of Bayes's rule and are collectively referred to as Bayes's Nets. Inference networks can take many forms and can capture an assortment of probabilistic interactions or nonindependencies among the variables represented on an inference network. Most of the current work on Bayes's Nets has involved the development of algorithms for the efficient propagation of probabilities throughout a network as new evidence arrives. But not so much attention has been paid to the fact that there are many logically distinguishable and recurrent forms and combinations of evidence that can serve to activate an inference network. Different forms of evidence require different methods for establishing the credibility or believability of evidence. This is a most important step in the fusion of evidence since credibility-related considerations form the very foundation for all subsequent arguments based on evidence. Part of my talk involves how these important credibility-related foundations are established in the analysis of inference networks. It is also true that evidence performs different roles in the analysis of inference networks. Some evidence directly instantiates nodes or probabilistic variables on an inference network. Such evidence is said to be directly relevant evidence. But other evidence, termed indirectly relevant or ancillary evidence, serves to justify the probabilistic strength of the arcs or linkages on an inference network. There is some controversy at present about the role of ancillary evidence in the analysis of Bayes's Nets that I will also address. In the process, I will show how many of the original insights Wigmore had 1913 about the construction of inference networks deserve more serious consideration than they are in fact receiving today.

Session MoB1: Plenary Panel Discussion

FUSION, Vision and Challenges

Organized by Rabinder Madan, Office of Naval Research, Arlington, USA,
James Myers, Ballistic Missile Defense Organization, USA,
Ivan Kadar, Consultant, Northrop Grumman Corporation, USA

Participants: Dr. Rabinder N. Madan - USA, Dr. Gabriel Ruchet - France, Lt. Col. Dr. James Myers - USA, Dr. Ben Wynne - UK, Dr. David Kleinman - USA, Dr. Jean-Pierre Le Cadre - France, Dr. Sebastiano Serpico - Italy, Dr. William D. Blair - USA, Dr. Anthony Hyder - USA, Dr. Ivan Kadar USA

Session MoC1: Target Detection and Recognition - 1

Chair: Pramod K. Varshney, Syracuse University, NY, USA
Co-Chair: Amy L. Magnus, Intelligent Information AFRL/IFTD, NY, USA

1 - A Suboptimum Permutation Test for Radar Detection in Log-Normal Clutter Environments

Francisco Alvarez-Vaquero, José L. Sanz-González, Universidad Politécnica de Madrid, Spain

Abstract:In this paper we have used a coherent log-normal model for the radar clutter: the in-phase and quadrature components of clutter have been modeled to give a log-normal amplitude distribution and a Gaussian distribution of the phase. We have compared this model with another one in which the distribution of the phase is uniformly distributed. Also, we present detectability curves of the permutation test under log-normal noise environments and different types of target models (nonfluctuating, Swerling I and Swerling II). We analyze the detector performance in terms of detection probability (Pd) versus signal-to-noise ratio (SNR) for different parameter values: the integrated pulse number N, the noise reference samples M, and the false alarm probability (Pfa). Finally, we show computer simulation results for correlated clutter, discussing the meaning of those simulations.

2 - Target Classification by Autoregressive Modeling Using Range Extent Profiles

Mahendra K. Mallick, Stefano Coraluppi, Alphatech, Inc., Burlington, MA, USA

Abstract: We present a novel algorithm based on autoregressive (AR) modeling for analyzing the separability and classification of ground target models using range extent data. Given a range extent profile of a target, we estimate the appropriate model order for the data using the Akaike information criterion (AIC). This prevents under-fitting or over-fitting of the data. Previous researchers have shown that even if the actual process is not an AR process, the AR model serves as a reasonable model for a wide class of practical problems. Error modeling for the range extent data is extremely difficult due to the complex nature of the scattering process, uncertainties in the channel, sensor state, target dynamics, and estimation of the range extent from a range profile. Therefore, our data driven approach serves as a useful algorithm for analyzing target model separability and classification. We apply the algorithm to simulated range extent data and obtain good classification results. We plan to test the algorithm further with real range extent data.

3 - On Parametric Detection of Small Targets in Sea Clutter

H.Einar Wensink, Hollandse Signaalapparaten B.V., Hengelo, The Netherlands

Abstract: A new algorithm is presented that instantaneously estimates the clutter characteristics in the environment of the radar cell that is processed. No a priori information is used, since the algorithm operates directly on the data of the incoming burst. It is a technique that adapts continuously and instantaneously to the environment. The algorithm, after having identified the sea clutter, rejects this sea clutter from the data; thus enhancing the probability of detecting small objects in a clutter environment. The final step is the actual detection that makes use of the advantages of the parametric representation. This results in a lower rate of false detections and as a consequence the later stages, like clustering and tracking, receive a more accurate input. The technique behind this algorithm uses recent developments in parametric time series analysis and performs well in suppressing sea as well as land clutter.

