Final Long Program

Monday - Tuesday - Wednesday - Thursday

Tuesday July 11

Session TuA1: Plenary talk

Possibility Theory in Information Fusion

Prof. Henri Prade, Institut de Recherche en Informatique de Toulouse (IRIT), France

Abstract: Possibility theory and the body of aggregation operations from fuzzy set theory provide some tools to address the problem of merging information coming from several sources. The approach to fusion is set-theoretic and the choice of conjunctive versus disjunctive fusion modes depends on assumptions on whether all sources are reliable or not. Quantified, prioritized and weighted and fusion rules are described. A possibilistic logic counterpart of these combination modes will be also briefly presented. The fusion of imprecise information is carefully distinguished from the estimation problem. Fuzzy extensions of estimation processes are also discussed. The approach, based on conflict analysis, applies to sensor fusion, aggregation of expert opinions as well as the merging of databases especially in case of poor, qualitative information.

Session TuB1: Tracking Systems - 1

Chair: Alfonso Farina, Alenia Marconi Systems, Roma, Italy
Co-Chair: Barbara La Scala, The Preston Group, Richmond, Australia

1 - A Micro-Density Approach to Multitarget Tracking

Keith Kastella, Veridian ERIM International, Ann Arbor, MI, USA

Abstract: This paper presents an approach to multitarget tracking based on recursive estimation of a conditional probability density functional for the MultiTarget Micro-Density (MTMD). The MTMD is a distribution that, when integrated over a region in target state space, gives the number of targets in that region. When target motion is governed by the Ito equation, the MTMD becomes a stochastic function that is characterized by a time-dependent probability density functional (PDFl) that obeys a type of Fokker-Plank equation (FPE) which is derived here. Bayes formula can be used to incorporate measurements into the PDFl to obtain the conditional PDFl. Numerical solution of the FPE and its Bayes’ formula update are illustrated in a brief numerical example.

2 - Tracking System Prediction Through Group Correlation Analysis

Brandon Bovey, Donald E. Brown, University of Virginia, Charlottesville, VA, USA

Abstract: This paper presents a method for predicting target locations based upon the correlation of movements for targets traveling in groups. The algorithm is designed for incorporation into a modern tracking system as a supplemental module, providing updated estimates of target velocities based upon the results of a correlation analysis. The approach is presented for application in the realm of ground-based tracking systems, although the approach is general and can be applied in any domain where target movements are correlated.

3 - A Hybrid-State Estimation Algorithm for Multi-Sensor Target Tracking

Stefano Coraluppi, Mark Luettgen, Craig Carthel, Alphatech Inc., MA, USA

Abstract: This paper describes a hybrid-state filtering algorithm that enables tracking of moving and stationary vehicles, on the basis of moving-target-indicator (MTI) measurements and SAR-based imagery detections. We use a hybrid-state model for vehicle dynamics with discrete states move and stop, and the discrete state influences the continuous-state dynamics through the process noise. We present a near-optimal recursive filter that is a hybrid-state extension to the well-known Extended Kalman Filter (EKF). We study the performance of the filter with a number of target trajectories. Our framework can be easily extended to include other sensor types, including EO-based imagery detections and signal intelligence measurements. Also, the filtering algorithm can be used as part of a multi-sensor multi-target tracking algorithm.

4 - Stochastic Estimation using a Continuum of Models

Jeffery Layne, Scott Weaver, U.S. Air Force Research Lab., WPAFB, OH, USA

Abstract: In this paper, we investigate a recursive multiple model tracking approach similar to the Generalized Pseudo–Bayesian 1 (GPB1) approach. However, here we consider a continuum of models rather than the discrete set that is usually implemented in the GPB1 method. By doing so better models are available to improve tracker performance and solve the symmetry problem inherent in most multiple model approaches.

Session TuB2: Target Tracking - 3 – Passive Sensors

Chair: Ivan Kadar, Consultant, Northrop Grumman Corporation, Bethpage, NY, USA
Co-chair: Edward Carapezza, Defense Advanced Research Projects Agency (DARPA), USA

1 - Bearings-Only Tracking using Data Fusion and Instrumental Variables

Y.T. Chan, Royal Military College of Canada, Kingston, Ontario, Canada
Terry A. Rea, National Defense Headquarters, Ottawa, Ontario, Canada

Abstract: This paper presents a recursive Measurement Instrumental Variables Bearings-Only Tracking (MIV-BOT) method for a stationary observer. A smoothing operation directly fuses multi-sensor bearing measurements by exchanging the measurements as the instruments in a pseudo linear estimator. The MIV-BOT formulation produces a smoothed velocity estimate parameterized to any position along the target trajectory, which is found from a single laser range finder measurement. Target range predictions, derived from the smoothed two-state velocity estimate, are then used as range measurements in two parallel Kalman filters. The result is a recursive, passive and unbiased fusion scheme. The theoretical development is investigated by Monte Carlo simulation in short tracking scenarios. Experimental results show that the fusion scheme produces reliable estimates for non-manoeuvring targets.

2 - A Hough Transform Track Initiation Algorithm for Multiple Passive Sensors

Kiril M. Alexiev, Ljudmil V. Bojilov, Bulgarian Academy of Sciences, Sofia, Bulgaria

Abstract: This paper concerns the data association problem. The data are received from several passive sensors in the presence of clutter and missed detections. The Hough transform algorithm initiates tracks and resolve the ambiguous measurement-target associations. An effective heuristic ghosts elimination technique is proposed in the paper, too. Numerical results are also presented.

3 - Passive Multisensor Multitarget Feature-aided Unconstrained Tracking: a Geometric Perspective

Ivan Kadar, Consultant, Northrop Grumman Corporation, Bethpage, NY, USA

Abstract: Novel, targets-to-sensors' geometry-based performance measure, bootstrap estimation algorithm and feature-aided association are described for the passive multisensor multitarget data association, position and velocity measurement estimation and coupled unconstrained association/tracking problem. The approach reduces computational complexity and ghost targets, and provides dynamically changing geometry dependent on-line estimation of both the target's velocity measurements and the computation of the associated correlated position and velocity measurement noise covariance matrix (R-matrix). Sequences of these estimates, along with position measurement estimate sequences, serve as inputs to a Kalman filter tracker, associating/forming/de-ghosting and maintaining tracks in Cartesian coordinates. Based on state estimates of targets, a relative geometric measure-of-merit is used to select sensors for optimum tracking performance. Previous approaches to the passive multisensor-multitarget position state estimation problem did not incorporate feature-aided gating and association, and used R-matrix formulations, based on Cramer-Rao lower bound computations, which do not explicitly exploit the effects of the changing geometry. An overall system construct embodying the above features is described. The tracking performance efficacy of the new algorithmic system is demonstrated in a simulated self-organizing network of synchronized acoustic Unattended Ground Sensors (UGS) using sequences of bearing measurement sets from triplets of UGS.

