Maneuvering target tracking problem is worthy of being interest because the estimation of target states requires an adaptive procedures of unknown target dynamics. This thesis is mainly focused on the development of theoretic methods on this type of problems. There have been many researches on this topic and they can be categorized as model-based adaptive filtering and input estimation. Among them, input estimation is an approach which determines if target maneuvers exist, and directly estimates the magnitude of the unknown maneuvers. One of the merits of input estimation is that target maneuvers are directly estimated from the available measurements regardless of the input level, without re-initializing any of filter parameters. However, the conventional input estimation methods assume that the input level is constant within the detection window. This has been the main drawback of the input estimation approaches since a realistic target maneuver may change in various fashions within the detection window. Consequently, estimation of the unknown target maneuvers based on such a strong assumption on the maneuver shapes gives only limited performance.
In this thesis, a new input estimation method is proposed to overcome the constant-level assumption on input signal. The proposed algorithm approximates the input signal as a linear combination of some elementary base functions of time. This formulation greatly simplifies the estimation procedure since only the coefficient vector of the base functions needs to be estimated. Due to this feature, a large detection window can be employed without significantly increasing the computational burden. By computer simulations, the performance of the proposed algorithm with a moderate selection of input shape is comparable to Bogler``s method, while the computation time is greatly reduced.