Final Long Program

Monday - Tuesday - Wednesday - Thursday

Wednesday July 12

Session WeA1: Plenary talk

Data Fusion in the Transferable Belief Model

Prof. Philippe Smets, IRIDIA - Université Libre de Bruxelles, Belgium

Abstract: When Shafer introduces his theory of evidence based on the use of belief functions, he proposed a rule to combine the belief functions induced by distinct pieces of evidence. Since then, theoretical justifications of this so-called Dempster's rule of combination have been produced and the meaning of distinctness has been assessed. We will presents practical applications where the fusion of uncertain data is well achieved by Dempster's rule of combination. It is essential that the meaning of the belief functions used to represent uncertainty be well fixed, as the adequacy of the rule depends strongly on a correct understanding of the context in which they are applied. Missing to distinguish between the upper and lower probabilities theory and the transferable belief model can lead to serious confusions, as Dempster's rule of combination is central in the transferable belief model whereas it hardly fits with the upper and lower probabilities theory.
In order to illustrate the possibilities that are offered by the transferable belief model when it comes to uncertain data fusion, we present some practical applications. For each of them, the transferable belief model seems well adapted whereas the classical probability theory might encountered problems, usually because of some missing information that probability theory requires and that is not available or worse, not existent.

Session WeB1: Resource Management - 1

Chair: Jean-Pierre Le Cadre, IRISA/CNRS, Rennes, France
Co-Chair : Frédéric Dambreville, IRISA/CNRS, Rennes, France

1 - A Fuzzy Logic Resource Manager and Underlying Data Mining Techniques

James F. Smith III, Robert D. Rhyne II, Naval Research Lab., Washington, DC, USA

Abstract: A fuzzy logic based expert system has been developed that automatically allocates electronic attack (EA) resources in real-time over many dissimilar platforms. The platforms can be very general, e.g., ships, planes, robots, land based facilities, etc. Potential foes the platforms deal with can also be general. This paper describes data mining activities related to development of the resource manager with a focus on genetic algorithm based optimization. The use of a database of scenarios prevents the algorithm from having too narrow a range of behaviors, i.e., it creates a more robust solution. The approach to optimization is a type of co-evolution, i.e., both friend and foe agents simultaneously adapt within a complicated environment perceived through various sensors such as: radar, electronic support measures, etc. New rule classes and the five components of the resource manager are discussed. Finally, the resource manager’s multi-platform response is examined for multiple scenarios.

2 - Optimal Passive Receiver Location for Angle Tracking

Kaouthar Benameur, Surface Radar Section, Defence Research Establishment, Ottawa, Ontario, Canada

Abstract: This paper presents a class of optimization problems dealing with selecting at each instant of time a strategy of measurements for a passive receiver. The basic problem is then to compute an optimal policy, during a specified observation time interval so that a prediction accuracy is optimized. This problem of mesurement strategy computation can be transformed into a deterministic control problem. It is shown that the optimal measurement policy can be precomputed before the measurements actually occur. A detailed description of this measurement strategy is presented in the case of a receding horizon observation interval.

3 - Sensor Management - Control and Cue

Gee Wah Ng, Khin Hua Ng, L. T. Wong, DSO National Laboratories, Singapore

Abstract: This paper presents the role of sensor management with respect to sensor system and the fusion process. The important roles and functions of sensor management are discussed. Multi-level classification of the sensor management is presented. The control and cueing aspects of sensor management are explored. An experiment is set to demonstrate sensor manager controlling electro-optical (EO) sensor's moving direction and sensor cueing process to track a target. A fuzzy controller is used in this experiment. The fusion system used an interactive multiple model (IMM) algorithm to give the estimated target direction..

4 - Scheduling Active and Passive Measurements

Jean-Pierre Le Cadre, IRISA/CNRS, Rennes, France

Abstract: Many tracking systems involve basically active and pssive subsystems. If it can be reasonably assumed that passive measurements have no "cost", this is not true for active measurements. So, a general problem is to scheduling active measurements, so as to combine them optimally with the passive ones. More generally, we are interested by optimizing controls in the estimation procedure.

Session WeB2: Tracking Course - 1

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

Short Course: Multitarget Tracking and Multisensor Fusion

Session WeB3: Image Fusion - 2

Chair: Amy L. Magnus, Intelligent Information AFRL/IFTD, NY, USA
Co-Chair: Brian O'Hern, Air Force Research Laboratory, Rome, NY, USA

1 - Fast Stereo Matching Using Multilevel Enhancement

Jaeweon Kim, Shishir Shah, Wayne State University, Detroit, MI, USA

Abstract: Stereo matching is a key problem in the area of computer vision and photogrammetry. Stereo is a useful method for machine perception and recovering 3D shape. Given a point in the left image, the problem is to find its corresponding point in the right image, which typically has additional image variations. Most existing techniques use a fixed window for the purpose of calculation of the correlation coefficient values. The fixed window approach leads to mismatch in areas lacking features and those without much intensity variations. This paper presents a new approach of correct and fast calculation for stereo matching. Some of the efficient and robust implementation aspects of the stereo matching algorithms using thresholding correlation values of previous pyramid structure that reduce the search area for finding the best matching point is addressed. The disparity for each scan line is chosen by selecting the position that gives the maximum correlation coefficient from the pyramid structure. However, sometimes this maximum value may be overridden by a spurious maximum. Ambiguities mainly come from image noise, lack of intensity variation or geometric distortion. For disambiguation, the adaptive window theory is applied to avoid mismatch in featureless area. Adjusting threshold level at each level of the pyramid structure reduces computing time and increases the accuracy for the next pyramid level. Results obtained on both synthetic and real image sets are presented.