Session TuB3: Image Fusion and Exploitation - 1 – Invited Session

Chair: Allen Waxman, MIT Lincoln Laboratory,Lexington, MA, USA
Co-Chair: William Streilein, MIT Lincoln Laboratory, Lexington, MA, USA

1 - Potential Utility and Needs for Imagery Fusion Technology

James Fahnestock, Chung Hye Read, U.S. National Imagery and Mapping Agency, USA

Abstract: The United States National Imagery and Mapping Agency (NIMA) is an information providing organization which must cope with large volumes of data supplied by various imagery sources. These image data types are or will be used to support policy makers and military commands. In supporting these decision-makers, accuracy and timeliness are critical for solving difficult problem sets. Fusion of various image data types provides increased data dimensionality that could lead to more robust solutions to these customers’ problems. NIMA needs to assure significant improvement in processing algorithms and achieve greater efficiencies to adequately perform our future mission functions. NIMA is transitioning to digital image data processes requiring softcopy exploitation of enormous data volumes. This, coupled with rapid changes in digital technologies is providing fertile ground for the application of fusion concepts to improve organizational performance. Imagery fusion products show potential for improvement of NIMA’s work on Assisted Feature Extraction/Assisted Target Recognition, Change Detection, Modality Understanding, Site Modeling, and Terrain Visualization. Like other data and imagery users, NIMA needs to make image fusion advances that will highlight important information and present more information than is evident in individual images. These increases in information should provide for improved efficiency, reduced uncertainty, and enhanced performance. As multi-modality image registration and fusion techniques mature, they could be critical enablers to NIMA’s future success. NIMA and the United States Imagery and Geospatial Information Service will benefit from advanced enabling technology research in these areas.

2 - AFOSR Research Programs in Image Fusion

John Tangney, U.S. Air Force Office of Scientific Research, Arlington, VA, USA

Abstract: The AFOSR basic research programs in image fusion are presented in context of Air Force technology needs for targeting, image exploitation, and autonomous systems. Programs include research involving human perception and neural processing in other biological systems, algorithms for fusion from multiple sources and platforms, and novel sensors. Available mechanisms for support of collaborative research will also be presented.

3 - Smart SensorWeb: Web-Based Exploitation of Sensor Fusion for Visualization of the Tactical Battlefield

Jeffrey Paul, U.S. Office of the Deputy Under Secretary of Defense for Science and Technology, Pentagon, Washington DC, USA

Abstract: Smart SensorWeb (SSW) is a recent DUSD(S&T) initiative inspired by extraordinary technological advances in sensors and microelectronics and by the emergence of the Internet as a real time communication tool. The overall vision for SSW is an intelligent, web-centric distribution and fusion of sensor information that provides greatly enhanced situational awareness, on demand, to Warfighters at lower echelons. Emphasis is on multi-sensor fusion of large arrays of local sensors, joined with other assets, to provide real-time imagery, weather, targeting information, mission planning, and simulations for military operations on land, sea, and air. This will require exploitation of advances in Sensor Fusion, in order to intelligently integrate imagery and other sources of information to give the battlefield commander real-time visualization – what he needs, when he needs it, with the military pay-off being Decision Dominance. This paper gives an overview of this new and exciting initiative, highlights the technology challenges in Sensor /Information Fusion and presents a program approach for near-term demonstrations and long-term solutions, involving the DoD, National Labs, commercial industry, and academia.

4 - Image Registration and Fusion in Remote Sensing for NASA

Jacqueline Le Moigne, James Smith, NASA Goddard Space Flight Center, USA

Abstract: With the increasing importance of multiple platform/multiple remote sensing missions, the integration of digital data from disparate sources has become critical to the success of these endeavors. In the near future, satellite remote sensing systems will provide large amounts of global coverage and repetitive measurements representing multiple-time or simultaneous observations of the same features by different sensors. Also, with the new trend of smaller missions, most sensors will be carried on separate platforms, resulting in a tremendous amount of data that must be combined. In meeting some of the Mission To Planet Earth objectives, the combination of all these data at various resolutions - spatial, radiometric and temporal - will allow a better understanding of Earth and space science phenomena. For example, for land cover applications, the combination of coarse-resolution viewing systems for large area surveys and finer resolution sensors for more detailed studies offer the multilevel information necessary to accurately assess the areal extent of important land transformations. High-resolution sensors, such as Landsat are very good for monitoring vegetation changes, e.g., changes in forest cover, when landscape features are local in scale. However, studies at a global or continental scale at high spatial and temporal resolutions would require the processing of very large volumes of data, and need to be performed with lower resolution sensors. It is therefore necessary to combine information from both types of sensors to conduct feasible, accurate studies.

Session TuB4: Evidential Reasoning Approach for Data Fusion

Chair: A. Appriou, ONERA, France

1 - Adding Decision Rule to the Shafer-Logan Algorithm for Hierarchical Identity Information Fusion

Anne-Laure Jousselme, D. Grenier, Univ. Laval, Canada
Éloi Bossé, DRE Valcartier, Val-Bélair, Québec, Canada

Abstract: In this paper, the Dempster-Shafer evidential theory is used in the form of the Shafer-Logan algorithm for fast computation when the information is hierarchically structured. Due to the hierarchical nature ofthe evidence, an algorithm proposed by Shafer and Logan is implemented which reduces the calculations from exponential to linear time proportional to the number of nodes in the tree. We present here main equations of the Shafer-Logan algorithm and give the owchart for implementation.We then add a decision rule based on the theory of utility. This decision rule offers a good way to take into account the hierarchical structure of the data, giving variable costs to nodes (propositions) depending on their level in the tree. Moreover, because of the form of the quantities in presence, a recursive computation is allowed which can be integrated asa last stage of the Shafer-Logan algorithm.