4 - Classification and Feature Selection with Fused Conditionally Dependent Binary Valued Features

Robert S. Lynch, Jr., Naval Undersea Warfare Center, Newport, RI, USA
Peter K. Willett, University of Connecticut, Storrs, CT, USA

Abstract: In this paper, the Bayesian Data Reduction Algorithm (BDRA) is compared to several neural networks to demonstrate classification performance and feature selection for fused binary valued features, where the statistical dependency (i.e., correlation or redundancy) between the relevant features of each class is varied. The BDRA uses the probability of error, conditioned on the training data, and a “greedy”approach (similar to a backward sequential feature search) for reducing irrelevant features from the data. Results are shown by plotting the probability of error as a function of the conditional probability between adjacent relevant features, where the number of relevant features is varied. In general, it is demonstrated that the performance difference between the BDRA and the neural networks depends on the statistical dependency between the features.

Session MoC2: Target Tracking - 1 – Track Fusion - 1

Chair: Chee-Yee Chong, Booz Allen & Hamilton, San Francisco, CA, USA
Co-Chair: X.Rong Li, University of New Orleans, LA, USA

1- Evaluating Hierarchical Track Fusion with Information Matrix Filter

Kuo-Chu Chang, George Mason University, Fairfax, VA, USA.

Abstract:This paper examines track fusion performance under various degrees of non-deterministicitity of the target dynamics, i.e., process noises. There are three approaches to state vector fusion, Weighted Covariance, Information Matrix, and Pseudo-Measurement. This paper focuses on performance evaluation of the Information Matrix form of state vector fusion. Closed form analytical solution of steady state fused covariance for hierarchical fusion architecture both with and without feedback have been derived. These results provide interesting insight into the mechanism of track fusion and greatly simplify the evaluation of fusion performance. In addition, availability of such a solution facilitates the trade-off studies for designing fusion systems under various operating conditions.

2 - Unified Optimal Linear Estimation Fusion - Part I: Unified Models and Fusion Rules

X. Rong Li, University of New Orleans, LA, USA
Yunmin Zhu, Sichuan University, Chengdu, Sichuan, China
Chongzhao Han, Xi’an Jiaotong University, Xi'an, Shaanxi, China

Abstract: This paper deals with estimation fusion; that is, data fusion for the purpose of estimation. Three estimation fusion architectures are considered: centralized, distributed, and hybrid. A unified linear model and general framework for these three architectures are established. Optimal fusion rules in the sense of best linear unbiased estimation (BLUE) and weighted least squares (WLS) are presented for cases with either complete, incomplete, or no prior information. These rulesare muchmore generalandflexible thanpreviousresults. For example, they are in a unified form that are optimal for all the three fusion architectures with arbitrariy correlation of local estimates or observation noises across sensors or across time. They are also in explicit forms convenient for implementation.

3 - Unified Optimal Linear Estimation Fusion - Part II: Discussions and Examples

X. Rong Li, University of New Orleans, LA, USA
Jie Wang, Xi’an Jiaotong University, Shaanxi, China

Abstract:Several unified optimal linear estimation/track fusion rules in the sense of best linear unbiased estimation (BLUE) and weighted least squares (WLS) have been presented in [3] for centralized, distributed, and hybrid fusion architectures. This paper presents their relationships, verifies these rules and demonstrate via computer simulation examples how these fusion rules can be used in cases with either complete, incomplete, or no prior information about the esti-matee (i.e., the quantity to be estimated).

4 - Problem Characterization in Tracking / Fusion Algorithm Evaluation

Chee-Yee Chong, Booz Allen & Hamilton, San Francisco, CA, USA

Abstract:The performance of a tracking/fusion algorithm depends very much on the complexity of the problem. This paper presents an approach for evaluating tracking/fusion algorithm that considers the difficulty of the problem. Evaluation is performed by characterizing the performance of the basic functions of prediction and association. The problem complexity is summarized by means of context metrics. Two context metrics to characterize prediction and association difficulty are normalized target mobility and normalized target density. These metrics should be presented along with the performance metrics in performance evaluation. The metrics also allow more efficient generation of input data for performance evaluation. Simple tests for basic tracking algorithm functions are presented.

Session MoC3: Sensor and Information Fusion

Chair: Jeffery Layne, US Air Force Research Lab. WPAFB, OH, USA
Co-Chair: Erik Blasch, US Air Force Research Lab. WPAFB, OH, USA

1 - Robust Data Fusion

J.F. Grandin, M. Marques, Thomson-CSF DETEXIS, Trappes-Elancourt, France

Abstract:This paper compares different fusion processes in terms of error probability and robustness. Simples fusion processes are studied in the general framework of two binary sensors (maximum likelihood and logical fusion functions like ‘or’, ‘and’…). In the sequel, the comparison is extended to likelihood vectors adding other fusion processes ( Bayes or entropy criteria, T-norms, T-conorms, means operators). Finally, different models of fusion processes combination informed with the imprecision and the reliability of information sources are proposed and demonstrate to have good robustness properties.

