JAIF -- Articles to Appear in a Future Issue

"Performance Prediction of Multisensor Tracking Systems for Single Maneuvering Targets"
William D. Blair, Paul A. Miceli, Georgia Tech Research Institute, USA

Studying the performance of multisensor tracking systems against maneuvering targets involves Monte Carlo simulations with the tracking algorithms implemented in a sophisticated computer simulation of the multisensor system. However, a simplified method for predicting the performance of a multisensor tracking system against maneuvering targets is needed for confirmation of the computer simulations, real-time command and control decisions such as multisensor resource allocation, and systems engineering of complex multisensor systems. The challenge of accurate performance prediction arises from the lack of covariance consistency of the Kalman filter when tracking maneuvering targets. In this paper, a method for performance prediction of a nearly constant velocity Kalman filter will be extended to tracking a maneuvering target with multiple dispersed sensors on an oblate earth. Given target position and acceleration as a function of time, the tracking performance of each sensor is expressed as a sensor-noise only covariance and maneuver lag or filter bias. In the fusion of the data from the multiple sensors, the SNO covariances fuse for a smaller covariance, while the maneuver lags fuse with a gain proportional to the inverse of the covariances for the sensor tracks. This method can also be used to predict the performance of a multisensor system that include one, two, and/or three dimensional sensors. The results of Monte Carlo simulations of multisensor tracking of a maneuvering target are used to illustrate the accuracy of methods for performance prediction.

"Track-to-Track Association with Augmented State"
R. Osborne III, Y. Bar-Shalom and P. Willett, University of Connecticut, USA

Association of tracks formed at different sensors is an ongoing area of interest in the field of information fusion and target tracking. In order to leverage additional information about a current target of interest which has been tracked at (an) additional sensor(s), track-to-track association (T2TA) must be performed. In addition to accurately identifying tracks with common origin, a desirable T2TA scheme will associate the tracks quickly, i.e., after only a few samples. A T2TA scheme is developed here which will take advantage of traditional kinematic state information as well as additional state information in the form of state augmentation. The main contribution is the use of two nonlinearly related state augmentations at the two sensors and accounting for their uncertainties. The results of T2TA are compared when using only kinematic state information, only state augmentation information, and the full augmented state. The full augmented state is shown to provide the most desirable association results, both in terms of accuracy and the number of samples needed to provide that accuracy.

"Possibilistic Medical Knowledge Representation Model"
H. Alsun, TELECOM Bretagne, France, L. Lecornu, TELECOM Bretagne and Latim, Inserm, France, C. Le Guillou, Latim, Inserm, France, B. Solaiman, TELECOM Bretagne, France

Medical Decision Support Systems involve two main issues: medical knowledge representation and reasoning mechanisms adapted to the considered representation model. This paper proposes an approach to construct a new medical knowledge representation model, based on the use of possibility theory. The major interest of using the possibility theory comes from its capacity to represent different types of information (quantitative, qualitative, binary, etc.), as well as different forms of information imperfections such as uncertainty, imprecision, ambiguity and incompleteness. Starting from the medical knowledge description, carried out by an expert giving the patient information representation, the proposed approach consists in building a possibilistic model including the Medical Knowledge Base (MKB). Moreover, the proposed approach integrates several possibilistic reasoning mechanisms on the considered information. The validation of the proposed approach is then conducted using Endoscopic Knowledge and Endoscopic Case Bases. The proposed representation, reasoning model and the obtained validation results show a real interest in order to achieve various goals of Medical Decision Support Systems such as classification, similarity estimation, etc.

"Profile-Free Launch Point Estimation for Ballistic Targets Using Passive Sensors"
R. Tharmarasa and T. Kirubarajan, McMaster University, Canada, N. Nandakumaran, Curtin University of Technology, Australia, Y. Bar-Shalom, University of Connecticut, USA, T. Thayaparan, Defence Research & Development Canada, Canada

This paper considers the estimation of the Launch Points (LP) of ballistic targets from two or more passive satellite-borne sensors by fusing their angle-only measurements. The targets are assumed to have a two stage boost phase with a free-flight phase between the two stages. Due to the passive nature of the sensors, there is no measurement during the free-flight motion. It is also assumed that measurements are available only after a few seconds from the launch time due to cloud cover. In the literature, profile-based methods have been proposed to estimate the target’s launch point and trajectory. Profile-based methods normally result in large errors when there is a mismatch between actual and assumed profiles, which is the case in most scenarios. In this paper, a profile-free method is proposed to estimate the target states at the End-of-Burnout (EOB) and LP. Estimates at the EOB are obtained by using forward-filtering with adaptive model selection based on boost phase changes. The LP estimates are obtained using smoothing followed by backward prediction. Uncertainties in the motion model and the launch time must be incorporated in the backward prediction. The LP estimate and the corresponding error covariance are obtained by incorporating the above uncertainties. Simulation results illustrating the performance of the proposed approach are presented.

