Solutions to the problem of filter initialization and divergence prevention are developed. These solutions are arrived at by introducing a lag of l so that l measurements beyond the time for which the state is being estimated are available. This means that it is possible to produce a local estimate of the state. This estimate can be compared with the filtered estimate of the state to check for divergence. If divergence is found, the filter can be reinitialized with this estimate. These solutions are then applied to the problem of tracking a maneuvering target. Approximate theoretical and actual simulated RMS position and velocity errors are given for the Bar-Shalom/Birmiwal trajectory. Starting from one system, we extracted l+1 corresponding filters. Since l+1 measurement sets were available to each of the filters, we were able to check for divergence and reinitialize when divergence was detected. To test the method, we used it on the maneuvering system problem with the Bar-Shalom/Birmiwal trajectory. The results show that this approach performs as well as, if not better, than the other available methods. We conclude that our method appears promising.
Kalman filter applications in highly maneuverable, intelligent target tracking
Kalman-Filter-Anwendungen bei sehr beweglicher, intelligenter Zielverfolgung
1983
8 Seiten, 6 Bilder, 8 Quellen
Conference paper
English