2 - Methods and Concepts for Air Situation Picture Generation

Eric Shynar, Uri Degen, Advanced Technology Ltd., Tel-Aviv, Israel

Abstract: There is some confusion in terminology concerning architectures and algorithms of Air Situation Picture Generation (ASPG). Therefore, two aspects concerning the ASPG are distinguished: the method of ASPG - Single or Multi Radar Tracking, and the concept - Distributed or Central, - of ASPG within an Air Defense Region, consisting of several Control and Reporting Centers. The evolution from Single Radar Tracking to Multi Radar Tracking and from Distributed Concept to Central Concept is discussed.

3 - Fusion Method For Physical Systems Based On Physical Laws

Nageswara S. V. Rao, David B. Reister, Jacob Barhen, Oak Ridge National Laboratory, Oak Ridge, TN, USA

Abstract: We consider a physical system described by a set of parameters. Each parameter is either measured by a number of sensors or estimated by a set of computer programs that use sensor measurements. As a result, the resultant parameter values could be widely varying. We propose a fusion method that combines the measurements and estimators based on the physical laws that relate the parameters. In comparison with the traditional fusion problems, there isnotraining set that provides the actual parameter values. Furthermore, since every parameter is measured or estimated, there are noparameters whose actual values are known. We propose a fuser based on the least violation of the physical laws that relate the parameters. Under certain smoothness conditions on the physical law, we show the asymptotic convergence of our method, and also derive distribution-free performance bounds based on finite samples. We illustrate the effectiveness of this method for a practical problem of fusing well-log data in methane hydrate exploration. For this problem, data fusion method resulted in an order of magnitude improvement in the accuracy compared to the best set of estimators for the key parameter of porosity.

4 - A Statistical Overview of Recent Literature in Information Fusion

L. Valet, Gilles Mauris, Philippe Bolon, LAMII / CESALP, Annecy, France

Abstract:The objective of this paper is to make a picture of the recent articles published on information fusion. Indeed, a great number of documents dealing with this technique are available in the literature. A classification scheme including application fields, fusion goals, fusion system architecture and mathematical tools is proposed. This overview of the last three years allows to compute the article distribution into each class. Finally, some elements of preliminary analysis of this classification are drawn.

Session MoC4: Information Modeling

Chair: Roger Reynaud, IEF / University Paris XI, Orsay, France
Co-Chair: Amy L. Magnus, Intelligent Information AFRL/IFTD, NY, USA

1 - Object Hypothesis Support in the Context of Knowledge-Based Fuzzy Possibilistic Fusion of Image Descriptions

Sotiris N. Raptis, S.G. Tzafestas, National Technical University of Athens, Greece.

Abstract:In the paper presented here a possibilistic image fusion scheme is investigated supported by a prior knowledge about the features’ distribution among objects-hypotheses. As a measure for possibilistic considerations fuzzy reasoning is used. The fusion strategy adopted here is original, in that individual features are studied separately instead of studying them independently of their own statistical behavior, as it is common in the literature. Moreover the possibilistic considerations are based on the fuzzy modeling of the features, again deviating from what is commonly seen in the literature which is adopting a feature vector logic. The investigation subjects are not the objects but their individual features although the final output favors specific object hypotheses. It is seen that applying a fuzzy-possibilistic reasoning for individual features is more efficient than considering objects as candidate hypotheses. Uncertainty is reduced when knowledge is embedded into the scheme. Prior knowledge is provided by the way the image descriptions are attributed to specific hypotheses. We therefore integrate into our computations a series of image descriptors ranging from skeleton components as edges, curves and closed contours to texture and image moments. These are fuzzily modeled. Fuzzy descriptors modeling is grounded on the idea that after applying a fuzzy classification or segmentation algorithm all pixels of the image will pertain to more than one neighboring image segments, prototypes or patterns. So any further computation on these pixels inherit initial pixels’ uncertainty provided by the segmentation procedure. Results can be interpreted as an object dependent feature behavior, which in turn leads to increased confidence in the right object hypotheses.

2 - Inquisitive Pattern Recognition

Amy L. Magnus, Intelligent Information AFRL/IFTD, Rome, NY, USA
Steven C. Gustafson, Air Force Institute of Technology, WPAFB, OH, USA

Abstract: In nature, inquisitiveness is the drive to question, to seek a deeper understanding, and to challenge assumptions. For the discrete world of computers, inquisitive pattern recognition is the constructive investigation and exploitation of conflict in information. Data fusion is a fertile proving ground for inquisitive technologies. Multi-source, multi-modal data inherently contain conflicting information. As data fusion incorporates capabilities for situation assessment, strategies to identify and resolve conflict become important. Inquisitive pattern recognition (IPR) is a persistent, unsupervised learning capability whose concepts include falsification---similar to the supervised learning technique of cross validation---and the classification of confusion in feature space. Coupled with knowledge base technologies, inquisitive pattern recognition enables a computer to acquire new experiences.