Correction to “A Critical Look at the PMHT”
D. Crouse, University of Connecticut, USA, L. Zheng, Beijing Institute of Technology, China, P. Willett, University of Connecticut, USA

This is an erratum on a JAIF manuscript of the first and third authors, an erratum pointed out by the second author.

"Distributed Tracking with a PHD Filter Using Efficient Measurement Encoding"
A. Aravinthan, McMaster University, Canada, R. Tharmarasa, McMaster University, Canada, K. Punithakumar, GE Healthcare, Canada, T. Kirubarajan, McMaster University, Canada  and T. Lang, General Dynamics Canada, Canada

Probability Hypothesis Density (PHD) filter is a unified framework for multitarget tracking, which provides estimates for the number of targets as well as the individual target states. Sequential Monte Carlo (SMC) implementation of a PHD filter can be used for nonlinear non-Gaussian problems. However, the application of PHD-based state estimators for a distributed sensor network, where each tracking node runs its own PHD-based state estimator, is more challenging compared with single sensor tracking due to communication limitations. A distributed state estimator should use the available communication resources efficiently in order to avoid the degradation of filter performance. In this paper, a method that efficiently communicates encoded measurements between nodes while maintaining the filter accuracy is proposed. This coding is complicated in the presence of high clutter and instantaneous target births. This problem is mitigated using adaptive quantization and encoding techniques. The performance of the algorithm is quantified using a Posterior Cram´er-Rao Lower Bound (PCRLB) that incorporates quantization errors. Simulation studies are performed to demonstrate the effectiveness of the proposed algorithm.

"Posterior Cramer-Rao Bounds for Doppler Biased Distributed Tracking"
X. Song, University of Connecticut, USA, P. Willett, University of Connecticut, USA and S. Zhou, University of Connecticut, USA

This paper investigates distributed tracking with range-Doppler coupling, where a range measurement of a target of interest is linearly biased by its range-rate (or Doppler). The coupling parameter can be zero, positive, or negative. The posterior Cramer-Rao bound (PCRB) is derived for distributed radar systems: multistatic and multiple-input multiple-output (MIMO) settings. In the multistatic case, a positive (lambda) leads to the lowest PCRB, the same as is true for monostatic tracking. The paper also compares the tracking performance of multistatic and MIMO configurations, where the latter utilizes two waveforms with (+/- lambda) parameters, respectively. Regarding the power-unlimited case, a MIMO radar can always outperform a multistatic one from a tracking perspective. However, if the total power is limited, the situation is somewhat different: the transmitter co-located configuration is worse than a multistatic one, while in the widely-separated case the better choice depends on geometry.

"Predetection Fusion in Large Sensor Networks with Unknown Target Locations"
R. Georgescu, University of Connecticut, USA, P. Willett, University of Connecticut, USA,
S. Marano, University of Salerno, Italy, and V. Matta, University of Salerno, Italy

Fusion of multisensor data can improve target probability of detection but suffers from a potentially increased false alarm rate. The optimal sensor decision rule in the case of multiple sensor systems and known target location is of course a likelihood ratio test. This approach, however, is not applicable to many practical scenarios, such as sonar, in which the location of the target is not known and hence the alternative hypothesis becomes composite. Therefore, we
propose predetection fusion and highlight its application to a variety of multitarget multisensor trackers. Additionally, the algorithm is motivated by the need for an efficient way to process the volume of data from large sensor networks that consist of low quality sensors. We thus propose predetection fusion as a contact sifting procedure followed by an Expectation Maximization step that refines the location of the estimated detections. Results are provided on a synthetic dataset and on a challenging realistic multistatic sonar dataset. The performance of predetection fusion is compared against the performance of the optimal multihypothesis GLRT approach.