文档介绍：I. INTRODUCTION
Tracking a maneuvering target using multiple
models has been shown to be highly effective in many
Multiple-Model Probability applications. In single-model filters, if the model
used by the filter does not match the actual system
Hypothesis Density Filter for dynamics, the filter will tend to diverge since the
actual errors fall outside the range predicted by the
Tracking Maneuvering Targets filter using its error covariance [7]. Maneuvering
targets might switch between different modes of
operation, and tracking using a single-model filter
might fail since the filter may be accurate for only
one mode of operation. In multiple-model approaches,
K. PUNITHAKUMAR, Student Member, IEEE
several filters, each matched to a different target
T. KIRUBARAJAN,
Member, IEEE motion mode, operate in parallel, and then the
A. SINHA overall state estimate is given by a weighted sum
McMaster University of the estimates from each filter. In many target
tracking problems with linear, Gaussian systems, the
interacting multiple model (IMM) estimator [7—9],
in which a bank of different hypothetical target
Tracking multiple targets with uncertain target dynamics
motion models is used, has been proven to have
is a difficult problem, especially with nonlinear state and/or
measurement equations. With multiple targets, representing better performance than the (single-model) Kalman
the full posterior distribution over target states is not practical. filter. For example, in the target tracking benchmark
The problem es even plicated when the number problem [6], in which performances of different
of targets varies, in which case the dimensionality of the state algorithms for tracking highly maneuvering targets
space itself es a discrete random variable. The probability pared, the IMM estimator outperformed the
hypothesis density (PHD) filter, which propagates only the Kalman and ®—¯ filters and yielded one of the best
first-order statistic