2 - Managing Inconsistent Intelligence

Johan Schubert, Defence Research Establishment, Stockholm, Sweden

Abstract: In this paper we demonstrate that it is possible to manage intelligence in constant time as a pre-process to information fusion through a series of processes dealing with issues such as clustering reports, ranking reports with respect to importance, extraction of prototypes from clusters and immediate classification of newly arriving intelligence reports. These methods are used when intelligence reports arrive which concerns different events which should be handled independently, when it is not known a priori to which event each intelligence report is related. We use clustering that runs as a back-end process to partition the intelligence into subsets representing the events, and in parallel, a fast classification that runs as a front-end process in order to put the newly arriving intelligence into its correct information fusion process.

3 - Applying Theory of Evidence in Multisensor Data Fusion: A Logical Interpretation

Laurence Cholvy, ONERA, Toulouse, France

Abstract: Theory of Evidence is a mathematical theory which allows one to reason with uncertainty and which suggests a way for combining uncertain data. This is the reason why it is usedas a basic tool for multisensor data fusion in situation assessment process. Although numerician people know quite well this formalism and its use in multisensor fusion, it is not the case for people used to manipulate logical formalisms. This present work intends to give them the key for understanding Theory of Evidence and its use in multisensor data fusion, first by giving a logical interpretation of this formalism, when the numbers are rational, and secondly, by reformulating, in a particular case, one model defined by Appriou for multisensor data fusion.

4 - An Evidential Markovian Model for Data Fusion and Unsupervised Image Classification

Laurent Fouque, Alain Appriou, ONERA, Châtillon, France
Wojciech Pieczynski, Institut National des Télécommunications, Evry, France

Abstract: In this paper, we deal with the fusion of information and the classification of images supplied by several sensors. By intrinsic characteristics of each sensors, provided informations are usually defined on different set of hypothesis, called frames of discernment. Adapted formalism need tobe used to compute the fusion process. We resolve this problem of multisensor image fusion and classification in an evidential framework, which is well adapted for the combination of knowledge defined on different frames of discernment. We present two models for merging available informations, a non contextual and a vectorial model which is defined by using a Markov chain structure torepresent a priori knowledge associated to labelling image. In the Markovian approach, we demonstrate that Markovian property is preserved after fusion, which enables us to apply the standard classification algorithms. We adopt an unsupervised context in which parameters estimation is done by using a mixture distribution algorithm, the ICE algorithm. We apply these models to satellite images.

Session TuB5: Civilian applications

Chair: Andreas Nürnberger, University of Magdeburg, Germany
Co-Chair: Jorge Marx-Gómez, University of Magdeburg, Germany

1 - Improvements of Pattern Recognition by using Evidence Theory. Application to Tag Identification.

Fabien Belloir, Alain Billat, Université de Reims Champagne-Ardenne, Reims, France

Abstract: In this paper we describe the improvements provided to a pattern recognition task by the use of the evidence theory when combining different classifier results. The application of this method concerns the identification of buried metal tags detected by an eddy current sensor. These tags are characteristic of the different contents (gas, water, …) of the buried pipes. We have developed classical, fuzzy and neural classifiers, each one giving a confidence level relatively to its decision. We show in this paper that an appropriate mass distribution coupled with a classical combination rule, without any a priori knowledge, provide a more important increasing of the performances than that obtained by the application of a simple weighted voting method.

2 - Performance Evaluation of a Fuzzy Fusion System for Subsoil Classification

L. Valet, Gilles Mauris, Philippe Bolon, LAMII / CESALP, Annecy, France
Naamen Keskes, ELF Aquitaine, Pau, France

Abstract: An information fusion system is proposed in this paper in order to segment seismic images in geological regions. The fusion of the attributes considered is made according to image interpreter knowledge, coded in the fuzzy subset theory formalism. The focus is on the problem of a quantitative performance evaluation of the fusion system according to the interpreter qualitative assessment, which has been translated into three main properties. The Baddeley distance is a potential operator for such a performance evaluation and the interesting results obtained with it are presented. But further developments are needed tocalibrate the Baddeley distance more precisely with interpreter behavior and thus to optimize the fuzzy fusion system in an automatic way.

3 - Model Based Fusion of Laser and Camera: Range Discontinuities and Motion Consistency

Jonas Nygårds, Åke Wernersson, Swedish Defence Research Establishment, Linköping, Sweden

Abstract: Consider a robot to measure or operate on man made objects randomly located in the workspace. The optronic sensing onboard the robot are a scanning range measuring time-of-flight laser and a CCD camera. The plane surfaces are modeled and parmeters extracted using the Radon/Hough transform. This extraction is very robust and motion is also included in a natural way. This paper gives additional results for range discontinuities. A multiple model framework for fusion of sensor information from laser and camera using parametric models of planar and cylindrical surfaces is suggested. An important issue is the mutual consistency between the motion, the range discontinuitym, occlusion and properties of the sensor combination. Typical applications are; Robust features for use during navigation in cluttered areas. Models for verification and updating of CAD-models when navigating inside buildings and industrial plants. Accumulating sensor readings into a map during operation of a telecommanded robot.

4 - Hybrid Approach to Forecast Returns of Scrapped Products to Recycling and Remanufacturing

Jorge Marx-Gómez, Claus Rautenstrauch, Andreas Nürnberger, Rudolf Kruse, University of Magdeburg, Germany

Abstract: Forecasting of scrapped products to recycling poses severe problems to recycling and remanufacturing companies due to uncertainties in available data. In this paper an extended prediction method to forecast return values (amount and time) of scrapped products to recycling is presented. The suggested model is based on important influencing factors and product life cycle data and has been applied to a case study (photocopiers) for evaluation. The approach employs a simulation study, the design of a fuzzy inference system for the prediction of the return in a specific planning period and the design of a neuro-fuzzy system for the prediction of return values with respect to time.