3 - A Symbolic Neuro-fuzzy Collaborative Approach for Inducing Knowledge in a Pharmacological Domain

M. Carmo Nicoletti, Arthur Ramer, University of New South Wales, Sydney, Australia
M. Aparecida Nicoletti, Fac. Ciências Farmacêuticas, Universidade de São Paulo, Brazil

Abstract: This paper discusses the experiments conducted with two conceptually different machine learning systems in a pharmacological domain related to the use of different excipients in drug production. It shows how a symbolic system can be used, in a collaborative way, to help a neuro-fuzzy system to induce a more appropriate set of fuzzy rules.

4 - Double Representation of Information and Hybrid Combination for Identification Systems

Alain Nifle, Thomson-CSF/ISR, Massy, France
Roger Reynaud, IEF / Université Paris XI, Orsay, France

Abstract: In an increasing number of classification systems, a priori and observations are both present as probabilistic and possibilistic continuous distributions to represent information in the most accurate and reliable way. We propose a method where information is simultaneously modeled in term of probability and possibility and is combined in a hybrid manner, without changing it in an entirely probabilistic form nor in an entirely possibilistic form. It thus defines a compatibility mass function. To extend the mechanism of validation windowing found in tracking algorithms, a possibilistic distribution is associated with each continuous probabilistic distribution, and contributes to build a mass function related to the validation-rejection of the observation. Thanks to these hybrid mechanisms of combination, we build a classifier respecting some constraints in the framework of evidence theory. In the paper we describe it and discuss its properties.

Session MoC5: System Design

Chair: Akira Namatame, National Defense Academy, Japan

1 - Control, Estimation and Abstraction in Fusion Architectures: Lessons from Human Information Processing

Carl B. Frankel, Organizational Measurement and Engineering, San Francisco, CA, USA
Mark D. Bedworth, Jemity / University of Central England, Birmingham, UK

Abstract:Human fusion offers important principles for the design of machine fusion. The most important of these is the separability of control, estimation and abstraction, without which competent real-time responsiveness cannot be assured. Another is to separate intention from embodiment, so that intentions can be flexibly embodied. Yet another, similar to human emotions, is the use of positive feedback events and subsequent feed-forward to produce timely responses to dynamic situations. Still another, leveraging the trajectory of human cognitive development, is to control scope of processing by layering data abstractions so that processing results can be immediately apprehended by human users. By applying these principles, machine fusion can both increase its competence and also increase the perception of its competence, increasing the likelihood that the machine will be received and treated as a trusted partner.

2- Information Fusion Method For System Identification Based On Sensitivity Analysis

Jacob Barhen, Nageswara S. V. Rao, Oak Ridge National Laboratory, TN, USA

Abstract:We consider the identification of a parametrized time-invariant non-linear plant using a smooth model such as a sigmoid non-linear network. There is measurement noise associated with the plant parameters as well as it's input and output. An initial plant model is obtained by utilizing the domain-specific knowledge in terms of the fundamental plant equations, which in general only partially capture the plant dynamics. Once the initial model is fixed, measurements are collected on the plant parameters and input/output. We show that the iid measurements can be fused with the initial plant model by recomputing the parameters. The updated parameters yield a more accurate identifier of the original plant both in parameter and input/output space. The method is based on empirical versions of the closed-form solutions derived in the nuclear engineering literature for an ideal version of the problem based the sensitivity analysis. We show the asymptotic convergence of our computational procedure as well as derive its finite sample results. We illustrate the method using an identifier based on a sigmoid feedforward neural network.

3 - Information Value Mapping for Fusion Architectures

Timothy J. Peterson, Malur K. Sundareshan, University of Arizona, Tucson, AZ, USA

Abstract: In assessing a fused sensor system, one considers the quality of the system architecture most often by the capabilities of the individual sensors, and the attributes of the fusion algorithm. Though it is possible to evaluate system performance in an idealized context, and model real-world perturbations as random disturbances, it may be advantageous to treat predictable events as deterministic. Additionally, a-priori information is not limited to constraints on the object of interest, but also can be applied to the scene in which the object resides. In this paper, the role of information value mapping in the design of efficient fusion architectures is presented. This concept provides a development tool for general scene modeling that attempts to capture the relative value of each sensor’s input related to a given location in the sensed scene. Some general guidelines for the development of information value maps will be outlined. Specific examples illustrate the modeling and implementation methods. The effects of including signal processing methods for modifying the sensor outputs, and hence the information value maps, will be given a particular focus. The use of currently popular image super-resolution algorithms is offered for an illustrative discussion of these methods.

4 - Decentralized Decision Networks and Their Emergent Properties

Saori Iwanaga, Akira Namatame, National Defense Academy, Yokosuka, Japan.