2 - A Multi-Level Bayesian Network Approach to Image Sensor Fusion

Amit Singhal, University of Rochester, NY, USA
Jiebo Luo, Christopher Brown, Eastman Kodak Company, Rochester, NY, USA

Abstract: Automatic main subject detection refers to the problem of determining salient or interesting regions in an image. We propose the use of a Bayesian network based approach to solving this problem in the unconstrained domain of consumer photographic images. Various image sensors, derived from classical computer vision literature, as well as other sources, can provide evidences about main subject regions in images. A traditional sensor fusion scheme, such as a Kalman filter, fuzzy logic, or simple Bayesian estimation, does not provide sufficient expressive power to capture the uncertainties and dependencies exhibited by such a system. We present a multi-level Bayesian network that accurately models the system and allows for sensor integration in an evidential framework. The multi-level Bayesian network performs better than a simple single-level Bayesian network at accurately combining various image sensor data to construct a belief map identifying main subject regions in the image. A subsequent study also shows that the multi-level Bayesian network performs better than a linear classification scheme as well as one based on neural networks.

3 - 3D Structure and Motion Recovery by Fusing Range and Intensity Image Sequences

Christophe Boucher, Jean-Charles Noyer, Mohammed Benjelloun, Université du Littoral Côte d’Opale, Calais, France

Abstract: We propose in this article a 3D dynamic reconstruction method based on the fusion of data from sequences of range and intensity images. The vision system acquires both images of the scene at every time. We use a line segment description of the objects of the scene, and more precisely a point representation of the segment described by its extremities. The detailed solution to this estimation problem lies on a global filter that matches the segments through the range/intensity image sequences and fuses both data to recover the 3D structure and motion of an object. The method performs well on synthetic image sequences from two sensors and is now applied to an experimental sequence of an object evolving according to a rotation/translation motion. The scene is viewed by a range camera which delivers a range image and an intensity (reflectance) image.

Session WeB4: Enabling Computation

Chair: H. Martínez Barberá, University of Murcia, Spain
Co-Chair: Roger Reynaud, IEF - University Paris XI, Orsay, France

1 - Ontology Based Design of Surveillance Systems with NUT

Vahur Kotkas, Jaan Penjam, Institute of Cybernetics at Tallinn Technical University, Estonia
Enn Tyugu, Royal Institute of Technology (KTH), Kista, Sweden

Abstract: This paper presents ontology-based programming using the NUT language as a notation for semantics of domain knowledge. The specification method and problem solving techniques are demonstrated on an example of modeling and management of a radar surveillance system in order to find optimal disposition and configuration of equipment. Structural synthesis of programs - a technique essential for the domain knowledge handling is briefly discussed.

2 - A Block-adaptive Blind Separation Algorithm for Post-nonlinear Mixture of Sub- and Super-Gaussian Signals

Yang Chen, Zhenya He, Southeast University, Nanjing, China

Abstract: The problem of blind separation of signal in post-nonlinear mixture is addressed in this paper. The post-nonlinear mixture is formed by a component wise nonlinear distortion after the linear mixture. hence a nonlinear adjusting part placed in front of the linear separation structure is needed to compensate for the didtortion in separating such sugnals. The learning rules for the post-nonlinear separation structure are derived by a maximum likelihood approach. An algorithm for blind separation of post-nonlinearly mixed sub- and super-Gaussian signals is proposed based on some previous works. Multilayer perceptrons are used in this algorithm to model the nonlinear part of the separation structure. The algorithm switches between sub- and super-Gaussian probability models during learning according to a stability condition and operates in a block-adaptive manner. The effectivness of the algorithm is verified by experiments on artificial and natural signals.

3 - Neural Networks for Sonar and Infrared Sensors Fusion

H. Martínez Barberá, A. Gómez Skarmeta, M. Zamora Izquierdo, J. Botía Blaya, University of Murcia, Spain

Abstract: The main goal of our work is to have a robot navigating in unknown and not specially structured environments, and performing delivery like tasks. This robot has both unreliable sonar and infrared sensors. To cope with unreliability a sensor fusion method is needed. The main problem when applying classical fusion methods is that there is no a priori model of the environment, just because the robot first carries on a map building process. There exist some simple methods for sensor fusion but, as we show, they do not address all the specific issues of our desired robot task. This way, we use neural networks for such fusion, and so we obtain more reliable data. We discuss some important points related to the training procedure of neural networks and the results we obtained.