Session TuC1: Tracking Systems - 2

Chair: Mohamad Farooq, Royal Military College, Canada
Co-Chair: Thiaglingam Kirubarajan, University of Connecticut, Storrs, CT, USA

1 - Mobile Radar Bias Estimation Using Unknown Location Targets

Yaakov Bar-Shalom, University of Connecticut, Storrs, CT, USA

Abstract: In target tracking systems using radars on moving platforms the locations of these platforms is available from GPS based estimates. However, these estimated locations are subject to errors that are, typically, stationary autocorrelated random processes, i.e., slowly varying biases. In situations where there are no known-location targets to estimate these biases, the next best recourse is to use targets of opportunity at fixed but unknown locations. It is shown that these biases can be estimated in such a scenario, i.e., they meet the complete observability condition. Following this, the achievable accuracy for a generic scenario is evaluated. It is shown that accurate georegistration can be obtained even with a small number of measurements.

2 - Bias Modeling and Estimation for GMTI Applications

Keith Kastella, B. Yeary, Veridian ERIM International, Ann Arbor, MI, USA
T. Zadra, R. Brouillard, E. Frangione, Orincon Corporation, San Diego, CA, USA

Abstract: This paper describes an approach to sensor bias modeling and estimation for ground target tracking applications using multiple airborne Ground Moving Target Indicator (GMTI) radar sensors. This approach was developed as part of the Precision Firecontrol Tracking (PFCT) segment of the DARPA Affordable Moving Surface Target Engagement (AMSTE) program. For airborne sensors, slowly varying platform location, heading and velocity errors lead to time-dependent measurement biases. Track accuracy can be improved by using a Kalman filter to estimate and correct the biases in real time, based on fixed reference points. The reference point location can be known a priori or estimated online as part of the bias correction algorithm. When the reference locations are known a prior, bias effects can be nearly completely eliminated. When the reference point is estimated online, significant performance improvement is obtained relative to uncorrected measurements.

3 - Over-the-Horizon Radar Multipath Track Fusion Incorporating Track History

Peter W. Sarunic, Mark G. Rutten, DSTO, Salisbury, Australia

Abstract: This paper describes an algorithm for associating and fusing multipath tracks in over-the-horizon radar (OTHR). The algorithm extends earlier work by using a model based approach to incorporate track history in its computation of association probabilities and fused estimate calculations, thus exploiting temporal as well as spatial relationships. The algorithm can be easily extended to achieve asynchronous fusion of non-OTHR tracks (e.g. microwave radar or GPS) with the multipath OTHR tracks.

4 - An Application of Generalized Least Squares Bias Estimation for Over-The-Horizon Radar Coordinate Registration

William C. Torrez, SNWSC, San Diego, CA, USA
Erik Blasch, Air Force Research Lab, WPAFB, OH, USA

Abstract: Target and sensor geometry for a particular suite of over-the-horizon radars and a single target are given. Systematic positional differences between tracks seen from two separate radar sites can be used to improve the estimation of ionospheric parameters. In this paper a description of over-the-horizon radar propagation is provided and a method, given the target/sensor geometry, is described for estimating the range and azimuth biases resulting from errors in modeling the ionospheric fluctuations. Using the Generalized Least Squares Estimation method, a quantitative analysis of the improvements in tracking target positions in regions of overlapping coverage is given .

Session TuC2: Target Tracking - 4 – Track Fusion - 2

Chair: James Llinas, Center for Multisource Information Fusion, NY, USA
Co-Chair: Ivan Kadar, Consultant, Northrop Grumman Corporation, USA

1 - Radar/ESM Tracking of Constant Velocity Target: Comparison of Batch (MLE) and EKF Performance

Isabelle Leibowicz, Philippe Nicolas, Laurent Ratton, Thomson-CSF/DETEXIS, Elancourt, France

Abstract: In this paper we provide a comparison of performance of batch (Maximum Likelihood Estimate MLE) and iterative (Extended Kalman Filter) techniques in the case of Radar/ESM tracking. Simulation results are obtained on a constant velocity single target scenario. Several issues are addressed, including sensor noise level, coordinate system influence (cartesian and Modified Polar Coordinate EKF), Radar activation scheduling, and we compare tracking accuracy to Cramer-Rao Lower Bound.

2 - Track Fusion of Distributed EFRLS State Estimators

Yunmin Zhu, Keshu Zhang, Sichuan University, Chengdu, Sichuan, China
X. Rong Li, University of New Orleans, LA, USA
Zhisheng You, Sichuan University, Chengdu, Sichuan, China

Abstract: We present two track fusion methods for distributed recursive state estimators of dynamic systems without knowledge of noise covari-ances. This estimator at every local sensor is to incorporate the dynamic matrix and the forgetting factor into the Recursive Least Squares (RLS) method to remedy the lack of knowledge of noises, which has been developed before and was called the Extended Forgetting Factor Recursive Least Squares (EFRLS) estimator. We prove that the aforementioned fusion methods are both exactly equivalent to the corresponding centralized EFRLS that use all measurements from local sensors. Therefore, the two track fusion methods have the same advantages as the corresponding centralized EFRLS does, such as they can perform a little bit worse than precisely specified Kalman filter (see the simulations in a referenced paper and this paper) and still well even if there exists correlation between sensors and cross-correlation between the process and measurement noise streams or temporal dependencies within those streams (cf. simulations in a referenced paper).

3 - Credibilist Multi-Sensors Fusion for the Mapping of Dynamic Environment

Dominique Gruyer, Cyril Royère, Véronique Berge-Cherfaoui, Heudiasyc-UMR CNRS, Université de Technologie de Compiègne, France

Abstract: In this article, we present how, starting from an credibilist multi-objects association algorithm we can carry out an multi-sensors fusion algorithm. The tracking algorithm carries out an data association between predicted information and observations. These information are imperfect. The algorithm takes into account the inaccuracy and the uncertainty of the data and the reliability of the sensors. Association is realized with the belief theory. This method can be applied to the fusion of several homogeneous data sources. The problem arises when information are heterogeneous. Here, we answer to this problem by using a decentralized architecture which breaks up into two stages. The first consists in having at first a local processing to each sensor. This local processing makes it possible to obtain a set of homogeneous data. The second stage uses these homogeneous data to carry out a global fusion. This fusion gives a representation and a global view of a dynamic environment around a reference vehicle the most faithful and most reliable by using all available information. Moreover, the very general approach employed shows the polyvalence of this algorithm which can be in any case used for the multi-objects association, the local tracking, the multi-sensors fusion and the global tracking.