Abstract:We address a new concept of decentralized decision networks that have a high degree of survivability when they are simultaneous software errors, hardware malfunctions or hostile attacks. The large-scale effects of locally interacting agents are called emergent properties of the system. Emergent properties are often surprising because it can be hard to anticipate the full consequences of even simple forms of interaction. We address a new framework and architecture for a distributed decision system to aid real-time dynamic decision-making in a highly competitive environment.

Session MoD1: Target Detection and Recognition - 2

Chair: Per Svensson, Defence Research Establishment, Stockholm, Sweden
Co-chair: Erland Jungert, Defence Research Establishment, Linköping, Sweden

1 - Evolutionary Control of an Autonomous Field

Mark W. Owen, SPAWAR, San Diego, CA, USA
Dale M. Klamer, Barbara Dean, Orincon Corporation, San Diego, CA, USA

Abstract: An autonomous field of sensor nodes needs to acquire and track targets of interest traversing the field. Small detection ranges limit the detectability of the field. As detections occur in the field, detections are transmitted acoustically to a master node. Both detection processing and acoustic communication drain a node’s power source. In order to maximize field life, an approach must be developed to control processes carried out in the field. In this paper we develop an adaptive threshold control scheme. This technique will minimize the power consumption while still maintaining the field-level probability of detection. The power consumption of the field of sensor nodes is driven by the false alarm rate at the individual sensor nodes in this problem formulation. The control law to be developed is based upon a stochastic optimization technique known as evolutionary programming. At the end of the paper, a set of results are presented with some conclusions.

2 - Target Detection and Identification using Neural Networks and Multi-Agents Systems

Roger Cozien, C. Rosenberger, P. Eyherabide, J. Rossettini, A. Ceyrolle, Saint-Cyr Coëtquidan Military School, Guer, France

Abstract: Our purpose is, in medium term, to detect in air images, characteristic shapes and objects such as airports, industrial plants, planes, tanks, trucks, … with great accuracy and low rate of mistakes. However, we also want to value whether the link between neural networks and multi-agents systems is relevant and effective. If it appears to be really effective, we hope to use this kind of technology in other fields. That would be an easy and convenient way to depict and to use the agents' knowledge which is distributed and fragmented. After a first phase of preliminary tests to know if agents are able to give relevant information to a neural network, we verify that only a few agents running on an image are enough to inform the network and let it generalize the agents' distributed and fragmented knowledge. In a second phase, we developed a distributed architecture allowing several mutli-agents systems running at the same time on different computers with different images. All those agents send information to a "multi neural networks system" whose job is to identify the shapes detected by the agents. The name we gave to our project is Jarod.

3 - Using Optimal Variables for Bayesian Network Classifiers

Fatima El-Matouat, Patrick Vannoorenberghe, Olivier Colot, Jacques Labiche, Université / INSA de Rouen, Mont-Saint-Aignan, France

Abstract: Using graphical models to represent independence structure in multivariate probability model has been studied since a few years. In this framework, Bayesian networks have been proposed as an interesting approach for uncertain reasoning. Within the framework of pattern recognition, many methods of classification were developped based on statistical data analysis. Belief networks were not considered as classifiers until the discovery that Naive Bayes, a very simple kind of Bayesian network, is surprisingly effective. In this paper, we propose to use belief networks classifiers with optimal variables that is to say networks which have to manage discrete and continuous variables.

4 - Modeling the Column Recognition Problem in Tactical Information Fusion

Johan Björnfot, Per Svensson, Royal Institute of Technology, Stockholm, Sweden.

Abstract: In this paper, we discuss the application of Hidden Markov Modeling (HMM) techniques to the column recognition problem, where a non-cooperative military unit consisting of a sequence of objects forms a transportation column. Here the task is to infer the object composition and organizational structure of the column from imperfect observations of individual objects, in combination with generic a priori information about the organizational structure of the non-cooperative forces. Good solution methods for the column problem would provide a significant contribution to the automatization of the aggregation process. To the best of our knowledge, the application of HMM to the column problem has not been previously proposed.

Session MoD2: Target Tracking - 2 – Multi-Model Approach

Chair: X. Rong Li, University of New Orleans, LA, USA
Co-chair: Mohamad Farooq, Royal Military College, Kingston, Ontario, Canada

1 - Model-Set Adaptation Using a Fuzzy Kalman Filter

Zhen Ding, Raytheon Systems Canada Limited, Waterloo, Ontario, Canada
Henry Leung, University of Calgary, Albert, Canada
Keith Chan, The Hong Kong Polytechnic University, Hong Kong

Abstract: In this paper, a fuzzy Kalman filter is proposed to combat the model-set adaptation problem since it is found to be able to extract more exactly dynamic information. The fuzzy Kalman filter uses a set of fuzzy rules to adaptively control the noise covariance and that makes it more suitable for real radar tracking. The proposed fuzzy Kalman filter is then combined with an IMM algorithm, hence, a fuzzy IMM (FIMM) algorithm is obtained. The performance of the FIMM algorithm is compared with that of an adaptive IMM (AIMM) algorithm using real radar target tracking data. Simulation result shows that the FIMM algorithm outperforms the AIMM algorithm in terms of both the root mean square prediction error and the number of track loss.