4 - Expanded Researching on Knowledge Discovery System

Bing-ru Yang, Dezheng Zhang , Jing Tang, Ying Zhou, University of Science and Technology, Beijing, China

Abstract: The development of database technology, every-increasing needs of enterprise management for the decision-making support and the comprehensive application of AI are driving the theories and technology of data mining and knowledge discovery growing rapidly. In recent years, data mining and knowledge discovery has been paying more attention to and all kinds of algorithm and tools are blooming. Their characteristics are overlapping of subjects, fusion of several techniques, generalizing of data mining and integrating of knowledge discovery. This paper discussed the development trend of knowledge discovery based on the summary and analysis of the theory of knowledge discovery in database and the actuality of technological method, and introduced the mechanism of total data mining process, researching in expansible structure and algorithm of KDD, and serially developing of software, according to the theory of double base cooperation proposed by author

Session WeB5: Guidance and Navigation - 1

Chair: Christian Musso, ONERA, Châtillon, France
Co-Chair: Subhash Challa, University of Melbourne, Australia

1- Comparative Analysis of Alternative Ground Target Tracking Techniques

Chih-Chung Ke, State University of New York at Buffalo, NY, USA
Jesús Garcia Herrero, GPSS-SSR-ETSIT, Universidad Politécnica de Madrid, Spain
James Llinas, State University of New York at Buffalo, NY, USA

Abstract: There have been a lot of studies addressing target-tracking problems, in which targets like aircraft and missiles can move freely in the air without hard spatial constraints. Tracking ground targets is much more difficult. Variable terrain structures not only limit the target kinematics, but also degrade the quality of measurement data. We are particularly interested in two among several categories of trackers for point ground targets that have no spatial extent. The first is Kalman-based approach in which a number of studies taking advantage of terrain information, such as elevation and road, have been proposed. The second is based on the theory of hidden Markov model (HMM) in which targets are assumed to move among discrete spatial cells at discrete time instants. It has been known that the terrain information may be incorporated to retrieve the actual history of target locations. This paper focuses on tracking a single ground target via both types of approaches. Various Kalman techniques were implemented. When the ground information, a road network in our study, is unavailable, the conventional Kalman filter and an interacting multiple-model (IMM) estimator with two models had roughly the same performance in terms of estimation errors. To reduce the errors due to transversal maneuvers, we developed three methods to take road structures into account. The first was to tune the variance of the process noise, depending on if the on-road target was moving straight or was ready for making a turn or maneuver. The second method, named curvilinear model, was a better choice that considered the target trajectory as a circular arc during the transition between road segments. The stability problem might occur if the jump of road orientation is too high. The third, which outperformed the other two for transversal errors, involved an additional stage to preprocess the sensor measurements. Besides, an adaptive HMM tracker was also developed under the same scenario. To save computation time, a partial scenario, which covered the entire road path of interest, was partitioned into several subscenarios. Each of them was associated with an HMM where the Viterbi algorithm was performed with road-based transition array. Joining all subtracks from each HMM formed the complete track. It turned out that the adaptive HMM tracker produced small transversal errors but large longitudinal errors, compared to Kalman-based techniques. Generally speaking, Kalman-based trackers are known to be efficient and robust, whose performance can be improved by the proposed methods. The adaptive HMM-based tracker can freely incorporate any terrain features, target kinematics, and even military doctrines into the transition array such that the track can be more accurate.

2 - Road Detection and Vehicles Tracking by Vision for an On-Board ACC System in the VELAC Vehicle

R. Chapuis, F. Marmoiton, R. Aufrère, F. Collange, J-P. Dérutin, LASMEA/CNRS, Université Blaise Pascal, Aubiére, France

Abstract: This paper presents a method designed to detect and track vehicles on highway in a safety improvement purpose. The goal of this kind of system is to regulate the speed of a vehicle so as to respect safety distances relative to vehicles ahead. The method is exclusively based on monocular computer vision and uses two algorithms. The first one is able to locate the lane borders in the image, and to deduce the 3D shape of the road axis. The second algorithm detects, tracks and compute the 3D location of vehicles ahead by using fixed lights embedded on these ones. By combining the results of the two algorithms, a fusion step permits to know were are the most dangerous vehicle according to its position, speed and circulation lane. The method has been implemented onour experimental vehicle VELAC and the whole system operates in real-time conditions.

3 - Temporal Sequence Recognition Using Uncertain Sensor Data

Michèle Rombaut, CREATIS, Lyon, France
S. Loriette-Rougegrez , J.M. Nigro, LM2S UTT, Troyes, France
I. Jarkass, Université Libanaise, Institut Universitaire de Technologie, Saida, Liban

Abstract: The problem addressed in this paper concerns the temporal sequence recognition for a dynamic system. Several formal models can be used such as rule based systems, or graphs such as transition graphs or Petri nets in order to describe the sequences to be recognized. Then, according to the inputs got from the system's sensors at different times, the goal is to evaluate the confidence into the fact that the sequence isinprogress. In this paper, the confidence is modeled by a distribution of mass of evidence proposed in the Dempster-Shafer's theory.