4 - Adaptive Track Fusion in a Multisensor Environment

Céline Beugnon, Tarunraj Singh, Mech. and Aerospace Engineering, SUNY at Buffalo, NY, USA
James Llinas, Center for Multisource Information Fusion, SUNY at Buffalo, NY, USA
Rajat K. Saha, Nova Research Corporation, Burlington, NY, USA

Abstract: The aim of this paper is to derive an adaptive approach for track fusion in a multisensor environment. The measurements of two sensors tracking the same target are processed by linear Kalman filters. The outputs of the local trackers are sent to the central node. In this node, a decision logic, which is based on the comparison between distance metrics and thresholds, selects the method to obtain the global estimate. Numerical simulation assess the influence of the thresholds and of the sensor noise ratio on the adaptive algorithm performance. The values of the thresholds govern the trade-off between accuracy and computational burden. The main advantage of the adaptive fusion is its ability to react to changes in the sustem characteristics.

Session TuC3: Image Fusion and Exploitation - 2 – Invited Session

Chair: Allen Waxman, MIT Lincoln Laboratory, USA
Co-Chair: William Streilein, MIT Lincoln Laboratory, USA

1 - Fusion Techniques for Environmental Database Construction

T. Laurençot, Thomson-CSF/ISR, Malakoff, France

Abstract: This talk will cover both 3D site modeling and terrain modeling using fusion of interferograms (obtained from radargrammetry) and disparity maps (derived from stereo imaging).

2 - Multi-Sensor 3D Image Fusion and Interactive Search

William Ross, Allen Waxman, William Streilein, J. Verly, Fang Liu, Michael Braun, Paul Harmon, and S. Rak, M.I.T. Lincoln Laboratory, Lexington, MA, USA

Abstract: Remotely sensed image datasets are growing rapidly with the ever-increasing variety, number, and coverage area of both airborne and satellite sensors. The timely exploitation of these data requires techniques for efficient visualization and search of multi-dimensional site information. Our approach has involved the development of biologically inspired algorithms for fusing multi-sensor imagery into unified color visualizations and for rapid search for multi-sensor signatures. Combined with interactive interfaces for 3D site fly-through, client requests for perspective views, and interactive training of search agents, these algorithms enable the efficient exploitation of a variety of challenging multi-sensor datasets. This first talk, of two that together present the entire system, will focus on describing and demonstrating our image fusion algorithms and web-based 3D site visualization interfaces, as well as fused signature searches for objects and infrastructure across site imagery.

3 - Fused Multi-Sensor Image Mining for Feature Foundation Data

William Streilein, Allen Waxman, William Ross, Fang Liu, Michael Braun, M. Aguilar, J. Verly, M.I.T. Lincoln Lab., Lexington, MA, USA
Chung Hye Read, National Imagery and Mapping Agency, Reston, USA

Abstract: The exploitation of remotely sensed multi-sensor imagery for agricultural, military, and civilian applications has become an important research area in recent years. Space-borne imaging satellites and air-borne sensors continue to produce an ever-increasing amount of data requiring timely investigation. In many applications it is necessary to be able to efficiently ‘mine’ this imagery for significant image features, such as roads, rivers, forests, and orchards, also known as Feature Foundation Data. In this talk we present our interactive Site Mining tool and the Image Map Interface, which together provide a powerful means by which an analyst can efficiently and successfully mine multi-sensor imagery for Feature Foundation Data in a web-based client-server environment. The Site Mining tool, based upon the fuzzy ARTMAP neural network, provides a point-and-click environment that enables the user to perform real-time mining of sensor-fused and contour- and texture-enhanced imagery in the context of color-fused imagery. The system reports detection confidence measures and indications of the relative utility of individual bands of the input data. The learned input vector weightings of the neural network classifier can be utilized to enhance the visualization of search targets in the color-fused imagery. The Image Map Interface enables the visualization of both raw and processed search results (e.g., road centerlines, bounded forest regions) and textual and graphical annotations in the context of geo-referenced color-fused imagery. We will demonstrate the use of the Site Mining tool and the Image Map Interface on commercial multispectral and hyperspectral image sets.

Session TuC4: Random Sets and Fuzzy Information

Chair : Shozo Mori, Information Extraction & Transport, VA, USA
Co-Chair : Gleb Beliakov, Deakin University, Australia

1 - Numerical Construction of Membership Functions and Aggregation Operators from Empirical Data

Gleb Beliakov, Deakin University, Clayton, Australia.

Abstract: A good choice of membership functions and aggregation operators is crucial for the behaviour of fuzzy systems. Goodness of fit to empirical data and flexibility in modelling various situations are the main criteria used by developers. This paper provides a general method for non-parametric representation of membership functions and aggregation operators using constrained spline functions. Tensor product monotone splines are used to approximate aggregation operators directly, while univariate splines are used to approximate their additive generators. Examples based on published empirical data are provided.

2 - Rule Discovery Based on Rough Set Theory

Yanyi Yang, Tee Chye Chiam, Nanyang Technological University, Singapore

Abstract: Nowadays , we are facing a big challenge to deal with huge amounts of data, how to extract useful information from it is an important issue. Rough set theory is a new mathematical approach to data analysis. Rough set theory is based on classification, it offers two fundamental concepts : reduct and core. In the paper, some basic ideas of rough set theory are presented and a new heuristic approach we used for rule induction is outlined by an illustrative example and the experiment results are also given.

3 - On a Family of Fuzzy Measures for Data Fusion with Reduced Complexity

Vicenç Torra, Institut d'Investigació en Intelligéncia Artificial, Catalunya, Spain

Abstract: Choquet integrals are one of the appropriate methods for fusing numerical information. They aggregate numerical values with respect to a fuzzy measure, a way to represent importance that is an alternative to the weights in a weighted mean. The use of fuzzy measures, although extremely flexible when compared with weighting vectors, presents some difficulties when used in real applications: to define a fuzzy measure to combine n values, 2n - 2 parameters have to be settled. In this work we present a family of fuzzy measures with reduced complexity and we show that they are either adequate for redundant information sources or complementary ones.

4 - Random Sets in Data Fusion: Formalism to New Algorithms

Shozo Mori, Information Extraction & Transport, Arlington, VA, USA

Abstract: Although connection between multi-target tracking and the random set theory was recognized during the course of development of multi-hypothesis tracking algorithms, it was only recent that such connection started to be discussed on a theoretical base and to be related several algorithms based on it. This paper describes the random-set formalism of a general theory of multi target tracking that was developed about twenty years ago, discusses recent theoretical and application developments, and explores further applications of random-set theory to data fusion.

