Generalized smoothing for multiple model/multiple hypothesis filtering: Experimental results

The estimation of the state of a dynamical system from corrupted sensor data is difficult when data association conflicts, possibly unresolved measurements, and a complex system dynamics must be taken into account. With some necessity this problem calls for multiple hypothesis/multiple model estimators. In this context we consider experimental results from a multiple-target air surveillance application. Particular emphasis is placed on retrodiction, a generalization of standard fixed-interval smoothing to multiple model/multiple hypothesis filtering. If a certain time delay is tolerable, retrodiction provides unique and accurate tracks from a complex filtering output. Even a small delay may significantly ease the surveillance mission.