4 - Recent Particle Filter Applied to Terrain Navigation

Christian Musso, Nadia Oudjane, ONERA, Châtillon, France

Abstract: A Recent particle method, the Local Regularized Rejection Particle Filter L2RPF is applied here in terrain navigation. An aircraft measures periodically the relative elevation. By means of a digital elevation map, the goal is to estimate the absolute position and velocity of the aircraft. Moreover, an inertial navigation system (INS) drives the generation of the particles (fusion of L2RPF and the INS). The conditional density of the state is recursively estimated. The proposed filter allows a precise correction step in a given computational time. For this problem, the Kalman filter is inadapted (multimodality) and batch methods are expensive (grid methods in a 6-dimentional state space).

Session WeC1: Resource Management - 2

Chair: Michel Prenat, Thomson-CSF Optronique, Guyancourt, France
Co-Chair : Véronique Cherfaoui, Heudiasyc CNRS, Compiègne, France

1- Selected Problems of MFR Resources Management

W. Komorniczak, J. Pietrasinski, Military University of Technology, Warsaw, Poland

Abstract: An idea of a Multi Function Radar resources management is described. In this kind of radar, in contrast to a classic solutions, a radar controller has to manage a lot of parameters and particularly, resources. Problems of radar resources management and its optimization are described in the paper. The multiplicity of limitations connected with resources to use, makes resources management process very complicated, but necessary. The resources to be controlled as well as optimization variables, parameters and goal function are characterized. The model of a system of radar resources controller simulation is presented. The priority assignment process for detected objects is discussed too. The methods of learning of priority assignment module are characterized.

2 - Detection with Spatial and Temporal Optimization of Search Efforts Involving Multiple Modes and Multiple Resources Management

Frédéric Dambreville, Jean-Pierre Le Cadre, IRISA/CNRS, Rennes, France

Abstract: This paper deals with optimization of splitable resources aimed to the detection of a moving target following a Markovian movement or a conditionally deterministic motion. Our work extends Brown's spatial optimization method. By use of a generalized linear formalism, we developed a method for optimizing both spatially and temporally (modeling resource renew), with management of multiple resource types or multi-modes resources. Such optimization involves also the fusion of several detection tools, in order to make them work together efficiently.

3 - Analysis of the Multisensor Multitarget Tracking Resource Allocation Problem

Pierre Dodin, Julien Verliac, Vincent Nimier, ONERA, Châtillon, France

Abstract: This paper deals with a study of the multisensor management problem. The main tool is the classical assignment formulation, using Kullback-Leibler entropy as costs. In order to use the benefit brought by the data fusion, coalitions or pseudo-sensors must be created at each step of time, creating an exponential calculus of all the possible sensor partitions. We compare this method and a predefinite strategy using different scenarios.

4 - Study of the Temporal Allocation of Two Passive Infrared Sensors in a Multitarget Environment

Marie De Vilmorin, Philippe Vanheeghe, Ecole Centrale de Lille, Villeneuve d'Ascq, France
Michel Prenat, Thomson-CSF Optronique, Guyancourt, France
E. Duflos, Ecole Centrale de Lille, Villeneuve d'Ascq, France

Abstract: The topic of this paper is the optimization of the management of the lines of sight of two infrared sensors in a multitarget environment. Some general trends for the elaboration of optimal strategies have been undercored by the study of a strategy based on the fusion of these passive sensors. The main result which has been used is the completion principle, taht is, the equivalence in Bearing Only Tracking (BOT) between the asymptotic behavior of the real maximum likelihood estimator and the one obtained by supposing a linear measurement model.

Session WeC2: Tracking Course - 2

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

Short Course: Multitarget Tracking and Multisensor Fusion

Session WeC3: Image Fusion - 3

Chair: Belur V. Dasarathy, Dynetics Inc., Huntsville, AL, USA
Co-Chair: Vladimir Petrovic, Manchester School of Engineering, UK

1 - A Real Time Pixel-Level Based Image Fusion Via Adaptive Weight Averaging

Eric Lallier, Mohamad Farooq, Royal Military College of Canada, Kingston, Ontario, Canada

Abstract: A novel pixel-level image fusion scheme for thermal and visual images is presented in this paper. The image fusion technique rests on physical characteristics of targets deemed of interest in a surveillance scenario. Each picture element (pixel), in both the thermal and visual images, is assigned a weight proportional the interest associated with it. Interest is defined as “not natural” or “man-made”. A weighted average of the intensity images representing the thermal and visual modalities is then performed for every corresponding pair of visual and thermal picture elements to obtain the fused image. For the thermal images, elements that are warmer or cooler than their environment (background) are deemed to be of “interest”. To this end, the thermal weights are associated with the divergence of the intensity of these pixels from the image mean intensity. For the visual images, the facts that the “targets of interest” are usually larger then the instantaneous field of view (IFOV) of the visual sensor and have a reflection behaviour that is more specular are used. The visual weight determination is based on the local variance in space and time of the intensity of the visual pixels. The performance of this technique is compared to a number of existing techniques in the literature. The results reveal that the proposed technique performs better than those in the literature. In addition, it also reveal that the proposed technique is more robust than those in the literature.