Session TuC5: Medical Applications

Chair: Basel Solaiman, ENST Bretagne, Brest, France
Co-Chair: Isabelle Bloch, ENST Paris, France

1 - Belief Function in Low Level Data Fusion: Application in MRI Images of Vertebra.

Laurent Gautier, Abdelmalik Taleb-Ahmed, Michèle Rombaut, Jack-Gérard Postaire, H. Leclet, Lab. d'Analyse des Systèmes du Littoral, Calais, France

Abstract: The work presented in this article was sponsored by the department of radiology of the "Institut Calot de Berck sur Mer". It was done in order to help doctors to monitor patients with spinal diseases. The objective is the reconstruction of each vertebra of the lumbar spine from a series of parallel views. From an initial segmentation, we are looking for the part of the image that better represents the vertebra anatomical contour, in order to give doctors a belief degree oneach part of this segmentation. To find the point of the cortex, the information of low level is used : gray level associated with spatial constraints. The originalities of the work are:

  • in the low level fusion process,
  • in the choice of the discriminating parameters for the expertise,
  • in the construction of the belief functions.

This allows us to obtain the most reliable decisions which are illustrated by experimental results.

2 - Fusion of Heterogeneous and Noisy Informations: Application to the Quantification of the Coronary Stenosis

Patrick Franco, Michel Menard, Pierre Loonis, Université de La Rochelle, France

Abstract: We propose an algorithm in order to evaluate the similarity between two space-time distributions. One is obtained by experiment, the other is estimated by a numerical calculus. These informations are heterogeneous; their location so as their density and their reliability are various. We have developed a first method which is right when nu-merical and experimental informations are closely linked. Nevertheless in real world problems, the initial conditions which induce the numerical information are vague. For the non linearity of the studied phenomena, the similarity degree of both informations are deeply degraded. Our approach is robust relatively to this noise. It is a part of an identification process of a coronary stenosis.

3 - Fuzzy Fusion and Belief Rethinking. Application to Esophagus Wall Detection on Ultrasound Images

Renault Debon, Basel Solaiman, Christian Roux, ENST Bretagne, Brest, France
J-M. Cauvin, LaTIM EA-2218, CHRU Morvan, Brest, France
M. Robazkiewcz, LaTIM EA-2218, CHRU La Cavale Blanche, Brest, France

Abstract: In medical ultrasound imaging, information is corrupted by inaccuracy (due to data, acquisition modality, noise), uncertainty (due to noise and missing data) and ambiguity (several anatomical structures having the same ultrasound respond). In this work, we propose a 3D segmentation method of esophagus inner and outer wall from endosonographic sequences (composed of separate slices uniformly distributed), which minimizes these information alterations thanks to the cooperation of different models. The proposed solution is based on the use of a stochastic optimization algorithm, fully adapted to our particular case: the goal is to find the optimal surface, which verifies regularity conditions and maximizes a given criteria. Moreover, this approach cooperates with a data fusion based processing, which allows a prior knowledge integration with its own inaccuracy. All these components are integrated in a coherent architecture hierarchically organized which allows belief rethinking. First results obtained on real images acquired in a medical center are presented.

Session TuD1: Situation Assessment

Chair: Stéphane Paradis, DRE Valcartier, Val-Bélair, Québec, Canada
Co-Chair: Driss Kettani, DRE Valcartier, Val-Bélair, Québec, Canada

1 - Fusion of Radar and EO-sensors for Surveillance

L.J.H.M. Kester, A. Theil, TNO Physics and Electronics Lab., The Hague, The Netherlands

Abstract: Fusion of radar and EO-sensors for the purpose of surveillance is investigated. All sensors are considered to be co-located with respect to the distance of the area under surveillance. More specifically, the applicability for such multi-sensor systems is examined for surveillance in littoral waters is examined. The sensor suite is a coherent polarimetric radar in combination with a set of camera’s sensitive in visible light, near infrared, mid infrared and far infrared. A fuse while track algorithm is the best candidate for these dissimilar co-located and not necessarily synchronized sensors.

2 - Fusion of Radar Tracks, Reports and Plans

O. M. Mevassvik, Arne Løkka, Norwegian Defense Research Establishment, Kjeller, Norway.

Abstract: This paper suggests a method that utilises non-real time information as an aid to improve maritime surveillance. Under certain conditions vessels move in accordance with preplanned routes and possibly also report their own position at certain positions or at certain times during the voyage. The method proposed consists of a statistical route model that describes the movement of the vessel, and includes refinement of the estimated movement based upon reports on the vessel. The estimated movement is then associated with radar tracks using multiple hypothesis techniques. This is due to the fact that the number of vessels with known route plans is small compared to the total number of vessels in an area. The associated radar tracks are also used to improve the estimated movement of the vessels. Generating possible solutions and selecting the best hypothesis is formulated as a constraint satisfation problem and implemented using a constraint programming technique.

3 - A Qualitative Spatial Model for Information Fusion and Situation Analysis

Driss Kettani, Jean Roy, DRE Valcartier, Val-Bélair, Québec, Canada.

Abstract: In this paper, we present a Qualitative Spatial Model that is particularly suitable for Information Fusion and Situation Analysis. Information Fusion and Situation Analysis are processes that lead to situation awareness. Many studies have shown that, in order to support the officer in gaining its situation awareness, a Situation Analysis Model must ensure a cognitive fit between the officer's mental approach and the system's interactions and processing. Spatial Reasoning is one of the main mental processes that the Officer performs to analyze a situation. It allows evaluating many key information such as objects' location, disposition, arrangement, distance, etc. that are required to assess a situation. Spatial Reasoning of the officer is mainly qualitative, so a Qualitative Spatial Model seems to be ideal to ensure a cognitive fit with the Officer' Spatial Model. In DREV, we have elaborated a Qualitative Spatial Model that is inspired from the human spatial reasoning approach and that it is particularly well suitable for the situation analysis process. It is based on the concept of influence area, which is a portion of space that people build around objects in order to contextually reason about space, evaluate metric measures, qualify positions and distances, etc. We use the concept of influence area to formally define major spatial relations such as neighborhood, distance and orientation, which are necessary to elaborate a spatial model. We show why and how our model is well appropriate to perform the situation analysis process with regard to the cognitive fit constraint. Finally, we describe other military applications that can benefit from such model.

