Performance Prediction of Multisensor Tracking Systems for Single Maneuvering Targets

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 [2], [1]. However, a simplified method for predicting the performance of a multisensor tracking system against maneuvering targets is needed to confirm the results of computer simulations, real-time command and control decisions such as multisensor resource allocation, and engineering of complex multisensor systems [3]. When reviewing current approaches to performance prediction, four traits are helpful in distinguishing the relative advantages between the approaches listed in the literature survey and the work presented in this paper. The first two traits deal with the ability to predict performance of multisensor or multitarget tracking algorithms. Most of the current approaches to performance prediction typically deal with a single sensor and single target. In addition, while some algorithms deal with either the multisensor or multitarget case, no work has been identified that is capable of handling both the performance prediction of multisensor and multitarget tracking algorithms. The third trait is ability to predict performance of a defined scenario. While some performance prediction techniques express an expected performance based on sensor parameters, none extend well to maneuvering target scenarios since deterministic changes in motion are not considered. Finally, the fourth trait to be considered is algorithm complexity. Engineering of complex systems often involves extensive parametric variability for which, traits three and four are critical. Algorithm scenario dependence and simplicity yield a high level of confidence when comparing against more complicated simulated algorithms. In this work, we present a simple to compute algorithm that accounts for scenario laydowns and can be extended to multiple sensors while assuming measurement sharing on a single maneuvering target. Cramer-Rao Bound (CRB) techniques are one of the basic tools for estimating performance. However, these techniques were originally designed for estimating deterministic parameters. More appropriate for the performance prediction of tracking systems, the Posterior Cramer-Rao Bound (PCRB) extends the CRB to provide a “measure” of system performance when both measurements and state are assumed to be stochastic processes [5], [4]. The work of [6] extends the PCRB to handle the prediction of multitarget systems under a set of assumptions. In addition, the work of [7] extends the PCRB to handle the prediction of an estimator for a single maneuvering target. However, none of the papers dedicated to an extension of the PCRB include comparisons to Monte Carlo simulations from a realistic tracker. Thus, the usefulness of each algorithm as it pertains to the performance of a multitarget tracker in a given scenario is not known.