2 - On the Effects of Sensor Noise in Pixel-Level Image Fusion Performance

Vladimir Petrovic, University of Manchester, UK
C. Xydeas, University of Lancaster, UK

Abstract: This paper considers image fusion under the condition that input image quality is reduced by sensor noise. The aim is twofold: i) to develop appropriate metrics which measure the effect of input sensor noise on the performance of a given pixel-level image fusion system and ii) to employ these metrics in a comparative study of the robustness of typical image fusion schemes whose input is corrupted by sensor noise.

3 - More the Merrier … or is it ? - Sensor Suite Augmentation Benefits Assessment

Belur V. Dasarathy, Dynetics Inc., Huntsville, AL, USA

Abstract: It is often implicitly assumed in the information fusion field that augmentation of a multi-sensor suite with additional sensors defacto enhances overall system performance because of the increase in the data being input to the fusion process. In this study, an explicit assessment of the validity of this assumption is made in terms of delineating the sensor characteristics domain wherein this is true (i.e., where fusion benefits do indeed increase) and quantitatively determine the extent of such benefits. Initially for illustrative purposes, a two-sensor suite augmented by a third sensor is used as a case study for this assessment. The consensus fusion logic, which is symmetric relative to the multiple sensors in the sensor suite, is employed as an example in this assessment process. The scope for generalizing this assessment to higher dimensional multi-sensor suites as well as other types of fusion logic is also discussed.

Session WeC4: Symbolic and Numeric Information: Hybrid Approaches - 1 – Invited Session

Chair: Galina Rogova, CUBRC, Buffalo, NY, USA

1 - Peircean Semiotics: A New Engineering Paradigm for Automatic and Adaptive Intelligent Systems Design

E.T. Nozawa, Lockheed Martin Aeronautics Company, Marietta, Georgia, USA

Abstract: The intent of this paper is to bring before the Information Fusion community highlights of Peircean Semiotics as defined by Charles Sanders Peirce. An attempt will be made to show that Peircean Semiotics has potentially a very revolutionary role to play in the further development of Information Fusion and in the advanced development of Artificial Intelligence, Cognitive Science, Natural Intelligence Science, and Information Processing Science and Technology in general. The transition to the Information Age and the explosive growth in knowledge is occurring at a time when the traditional humanistic philosophies and scientific philosophies have run their course and are unable to provide adequate guidance for the development of Advanced Automated Reasoning-Based Belief systems that are simultaneously Open-Systems and Human-Centered Systems.

2 - Semeiotic Data Fusion

Robert W. Burch, Texas A&M University, College Station, TX, USA

Abstract: Approaches to problems of data fusion typically involve methods, such as statistical inference and Baysian analysis, that are quantitative in nature. Quantitative methods, however, are often poorly suited for dealing with types of data that are essentially qualitative and relational in nature, for example, organizational structure, economic and political relations, and psychological, sociological, and historical conditions. A directly qualitative/relational approach to data analysis arose in Moscow during the last decade of the Societ period, in the work of Professor Victor Konstantinovich Finn. This approach currently thrives in Finn's Intellectual Systems Laboratory in VINITI (The All-Russian Institute for Scientific and Technical Information). By using a "semeiotic" approach--that is, one based on mathematical logic and the formal analysis of concepts--Finn's so-called "JSM Method" is a novel tool for data fusion that has proved itself effective in a number of applications. These range from chemistry and pharmacology to sociology and labor relations. The present paper presents basic ideas of Finn's JSM Method of Automatic Generation of Hypotheses.

3 - HyM: a Methodology for the Development of Integrated Hybrid Intelligent Information Systems

Simon Kendal, X. Chen, A. Masters, University of Sunderland, UK

Abstract: HyM is a hybrid methodology for the development of large-scale and complex integrated hybrid intelligent information systems, which combines traditional information system development methods with knowledge-based system development methods. The methodology is an integration of four existing methods using two integration process approaches: intra-process and inter-process. In the requirements analysis phase, a structured method is applied to function analysis, an information modeling method is applied to data analysis, and a knowledge acquisition method is applied to knowledge analysis. An intra-process approach is then used to integrate these techniques together using consistency rules. In the design phase, the inter-process approach is used to transform the requirements analysis to object-oriented design by a transformation method. Finally, the object-oriented method is applied to the design and implementation of hybrid information systems. This methodology takes advantage of the four individual methods to overcome the limitations of each. It is applicable to the development of traditional information systems, knowledge-based systems, and large and complex co-operative and intelligent information systems.

Session WeC5: Guidance and Navigation - 2

Chair: Jose R. Casar, Universidad Politécnica de Madrid, Spain
Co-Chair: Itzhack Y. Bar-Itzhack, Technion Institute of Technology, Haifa, Israel

1- Data Fusion Versus Passive Filtering for Angular Velocity Estimation

Itzhack Y. Bar-Itzhack, Technion-Israel Institute of Technology, Haifa, Israel

Abstract: This paper presents two approaches to estimating the angular velocity of a spacecraft. One approach, which is basically a data fusion approach, uses passive filtering, while the other uses active filtering. According to the first approach differentiated spacecraft attitude data is processed to yield noisy angular velocity information which is then passed through a low pass filter. According to the second approach, an active filter that yields the angular velocity, processes either the differentiated attitude data, or just the attitude data. The active filter blends the information rendered by the spacecraft dynamics with information on either the differentiated attitude or on the attitude itself. Examples are presented which use real spacecraft data.