2 - Tracking Closely Maneuvering Targets in Clutter with an IMM-JVC Algorithm

Alexandre Jouan, Lockheed Martin Canada, Montréal, Québec, Canada
Benoit Jarry, Hannah Michalska, McGill University, Montréal, Québec, Canada

Abstract:The tracking of closely maneuvering targets represents a challenge for both the contact-to-track association and the positional estimation algorithms. Previous simulations have shown that the coupling of an association scheme using the Jonker-Volgenant-Castanon (JVC) optimization with an Interacting Multiple Model (IMM) positional estimator gives superior tracking performance than other tested combinations such as the JVC-Adaptive Kalman Filter (JVC-AKF), the Nearest Neighbor (NN)-AKF (NN-AKF) or the NN-IMM. However, the efficiency of the JVC optimization scheme will depend on how the assignment matrix is built. This highlights the role played by both the construction of the buffers of contacts and the selection of the tracks likely to be updated with the incoming contacts. After a brief recall of the IMM-JVC formalism, this paper presents an analysis of the JVC output and identifies additional functionalities that should be activated to improve its performance. Simulation results are obtained on a scenario that involves two closely maneuvering aircraft. Sensor reports are contaminated with randomly simulated clutter.

3 - Comparing an Interacting Multiple Model Algorithm and a Multiple-Process Soft Switching Algorithm: Equivalence Relationship and Tracking Performance

Tan-Jan Ho, Mohamad Farooq, Royal Military College of Canada, Kingston, Ontario, Canada

Abstract:In this paper, we show that a relationship exists between a multiple-model soft- switching filter and an interacting multiple-model (IMM) filter. By assuming that each model transition probability is equally likely, the constraints imposed on the mixing and fusion weights in the proposed IMM filter are similar to those in the multiple-model soft-switching filter obtained in a referenced paper. Thus, the constraints derived from a probabilistic approach can be reduced to those obtained via a deterministic approach. Without the aid of an additional constraint, the IMM-type filter uses the real-time information of the innovations and their covariances to choose filter weights lying within 0 and 1. The results of this paper show that the multiple-process soft-switching filter is a special case of an IMM filter. This will be further substantiated through simulations.

4 - Variable- and Fixed- Structure Extended IMM Algorithms Using Coordinated Turn Model

Emil Semerdjiev, Ludmila Mihaylova, Bulgarian Academy of Sciences, Sofia, Bulgaria
X. Rong Li, University of New Orleans, LA, USA

Abstract:A new variable structure (VS) Extended Interacting Multiple Model (EIMM) technique is developed in the paper. Fixed structure (FS) and VS EIMM algorithms using extended constant velocity and extended coordinated turn (ECT) models, are proposed. The ECT model includes the difference between the unknown current turn rate and its fixed value assumed in each model for the IMM algorithm. Due to the estimated turn rate, significant self-adjusting abilities are provided to the designed EIMM algorithms, which give very good overall accuracy and consistency. Both EIMM algorithms are compared to a particular VS adaptive grid IMM algorithm. It is shown that the VS IMM algorithms show better mobility, while the FS EIMM algorithm possesses better consistency.

Session MoD3: Image Fusion - 1

Chair: Pramod K. Varshney, Syracuse University, NY, USA
Co-chair: Wojciech Pieczynski, INT, Evry, France

1 - Different Focus Points Images Fusion Based on Wavelet Decomposition

Xuan Yang, Wanhai Yang, Jihong Pei, Xidian University, Shaanxi, China

Abstract: A new technique is developed for the data fusion of two images. Two spatially registered images with differing focus points are fused by deciding clear objects. At first, an impulse function is defined to describe the image quality of an object. Then the clear region is decided by analyze the wavelet decomposition components of two primary images and two blurred images. The results of the comparison show this method performing better in preserving edge information for the test images than that of the other image fusion methods.

2 - A Pyramid Approach For Multimodality Image Registration Based On Mutual Information

Hua-mei Chen, Pramod K. Varshney, Syracuse University, NY, USA

Abstract:A pyramid approach for multimodality image registration based on mutual information is presented. The image pyramid is obtained by using the wavelet transform. An exhaustive search algorithm at the coarsest level of the image pyramid is developed. Image partitioning and gray level intensity binning are used to increase the fidelity of the process. Because of the fact that image partitioning is used, our algorithm has the potential to be parallelized and implemented on a multiprocessor computer. Our algorithm has been applied on remotely sensed images. The results show that our algorithm is very promising.

