Locating, Classifying and Countering Agile Land Vehicles by David D. Sworder & John E. Boyd

Author:David D. Sworder & John E. Boyd
Language: eng
Format: epub
Publisher: Springer International Publishing, Cham

The (2T--GWE algorithm meets all of the system specifications. A friendly target can be identified quickly from either direct measurements or the elusive strategy of its motion.

5.5 Conclusion

This chapter presents several illustrative examples of use of the GWEwhere the architect’s intent is to track a target, classify it, or both. All of the examples use the same measurement set: either or as the case may be. The most favorable algorithm for tracking in this application is the EKFC. Target classification is not an issue: the cooperative EKF knows the proper target class, and likewise with target motion mode (it also knows the current motion regime). The tracking error is roughly that of the minimum variance estimator though the nonlinear measurements and the pseudo-noise adjustments argue against taking that too literally.

The least favorable algorithm in this application is the nominal EKF—either with or without pseudo-noise augmentation. The basic EKF-tracker knows neither target motion regime nor the existence of alternative target classes with distinguishing motion characteristics. Instead, the EKF uses a single, centered kinematic model within a prespecified target class. In the cases studied here, the specific engagement model is that of conventional CV motion with pseudo-noise introduced to account for the modal uncertainty. The EKF is slow to respond to regime transitions because it has no reason to expect them. The response delays in the nominal EKF can be reduced by increasing the pseudo-noise, but the notional 1σ-error ellipses become correspondingly, and in most cases incorrectly, larger. The response of the nominal EKF gives a credible lower bound on the performance of a tracker in this application.

We would expect the performance of the GWE to lie between normative EKF and the EKFC. The GWE does not ignore the regime changes as does the EKF, nor does it know the regime as does the EKFC. Rather, it acknowledges the possibility of regime events and tries to classify them as they occur. The quality of this classification depends upon the data set at the tracker and on the analytical model that the GWE uses to cue its response to patterns in the observations.

The basic one-target, Markov tracker, (M--GWE with a string depth of three employs 27 separate local models. We would expect its performance to be superior to the nominal EKF, but less good than the cooperative EKF. The latter can be viewed as a multiple model tracker that always uses the correct local model, reported by the cooperative target, giving it a continuous regime string with length k at time t = kT—far longer than the truncated regime histories of length three (6 s) available to the GWE.

There are two useful metrics that contrast the performance of the tracking algorithms. The first is the median radial tracking error. In this application, the median error is superior to the mean error since the latter is overly sensitive to the small number of big errors subsequent to a regime change. The maximum tracking error is also important. In most cases, the tracking window has fixed size. Any error that exceeds the bounds of the window will result in the target being lost.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.