2 - Multisensor Data Integration in the NASA/Stanford Gravity Probe B Relativity Mission

M.I.Heifetz, G.M.Keiser, A.S.Krechetov, A.S.Silbergleit, Stanford University, CA, USA

Abstract: Gravity Probe B (GP-B) is a gravitational experiment designed to measure two predicted by General Theory of Relativity precessions of a free-falling gyroscope placed in a polar orbit about the Earth. The frame-dragging effect (drift perpendicular to the orbital plane) has never been directly measured before, while the geodetic effect (drift in the orbital plane) will be measured with an unprecedented accuracy. GP-B Data Analysis is an example of the multi-sensor information fusion: it requires the integrated processing of the data from nine physical sources of information: four science gyroscopes, the science telescope, the attitude control system of the spacecraft, on-board GPS receivers, NASA/JPL Earth ephemerides, and the astronomical data on the proper motion of the reference star. This paper presents the core filtering approach to the state-estimation of the GP-B system. Two-step nonlinear filter (that may be applied for the wide class of nonlinear measurements) is discussed and the specifics of its implementation for the GP-B data analysis is presented.

3 - Fault Detection and Isolation using Interval Analysis: Application to Vehicle Monitoring

Pascal Bouron, Dominique Meizel, Université de Technologie de Compiègne, France

Abstract: This paper gives an example of using set membership techniques for detecting component fault and model failures and isolating the cause of the fault. Set membership estimation techniques can inherently detect model failure when the estimated set becomes empty. This property is here applied for fusing parity equations generated by an analytic redundancy study. For each parity equation, one defines a symbolic indicator that individually characterizes a certain or possible failure. Defining a {cause/effect} array makes it possible to isolate the certain or possible causes of the defect. The method is developped within a pedagogical example of the kinematic model of a vehicle.

4 - ADS Bias Cancellation Based on Data Fusion with Radar Measurements

Juan A. Besada, Jesus Garcia, Gonzalo de Miguel, Jose R. Casar, Universidad Politécnica de Madrid, Spain
Gonzalo Gavin, INDRA-DTD, Madrid, Spain

Abstract: This paper describes a complete tracking function for air traffic control based on the fusion of radar and ADS-B messages. For this system, the most important terms to be corrected are low frequency errors from both radar sensors and ADS measurements, because the other components can be easily lowered through filtering. In the paper we propose innovative methods both for radar registration, and for low frequency ADS-B errors removal. Those two processes form the core of the tracking function. Additionally, very accurate measurement conversions are included, to avoid corrupting the estimations. The results clearly show the ability of the system to improve ADS-B based tracking and to obtain accurate estimations of radar biases. This second aspect is very interesting for improving tracking non ADS-B equipped aircraft.

Session WeD1: Target Tracking - 6 – Track Fusion - 3

Chair: Thiaglingam Kirubarajan, University of Connecticut, Storrs, CT, USA
Co-Chair: Peter K. Willett, University of Connecticut, Storrs, CT, USA

1- Multisensor Multitarget Tracking with Central-to-Local Feedback

Carl G. Looney, Yaakov L. Varol, Sheng Tang, Computer Science Department, University of Nevada, Reno, USA

Abstract: This paper investigates the effects on central tracks of the feedback of central track data to local sensing and tracking stations. The sensors are local radar stations that also initiate and update local tracks from noisy measurements of range, azimuth and elevation using abg-filters. They send their current track data to the central tracking station, which fuses its own track predictions with the local data to update its central tracks via abg-filters. We study the two cases of with and without missing data. Our simulations use low, medium and high levels of noise power. They show that the feedback reduces the error between the central tracks and the actual trajectories by an average of about 50%.

2 - Effects of Cross-Covariance and Resolution on Track Association

B. La Scala, The Preston Group, Richmond, Australia
A. Farina, ALENIA Marconi Systems, Italy

Abstract: This paper considers the effects on track association of two features that are often neglected in analysis. These two features are the cross-covariance between the two tracks of the same target; and the effects of a loss of resolution in a sensor. The sensors modelled in the paper are non-homogeneous in both state and measurement space and a method for calculating the cross-covariance for such dissimilar sensor is derived. Also, a simple resolution model is proposed. The paper uses Monte Carlo simu-lations to illustrate the effects of these two features on a number of track association methods.

3 – MILORD, an Application of Multifeature Fusion for Radar NCTR

V. Nimier, A. Bastière, ONERA, Châtillon, France
N. Colin, M. Moruzzis, Thomson-CSF/ Airsys, Bagneux, France

Abstract: Among the various topics addressed by data fusion, the application to Target Recognition by Radar is one of major interest because it is expected that good performance will come out from a process in which several complementary information will be merged. In the framework of the PEA ("Programme d'Etudes Amont") MILORD ("Moyen d'Identification Lointaine d' Objectifs Radar Désignés"), studies are currently conducted for defining the best techniques to be used for this function. This paper summarises the current results which were obtained so far in this domain. After having presented, in Chapter 2, the objectives and interests of fusion within the domain of Target Recognition by Radar, major fusion techniques are discussed in Chapter 3. Application of fusion to MILORD and current results are then summarised in Chapter 4, before conclusions which are drawn in Chapter 5.