Session TuD2: Target Tracking 5 - Data Association

Chair: Jean Dezert, ONERA, Châtillon, France
Co-Chair: X. Rong Li, University of New Orleans, LA, USA

1 - Data Association with Believe Theory

Cyril Royère, Dominique Gruyer, Véronique Cherfaoui, Heudiasyc, CNRS, Université de Technologie de Compiègne, France

Abstract: In this paper, we present a method based on believe theory to realise the identification of an object. We consider applications where the size of the frame of discernment is large and we propose generalisation for believe mass computing. In order to tacking into account of unknown hypothesis, we introduce a new framework for Dempster’s combination: it is called the extended open world. This framework offers the possibility to have an opinion about the conflict between the experts and about the opportunity to introduce a new hypothesis in the frame of discernment. Some results highlights advantages of this framework in the case of pignistic decision.

2 - An LP-based Algorithm for the Data Association Problem in Multitarget Tracking

P. Storms, Hollandse Signaalapparaten B.V., Hengelo, The Netherlands
F. Spieksma, University of Maastricht, The Netherlands

Abstract: In this work we present a linear programming (LP) based approach for solving the data association problem (DAP) in multiple target tracking. It is well-known that the DAP can be formulated as an integer program. We present a compact formulation of the DAP. To solve practical instances of the DAP we propose an algorithm that uses an iterated K-scan sliding window technique. In each iteration we solve the Linear Programming relaxation of an integer program and next apply a greedy rounding procedure. Computational experiments indicate that the quality of the solutions found is quite satisfactory.

3 - Data Association through Fusion of Target Track and Identification Sets

Erik Blasch, Air Force Research Lab, WPAFB, OH, USA
Lang Hong, Wright State University, Dayton, OH, USA

Abstract: A joint probability data association tracking algorithm typically associates only position measurements. With multiple-interacting targets in the presence of clutter, data association can be confused by spurious measurements. In this paper, we propose a set-based track and identification data association (SBDA) technique to leverage object identification information. We investigate the SBDA technique for a scenario in which a tracker has access to both coarse position measurements and belief identification information to enhance data association.

4 - Track Formation in Clutter Using a Bi-Band Imaging Sensor

Jean Dezert, ONERA, Châtillon, France
Thiaglingam Kirubarajan, University of Connecticut, Storrs, CT, USA

Abstract: In this paper we present an extension of the Markov-chain-based performance evaluation technique for a bi-band two-stage sliding window cascaded logic 2/2 x m/n for track formation in clutter. This work has been motivated by a ballistic target surveillance problem based on a bi-spectral satellite observation system. We show how to combine an AND and OR fusion decision logic within the classical performance evaluation approach and how this can result in better performance and serve as a useful tool in satellite tracking system design.

Session TuD3: Image Fusion and Exploitation - 3 – Invited Session

Chair: Allen Waxman, MIT Lincoln Laboratory, Lexington, MA, USA
Co-Chair: William Streilein, MIT Lincoln Laboratory, Lexington, MA, USA

1 - Fusion of Multi-Sensor Imagery for Night Vision: Color Visualization, Target Learning and Search

David A. Fay, Allen Waxman, M. Aguilar, D.B. Ireland, J.P. Racamato, W.D. Ross, W.W. Streilein, M.I. Braun, M.I.T. Lincoln Laboratory, Lexington, MA, USA

Abstract: We present methods and results for fusion of imagery from multiple sensors to create a color night vision capability. The fusion system architectures are based on biological models of the spatial and opponent-color processes in the human retina and visual cortex, implemented as shunting center-surround neural networks. Real-time implementation of the dual-sensor fusion system combines imagery from either a low-light CCD camera or a short-wave infrared camera, with thermal long-wave infrared imagery. Results are also shown for extensions of this fusion architecture to include imagery from all three of these sensors, Visible/SWIR/LWIR, as well as a four sensor system fusing imagery from Visible/SWIR/MWIR/LWIR cameras. We also demonstrate how the results from these multi-sensor fusion systems are used as inputs to an interactive tool for target designation, learning, and search based on a Fuzzy ARTMAP neural network.

2 - Image Fusion of High Resolution LWIR and IITV Sensors for Pilotage

Anthony L. Leatham, Luan Do, Raytheon Electronic Systems, McKinney, TX, USA

Abstract: Infrared, image intensified, and low light level CCD have well recognized uses, capabilities and limitations. Several government and industry studies objectively evaluated the relative merits of these sensors as applied to the day and night pilotage missions. These studies found that each sensor excelled under different conditions and environments. Most pilots preferred having at least two different types of sensors available, since they are sometimes complement each other. The ultimate goal of image fusion is to provide an automated method integrating the various image information from different sensors to yield a high quality real-time presentation. Ideally, such a composite should retain the essential information from each sensor while minimizing any artifacts or distortions so that the end result is a seamless representation of reality. By putting together several technologies, image fusion offers an overall improved single image representation of thermal, visible and color, etc.

3 - Perceptual Evaluation of Different Night-time Imaging Modalities

Alexander Toet, N. Schoumans, J.K. Ijspeert, TNO Human Factors, The Netherlands

Abstract: Human perceptual performance was tested with images of night-time outdoor scenes registered both with a dual-band image intensified low-light CCD camera (IICCD), and with thermal middle (3-5 micron) and long (8-10 micron) wavelength band infrared (IR) cameras. Fused imagery was produced by combining the individual bands using different color and grayscale fusion schemes. The number of correct responses and the reaction time of human subjects was measured for each (individual and fused) type of imagery, and for different observation tasks ranging from global scene recognition (situational awareness) to the perception of small details (target detection). The results show that the IICCD imagery contributes most to global scene recognition, horizon detection, and the identification of water, roads and buildings. IR imagery serves best for the detection and recognition of humans and vehicles. Color fused imagery yields the best overall scene recognition performance.