3 - 2-D Image Fusion by Multiscale Edge Graph Combination

Stavri G. Nikolov, Dave R. Bull, C. Nishan Canagarajah, University of Bristol, UK
Mike Halliwell, Peter N. T. Wells, Bristol General Hospital, UK

Abstract: A new framework for fusion of 2-D images based on their multiscale edges is described in this paper. The new method uses the multiscale edge representation of images proposed by Mallat and Hwang. The input images are fused using their multiscale edges only. Two different algorithms for fusing the point representations and the chain representations of the multiscale edges (wavelet transform modulus maxima) are given. The chain representation has been found to provide numerous new alternatives for image fusion, since edge graph fusion techniques can be employed to combine the images. The new framework studies different levels, i.e. pixel and feature level, of image fusion in the wavelet domain.

4 - Robust Multisensor Image Registration with Partial Distance Merits

Xiangjie Yang, Yunlong Sheng, Image Science Group, Univ. Laval, Ste-Foy, Québec, Canada
Léandre Sévigny, DRE Valcartier, Courcelette, Québec, Canada
Pierre Valin, Lockheed Martin Canada, Montréal, Québec, Canada

Abstract: Challenge in the registration of battlefield images in visible and far-infrared bands is the feature inconsistency. We use a contour-based approach and propose robust free-form curve-matching algorithms using the adaptive hill climbing and the iterative closest point algorithm. Both algorithms do not requires explicit curve feature correspondence, are designed to be robust against outliers. The originality of this work is the use of the mean partial distance as the objective function in the iterative closest point algorithm, so that outliers can be easily handled by using rank order statistics. A fast algorithm using the segment representation of Voronoi diagram for the nearest point transform and the distance transform is used.

Session MoD4: Bayesian and Belief Fusion Approaches

Chair: Wojciech Pieczynski, INT, Evry, France
Cochair: Mohammed Benjelloun, Université du Littoral Côte d'Opale, Calais, France

1 - A Conservative Approach to Distributed Belief Fusion

Churn-Jung Liau, Institute of Information Science, Academia Sinica, Taipei, Taiwan

Abstract: In this paper, we develop logics for merging beliefs of agents with different degrees of reliability. The logics are obtained by combining the multi-agent epistemic logic and multi-sources reasoning systems. Every ordering for the reliability of the agents is represented by a modal operator, so we can reason with the merging information under different situations. The approach is conservative in the sense that if an agent's belief is in conflict with those of higher priorities, then his belief is completely discarded from the merged result. We consider two strategies for the conservative merging of beliefs. In the first one, if inconsistency occurs at some level, then all beliefs at the lower levels are discarded simultaneously, so it is called level cutting strategy. For the second one, only the level at which the inconsistency occurs is skipped, so it is called level skipping strategy. The formal semantics and axiomatic systems for these two strategies are presented.

2 - A Generic Framework for Resolving the Conflict in the Combination of Belief Structures

Eric Lefevre, Olivier Colot, Patrick Vannoorenberghe, Denis de Brucq, Université/INSA de Rouen, France

Abstract: Within the framework of Dempster-Shafer theory of evidence, the data fusion is based on the building of single belief mass by combination of several mass functions resulting from distinct information sources. This combination called Dempster's combination rule (or orthogonal sum) has several interesting mathematical properties like commutativity or associativity. Unfortunately, it badly manages the existing con ict between the various information sources at the step of normalization. In this paper, we introduce traditional combination operators used within the framework of evidence theory. We propose other combination operators allowing an arbitrary redistribution of the con icting mass on the propositions. These various combinations operators were tested on sets of synthetic and real masses.

3 - Optimal Segmentation by Random Process Fusion

Serge Reboul, Damien Brige, Mohammed Benjelloun, Université du Littoral Côte d’Opale, Calais, France

Abstract: We introduce in this article an optimal segmentation method of nonstationary random processes. Segmentation of a non stationary process consists in assuming piecewise stationarity and in detecting the instants of change. We consider here that all the data from all the sensors are avaible in a same time and perform a global segmentation. The bayesian fusion method we propose for the segmentation is based on the introduction of a joint prior model for the simultaneously segmentation and estimation of data coming from a set of sensors. We build a change process and define its prior distribution for the data fusion. That allows us to propose the MAP estimate as well as some minimum contraste estimate as a solution. We define, in the parametric processes distribution case, the expression and signification of all the segmentation’s parameters. We compare the performance of our detection method in the case of two or three sensor. Application to the fusion of wind data velocity and direction is proposed.

4 - Pairwise Markov Chains and Bayesian Unsupervised Fusion

Wojciech Pieczynski, INT, Evry, France

Abstract: We propose a new model called a Pairwise Markov Chain (PMC), which generalises the classical Hidden Markov Chain (HMC) model. The PMC model is more general than HMC in that the process one wants to estimate is not necessarily a Markov process. However, PMC allows one to use the classical Bayesian restoration methods like Maximum A Posteriori (MAP), or Maximal Posterior Mode (MPM). So, akin to HMC, PMC allows one to restore hidden stochastic processes, with numerous applications to speech recognition, multisensor image segmentation, among others. Furthermore, we propose a new method of parameter estimation, which allows one to perform unsupervised restoration with PMC. The method proposed is valid even with non Gaussian and possibly correlated noise. Furthermore, the very form of the statistical distribution of the noise need not be known exactly. All that is required is that for each class the form of the noise distribution belongs to a given set of forms.