4 - Multi-Mode Detection with Markov Target Motion

Danna Sinno, D. Cochran, D.R. Morrell, Arizona State University, Phoenix, AZ, USA

Abstract: This paper addresses the problem of configuration of a detection system offering multiple modes of operation that differ in their detection performance and geographical coverage. A technique for optimal mode selection based upon minimizing Bayesian risk is formulated and demonstrated for the case of a two-mode system with a moving target. The dynamics of the target are described by a Markov model.

Session WeD2: Tracking Course - 3

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

Short Course: Multitarget Tracking and Multisensor Fusion

Session WeD3: Remote Sensing

Chair: Basel Solaiman, ENST-Bretagne, Brest, France
Co-Chair: Belur V. Dasarathy, Dynetics, USA

1 - Multimodality Image Registration and Fusion Using Neural Network

Mostafa G. Mostafa, Aly A. Farag, Edward Essock, University of Louisville, KY, USA

Abstract: Three-dimensional (3-D) digital models of terrain region are essential for various remote sensing applications. Multimodality image registration and fusion are essential steps in building 3-D models from remote sensing data. In this paper, we present a neural network technique for the registration and fusion of multimodality remote sensing data for the reconstruction of 3-D models of terrain regions. A FeedForward neural network is used to fuse the intensity data sets with the spatial data set after learning its geometry. Results on real data are presented. Human performance will be assessed on several perceptual tests in order to evaluate of the fusion results.

2 - Radar Image Fusion by Multiscale Kalman Filtering

Giovanni Simone, Francesco C. Morabito, University of Reggio Calabria, Italy
Alfonso Farina, Alenia Marconi Systems, Roma, Italy

Abstract: In this paper, we propose the application of the Multiscale Kalman Filter (MKF) to the fusion of images of the same scene, acquired by different radars operating with different resolutions. The images have been spatially registered and they have been processed by a MKF processor, to compute an output image with improved information carried by each input image. The full fledged Multiscale Kalman Filter is described, and a basic model is tested to fuse AIRSAR and SIR-C images.

3 – Wide Area Fire Surveillance by Infrared Digital Signal Processing

L. Vergara, P. Bernabeu, J.Igual, Universidad Politécnica de Valencia, Spain

Abstract: A general scheme for automatic signal detection in scanning surveillance systems is proposed. We consider the problem of deciding if we have an alarm, based on the data measured in the cell under analysis in different consecutive scanning time instants. The proposed scheme includes a linear predictor and a subspace model for the signal to be detected, in an effort to maximise the detector performance. Although it may have general applicability in scanning surveillance problems, we focus the work on the problem of wide-area uncontrolled fire detection by means of infrared radar.

4 - Segmentation of Airborne Hyperspectral Images by Integrating Multi-Level Data Fusion

Marc Lennon, M.C. Mouchot, Grégoire Mercier, Basel Solaiman, ENST-Bretagne, Brest, France
L. Hubert-Moy, Université de Rennes, France

Abstract: This paper deals with the extraction of bocage network from hyperspectral images acquired with the Compact Airborne Spectrographic Imager (CASI). The strategy of segmentation integrates several levels of data fusion allowing to take a decision concerning the membership of each pixel to the bocage network from the large set of original data. The first level leads to quantify the membership of each pixel to specific features of the bocage. It includes data fusion based on physical properties, geometric context-dependant fuzzy fusion with an original consistency measure and geometric fusion of decisions. The second level is a fuzzy fusion of methods allowing to quantify the membership of each pixel to the bocage network. Finally, the third level consists in post-processing the data with a context-dependent fusion of decisions to obtain the final map of the bocage.

Session WeD4: Symbolic and Numeric Information: Hybrid Approaches - 2 – Invited session

Chair: Galina Rogova, CUBRC, Buffalo, NY, USA

1 - Dual aspects of a Multi-Resolution Grid-Based Terrain Data Model with Supplementary Irregular Data Points

Fredrik Lantz, Erland Jungert, Swedish Defense Research Establishment, Linköping, Sweden

Abstract: Digital terrain data models in high resolution are required in applications for visualization but also, e.g. for identification of various types of terrain features. These two aspects are in a way contradictory since the former application require a large number of data points to represent the high resolution, while the latter cannot deal with such a large number of data points without high demands for heavy computational powers. A solution to this problem is a structure that includes quantitative characteristics for visualization and a qualitative representation for feature analysis. A digital terrain data model characterized with these dual aspects has been designed and will be presented in this work.