Session TuD4: Fuzzy Mathematical Programming for Fusion – Invited Session

Chair: Mustafa Günes, University of Dokuz Eylül, Buca-Izmir, Turkey
Co-Chair: Vedat Pazarlioglu, University of Dokuz Eylül, Buca-Izmir, Turkey

1 - Fuzzy Approaches to the Production Problems: the Case of Refinary Industry

Mustafa Günes, University of Dokuz Eylül, Buca-Izmir, Turkey

Abstract: The Fuzzy principle states that everything is a matter of degree. So far many business production problems solved by Operational Research Optimization Techniques,under the considerations of some assumptions. In the current literature,still we have several applications of fuzzy linear,integer,goal and other programming applications.The main aim of this study is to add new application to the literature and to solve the refinary production problem by using the fuzzy principles. In application the real refinary model developed and an alternative fuzzy model solutions critisized to determine which one is better then the others. Finally, comparing the classical solution by the one of the best solution of the Fuzzy Models displayed that,one can obtain more suitable output of the models than traditionals.

2 - Fuzzy Multiplecriteria Assignment Problems for Fusion on Hungarian Algorithm

Ibrahim Güngör, University of Süleyman Demirel, Isparta, Turkey
Mustafa Günes, University of Dokuz Eylül, Buca-Izmir ,Turkey

Abstract: In reality , it is possibility to encounter the assignment problems including multiple purposes and whose purposes featuring in a fuzzy way . In that research, 0-1 linear goal programming models of fuzzy multiple criteria assignment problems representing different-structured purposes are made up . Furthermore , in searching Hungarian algorithm that the solution of classic assignment problems obtained by chancing Cij coefficients suitably according to fuzzy purposes in some fuzzy multiplecriteria assignment problems .

3 - The Hedonic Price Index Model for Fusion on Car Market

Vedat Pazarliolu, Mustafa Günes, University of Dokuz Eylül, Buca-Izmir, Turkey

Abstract: This paper involves two subparts: in the first session of the study research focused on the associations between price and other effective variables. So that in this part, main aim objective is to determine how changes the automobile prices then to suggest the optimum decision criteria to the buyers. In literature this kind of model studies are called as "Hedonic Price Model". In real applications we always carry out the ambiguity and fuzziness. In the second part of study the developed "Hedonic Price Model" has been fuzzified and new addition decision information presented for the buyer of cars. At the end of study, result of analysis displayed that the Fuzzy Hedonic Price Index Model for fusion will also give more widely vision then classical approaches.

4 - Aggregating Truth and Falsity Values

Marcin Detyniecki, Bernadette Bouchon-Meunier, LIP6, University of Paris VI, Paris, France

Abstract: The problem of aggregating truth values is at the core of the studies in fuzzy logic. But it is to notice that the purpose of this aggregation is to compute the truth value of a logical phrase. Here we are interested in the aggregation of different truth values observed for the same logical phrase. We propose an axiom set for the aggregation of truth values, which leads to the characterization of two truth-aggregation families, a prudent and an enthusiastic. The first one has a cautious attitude choosing between two observed values the one which is more uncertain. The second one has an enthusiastic behavior and will reinforce the result if it observes twice the truth or twice the falsity. When observing falsity and truth the operator gives a compensated value. We finish by expounding the use of these operators and their relationship with the traditionally used truth-aggregation operators: the t-norms and t-conorms. Actually the presented operators should be used for the aggregation of different observed truth values for the same phrase vs. the calculus of the truth of a logical phrase.

Session TuD5: Fault Diagnosis and Condition Monitoring

John MacIntyre, University of Sunderland, UK

1 - D-S Evidence Theory Applied to Fault Diagnosis of Generator Based on Embedded Sensors

Du Qingdong, Xu Lingyu, Zhao Hai, Northeastern University, Shenyang City, China.

Abstract: In the monitoring system of power plant, method of gathering real time data of sensors is often adopted. It not only increases the communication burden of monitoring system but also results in error transmitting due to worse electromagnetism environment; For conquering these shortcomings, we adopt an new approach --- using embedded multisensors and D-S evidence theory. This method has been applied in the monitoring system of JiLin FengMan Power Plant successfully.

2 - Fusing Diagnostic Information Without A Priori Performance Knowledge

M. Garbiras, K. Goebel, GE Corporate Research and Development, Niskayuna, USA

Abstract: Diagnostic information fusion is the method by which one would determine a system’s state for those instances where several different diagnostic tools, and possibly other sources, are used for state estimation. Because system state predictions from different diagnostic tools will disagree at some extent, if not completely contradict one another, a robust fusion tool is necessary to produce a reliable assessment of system state. This paper addresses the need for a reliable solution to the problem of diagnostic information fusion, particularly with the absence of a priori knowledge of diagnostic tool performance. Tool performance specifications are often times hard to come by, in particular where data about events are sparse or where e comprehensive evaluation cannot be performed. In response, a fusion process, using a set of neural networks, was developed to distinguish recognizable patterns from the output of the individual diagnostic tools. We apply this fusion concept to data that were gathered from a high-speed milling machine and processed by several previously developed diagnostic tools.

3 - An Architectural System Solution for On-line Technical Diagnosis

Monica Alexandru, C. Bigan, Politehnica University of Bucharest, Romania

Abstract: In this article a generic distributed system architecture for process monitoring, fault diagnosis and assisted maintenance is proposed. The diagnosis system aims identifying failures as and when they happen in normal operation. It integrates different fault detection and isolation techniques such as model based methods, heuristic and rules-based reasoning. The actual developing supervisory intelligent systems has to:

  • detect and interpret the abnormal conditions that will cause an incident
  • determine what kind of action should be taken and resume the process to normal conditions
  • find reasons of equipment malfunctions and schedule a maintenance plan

4 - A Data Fusion Concept for a Query Language for Multiple Data Sources

Erland Jungert, Swedish Defense Research Establishment, Linköping, Sweden

Abstract: Query languages for multiple sensor data sources require a technique that allows fusion of the different types of sensor data that may be part of the queries. A method that generally can carry out the fusion process must therefore be developed. Such a process must be able to collect, transform and organize the information subject to fusion. It is of great importance that the uncertainties in the information are possible to deal with. Besides this the applied data fusion method should be replaceable. This paper describes a fusion process designed for the query language SQL.

bateau mouche

Tuesday, 11 - 8: 00pm

Conference dinner

Monday - Tuesday - Wednesday - Thursday

Last Updated: June 13, 2000
Web site by: dezert@onera.fr (content), gaultier@onera.fr (form)
copyright © ISIF 2000