Session MoD5: System Design and Applications - Invited Session

Chair: Mark Bedworth, Jemity / University of Central England, Birmingham, UK

1 - Data Fusion System Engineering

Alan N. Steinberg, Veridian ERIM International, Chantilly, VA, USA

Abstract: The paper reports on methods for cost-effective development and integration of multi-sensor fusion technology. The methods presented derive from the Project Correlation Data Fusion Engineering Guidelines with significant evolution in current efforts for DARPA (Defense Advanced Research Agency), BMDO (Ballistic Missile Defense Organization) and elements of the U.S. intelligence community.
Approaches for four types of research are distinguished: Requirement-Driven (Find solutions to mission problems) ; Acquisition-Driven (Prototype evaluation of techniques to support system acquisition & technology insertion) ; Phenomenology-Driven (Discover exploitation potential in untried combinations of multi-sensor/ multi-discipline phenomenology ) ; Technology-Driven (Proof-of-concept evaluation to support technology road-map).
Methods are presented for each of these. In particular, methods for requirement-driven and acquisition-driven developments are discussed in terms of the following phases: Problem decomposition - assigning the role for data fusion, as well as for other system functions (sensors, communications, response assets, human operators, etc.) ; Data Fusion Tree Design - partitioning process among C3 nodes and into processing nodes; interaction with sensors,/sources, resource management nodes, and information users ; Data Fusion Node Design - data input/output, allocation to human/automatic processes, technique selection ; Detailed Design and Development - pattern application, algorithm tailoring, software adaptation and development ; Multi-Sensor Test and Evaluation - metrics, test environments and procedures.
Diverse examples of system development experiences are presented, with lessons learned regarding applicability of specific system engineering methods and data fusion techniques.

2 - An Architecture for the Integration of All Levels of Data Fusion

Stephane Paradis, Jean Roy, DRE Valcartier, Québec, Canada

Abstract: Data fusion is a key enabler for the situation analysis process. The study and the integration of all data fusion levels require both a suited architecture, and a modeling and simulation environment that allows the technological demonstration of the situation analysis concepts to be investigated, and also the demonstration of their supportive contribution to the situational awareness state. This paper discusses both of these aspects. In particular, it focuses on design issues related to development of a simulation environment for the analysis of the tactical situation. Finally, the paper describes a bootstrapping effort to develop a baseline of this environment. This baseline has been successfully used to support the analysis of threat evaluation algorithms for the command and control system of the IROQUOIS class ships.

3 - Practical Trade-offs in Fusion Architecture Design

Elisa Shahbazian, Rodney Hallsworth, Daniel Turgeon, Lockheed Martin Canada, Canada

Abstract: Since 1991, the Research and Development (R&D) group at Lockheed Martin Canada (LM Canada) has been developing and demonstrating the application of Multi-Source Data Fusion (MSDF) techniques for target tracking and identification within the Naval Command and Control (C2) for the HALIFAX Class frigates. The current C2 as well as the sensor suite of the HALIFAX Class were designed in the early 80s and based around a proprietary hardware architecture. The sensor data is pre-processed and provided to the C2 in real time. Considering the late 70s and 80s design of the sensor interfaces, where a data fusion within the C2 was not a commonality, not all of the information beneficial for a data fusion system is provided to the C2. After a sequence of simulation and modelling efforts for an MSDF capability within the HALIFAX Class C2, this project is now at a point where real data captured from a ship trial on the HALIFAX Class is being injected into the MSDF. This is an on-going activity and a number of iterations are foreseen before the MSDF becomes part of HALIFAX Class C2. This paper provides a summary of lessons learned in this exercise.

4 - Decisions in Condition Monitoring - An Exemplar for Data Fusion Architecture

Paul Hannah, Andrew Starr, Andrew Ball, University of Mancheter, UK

Abstract: This paper aims to demonstrate the strategy and structures involved in making decisions based on condition data, and to draw parallels with data fusion models. A new df model is demonstrated, and examples are drawn from condition monitoring applications. In particular, the work here presented introduces a new framework for the application of data fusion solutions to the analysis of engineering problems. A review of frameworks used in data fusion applications is presented, along with important factors to consider in the lay out of a robust process model, to host a coherent and effective data fusion problem-solving strategy. The main theme of the work focuses on the development of an intelligent multi-sensored engine. The partners involved in this research effort aim to develop a robust methodology for sensing and analysis under harsh environments, stressing its application to the fields of combustion and fault diagnostics analysis. The proposed process model will be used to facilitate the implementation of a common strategy to tackle these problems.

Monday - Tuesday - Wednesday - Thursday

Last Updated:June 6, 2000
Web site by: (content), (form)
copyright © ISIF 2000