Adaptive Tracking Filter for Maneuvering Targets

Abstract

A general method of continually restructuring an optimum BayesKalman tracking filter is proposed by conceptualizing a growing tree of filters to maintain optimality on a target exhibiting maneuver variables. This tree concept is then constrained from growth by quantizing the continuously sensed maneuver variables and restricting these to a small value from which an average maneuver is calculated. Kalman filters are calculated and carried in parallel for each quantized variable. This constrained tree of several parallel Kalman filters demands only modest om;puter time, yet provides very good performance. This concept is implemented for a Doppler tracking system and the performance is compared to an extended Kalman filter. Simulation results are presented which show dramatic tracking improvement when using the adaptive tracking filter.The Kalman filter has been accepted as one of the best methods of providing the motion analysis of the steadystate parameters of a moving target. Although developed specifically for linear models, the Kalman filter has found wide application for nonlinear observation equations and nonlinear motion through linearization techniques (i.e., extended Kalman filter). Maneuvering target tracking has also received considerable attention. The likelihood of a maneuver can be accommodated in the design by increasing the modeled plant noise, but this, of course, degrades the accuracy during steady conditions. It also makes the estimator very prone to divergence, particularly if nonlinear equations have been approximated by an extended Kalman filter expansion. A number of adaptive approaches have been investigated to resolve this design dilemma by introducing rnodifications to the Kalman filter to get it through the maneuver without diverging or otherwise severely distorting the estimate. Some of these devices are 1) detecting the maneuver and modifying the Kalman filter by reinitializing the covariance matrix, 2) detecting the maneuver and changing the state variables, 3) using a fading memory concept, and 4) using statistical forcing functions to represent target maneuvers. These methods have the limitations of requiring a specific detection mechanism of being expensive to implement and of requiring knowledge of the target not always available.The performance of these techniques also depends upon the target model for the maneuver. For the process proposed here, the target was assumed to experience a very rapid maneuver that occurs only rarely during the course of travel. For the reasons cited above and because of the peculiarity of such a model, the recent literature has explored adaptive parallel Kalman filter structures [1] - [5] . The Bayes-Kalman tree structure proposed here is a variation on those developments.The technique developed in its pure form generates a continuously expanding set of Kalman filters responding to a continuously expanding setof options available to the maneuvering target. The proposed implementation limits the expansion to a discrete set of values for the maneuver var...

Contact Localization and Motion Analysis in the Ocean Environment: A Perspective

“…Residual error characterization remains an active area of research; one in which detection of the bias has been stressed. Much of the attention has centered on adaptation to biasing caused by contact maneuvers [17,[42][43][44][45]. However, increasing attention is being paid to biasing due to sensor positioning [46] and environmental effects [14], and also on the effects of certain types of random errors [47,48].…”

Target Tracking Based on Bearing Only Measurements

“…In its pure form, this approach requires an infinitely growing bank of paralluý 'ilters, N initially, N2 for the second measurements, etc. Different approaches to reduce this described growth are proposed in the literature, in order to get practical, realizable filters [14], [10], [21], [22], [28], [29].…”

A Survey of Manoeuvring Target Tracking Methods

“…All of these approaches increase the effective filter bandwidth so as to respond more rapidly to the measurement data during a manoeuvre; but they are risky in the presence of clutter since they increase the size of the validation gate. Some of the more recent approaches to manoeuvring target tracking include: adaptive Kalman filters [61,39], filters using correlated and semi-Markov process noise [120,119,60] usually implemented using multiple model (partitioning) filters and filter banks [131,58,95,114,125,141]; filters based on Poisson and renewal process models of acceleration [86,128]; input estimation [35,36] and input and onset-time estimation [33,38,116]; variable dimension filters [13]; track splitting filters with a finite memory constraint [147]; the generalised pseudo-Bayesian (GPB) algorithm [68,1]; and the interacting multiple model (IMM) algorithm [31,28,96]. A second-order extension of the IMM algorithm was developed in [26].…”