Multitarget Tracking Algorithm Parallelization for Distributed-Memory Computing Systems.
ABSTRACT We present a robust scalable parallelization of a multitarget tracking algorithm developed for air traffic surveillance. We couple the state estimation and data association problems by embedding an interacting multiple model (IMM) state estimator into an optimization-based assignment framework. A SPMD distributed-memory parallelization is described wherein the interface to the optimization problem, namely computing the rather numerous gating and IMM state estimates, covariance calculations, and likelihood function evaluations (used as cost coefficients in the assignment problem), is parallelized. We describe several heuristic algorithms developed for the inherent task allocation problem wherein the problem is one of assigning track tasks, having uncertain processing costs and negligible communication costs, across a set of homogeneous processors to minimize workload imbalances. Using a measurement database based on two FAA air traffic central radars, courtesy of Rome Laboratory, we show that near linear speedups are obtainable on a 32-node Intel Paragon supercomputer using simple task allocation algorithms.
- [Show abstract] [Hide abstract]
ABSTRACT: The focus of this paper is to present the results of our investigation and evaluation of various shared-memory parallelizations of the data association problem in multitarget tracking. The multitarget tracking algorithm developed was for a sparse air traffic surveillance problem, and is based on an Interacting Multiple Model (IMM) state estimator embedded into the (2D) assignment framework. The IMM estimator imposes a computational burden in terms of both space and time complexity, since more than one filter model is used to calculate state estimates, covariances, and likelihood functions. In fact, contrary to conventional wisdom, for sparse multitarget tracking problems, we show that the assignment (or data association) problem is not the major computational bottleneck. Instead, the interface to the assignment problem, namely, computing the rather numerous gating tests and IMM state estimates, covariance calculations, and likelihood function evaluations (used as cost coefficients in the assignment problem), is the major source of the workload. Using a measurement database based on two FAA air traffic control radars, we show that a “coarse-grained” (dynamic) parallelization across the numerous tracks found in a multitarget tracking problem is robust, scalable, and demonstrates superior computational performance to previously proposed “fine-grained” (static) parallelizations within the IMMIEEE Transactions on Parallel and Distributed Systems 11/1997; · 1.80 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: To date, there has been a lack of efficient and practical distributed- and shared-memoryparallelizations of the data association problem for multitarget tracking. Filling this gap is oneof the primary focuses of the present work. We begin by describing our data association algorithmin terms of an Interacting Multiple Model (IMM) state estimator embedded into anoptimization framework, namely, a two-dimensional (2D) assignment problem (i.e., weightedbipartite matching). Contrary to conventional wisdom, we show that the data association (oroptimization) problem is not the major computational bottleneck; instead, the interface to theoptimization problem, namely, computing the rather numerous gating tests and IMM stateestimates, covariance calculations, and likelihood function evaluations (used as cost coefficientsin the 2D assignment problem), is the primary source of the workload. Hence, for both ageneral-purpose shared-memory MIMD (Multiple Instruction Multiple Data) multiprocessorsystem and a distributed-memory Intel Paragon high-performance computer, we developedparallelizations of the data association problem that focus on the interface problem. For theformer, a coarse-grained dynamic parallelization was developed that realizes excellent performance(i.e., superlinear speedups) independent of numerous factors influencing problemsize (e.g., many models in the IMM, denseycluttered environments, contentious target-measurementdata, etc.). For the latter, an SPMD (Single Program Multiple Data) parallelization wasdeveloped that realizes near-linear speedups using relatively simple dynamic task allocationalgorithms. Using a real measurement database based on two FAA air traffic control radars, weshow that the parallelizations developed in this work offer great promise in practice.Annals of Operations Research 01/1999; 90:293-322. · 1.03 Impact Factor