2 - An Agent based Combat Information Processing System

Phillip Emmerman, Uma Movva, Larry Tokarcik, US Army Research Laboratory, Adelphi, MD, USA
Carolin Gasarch, Timothy J. Rogers, V.S. Subrahmanian, University of Maryland, College Park, MD, USA

Abstract: Tactical battlefield applications require that agents monitoring battlefield events be able to dynamically react to events and autonomously take actions that are in the best interest of the agent (and the command and control system to which the agent belongs). In this paper, we describe how IMPACT - Interactive Maryland Platform for Agents Collaborating Together - has been used to agentize selected parts of a large scale battlefield visualization demonstration system entitled the Combat Information Processor (CIP). The CIP has been designed by the US Army Research Laboratory to demonstrate and experiment with large scale, tactical battlefield visualization concepts. We describe how this agentization improves the functionality of CIP significantly.

3 - Automatic Air Target To Air Line Association

Martin Oxenham, Defense Science and Technology Organisation, Salisbury, Australia

Abstract: Australia's Strategic Policy states that surveillance of the sea-air-gap to the north of Australia is a key element of Australia's defence strategy. This surveillance is primarily afforded by a suite of ground-based microwave and over-the-horizon (OTH) radars. The fusion of the data and tracks from these sensors is an area of active research in support of air picture compilation. To ensure the best results from the fusion process, it is critical that the estimates of the target state from each constituent sensor be as accurate as possible. In this paper, the theoretical aspects of the problem of automatically associating air targets with airlanes are investigated with the aim of providing a means of removing possible biases in OTHR target state estimates caused by ionospheric effects. Furthermore, some of the benefits of automatic air target to airlane association for situation assessment are briefly examined.

Session WeD5: Guidance and Navigation - 3

Chair: Patrick Maupin, LIVIA, Montréal, Québec, Canada
Co-Chair: Stéphane Paradis, DRE Valcartier, Val-Bélair, Québec, Canada

1 - Time Recovery through Fusion of Inaccurate Network Timing Assistance with GPS Measurements

Jari Syrjärinne, Nokia Mobile Phones, Tampere, Finland

Abstract: In this paper fusion of inadequate timing information from GPS space vehicles and from a cellular network is studied. Neither of the sources is accurate enough to be used for exact timing alone, but using the time recovery method proposed by the author, exact timing can be eventually derived from the fusion of these sources. Naturally, the positioning accuracy is also positively affected when accurate time is recovered. The proposed method is based on minimization of a quality of fit value. The quality of fit value used in this paper is sum of squared residuals obtained from a Least Mean Squares based positioning procedure, widely used in point-solution type GPS applications. The proposed method is tested and evaluated using both simulated and real data.

2 - Fusion of Heterogeneous Sensors for the Guidance of an Autonomous Vehicle

Jan C. Becker, Technical University, Braunschweig, Germany

Abstract: This paper describes the sensor fusion system of an autonomous vehicle for automated vehicle testing. The vehicle sensor system for object-detection consists of a stereo vision sensor, four laserscanner (lidar) and a radar sensor. The sensor system is designed to totally cover the vehicle environment with a high redundancy in front of the vehicle. The sensor fusion system of the vehicle consists of a data alignment, a data association and a state estimation module. An adaptive information filter is used for the fusion of the associated targets from different sensors. The fused targets are input to the path planning and guidance system of the vehicle to generate a collision free motion of the vehicle.

3 - Sensor Data Fusion Using Kalman Filter

Jurek Z. Sasiadek, P. Hartana, Carleton University, Ottawa, Ontario, Canada

Abstract: Autonomous Robots and Vehicles need accurate positioning and localization for their guidance, navigation and control. Often, two or more different sensors are used to obtain reliable data useful for control system. This paper presents the data fusion system for mobile robot navigation. Odometry and sonar signals are fused using Extended Kalman Filter (EKF) and Adaptive Fuzzy Logic System (AFLS). The signals used during navigation cannot be always considered as white noise signals. On the other hand, colored signals will cause the EKF to diverge. The AFLS was used to adapt the gain and therefore prevent the Kalman filter divergence. The fused signal is more accurate than any of the original signals considered separately. The enhanced, more accurate signal is used to guide and navigate the robot.

4 - Multiple Correspondence Analysis for Highly Heterogeneous Data Fusion. An Example in Urban Quality of Life Assessment.

Patrick Maupin, LIVIA, Ecole de technologie supérieure, Montréal, Québec, Canada
Philippe Apparicio, Université du Maine, Le Mans, France and INRS-Urbanisation, Montréal, Canada
R. Lepage, LIVIA, Ecole de technologie supérieure, Montréal, Québec, Canada
Basel Solaiman, ENST-Bretagne, Brest, France

Abstract: The aim of this paper is to present the preliminary results obtained through a composite data processing system using Multiple Correspondence Analysis in order to perform data fusion and deduce rules of spatial organisation. After having described a case study on the cartography of Ambrosia artemisiifolia (common ragweed), currently investigated on the Urban Community of Montreal (Canada) territory, the authors will present an overview of the methods implemented for the digitalisation and interpolation of data and discuss on methodological problems raised in the curse of the study. Finally, avenues for future research will conclude this paper.

Monday - Tuesday - Wednesday - Thursday

Last Updated: June 20, 2000
Web site by: (content), (form)
copyright © ISIF 2000