Tracking Highly-Maneuvering Targets and Data Fusion from IMM Estimator Tracks

Maneuvering targets are of special interest in remote tracking because of the difficulty in modeling the changes in their dynamic behavior. Sharp maneuvers present challenges. For some targets, such as those moving in a 2D plane, the proposed White Noise Turn Rate (WNTR) model should be included as one or more modes in an Interacting Multiple Model (IMM) estimator. This model is similar to the conventional coordinated-turn (CT) model, also known as the Wiener Process Turn Rate model, but with non-zero mean turn rate process noise, instead of zero-mean turn rate process noise. Such an IMM design, possibly coupled with a CT mode as well to accommodate longer turns with low turn rate, results in better estimation accuracy and improved covariance consistency. More sophisticated targets, such as those guided by feedback control systems in 3D space headed for a known or approximately-unknown destination, should be modeled using a guidance and destination model. The Proportional Navigation based estimation model discussed below describes the motion of such targets with the ability to model maneuvers as an approximately-Gaussian process noise in the target\u27s destination, where the destinations can switch to ``fake\u27\u27 ones if the target exhibits evasive maneuvers on its way to its true destination. It is based on the concept of the Proportional Navigation guidance feedback control strategy implemented in real-world targets, where such a target reaches its destination by continuously attempting to reduce the angle rate of the line of sight (LOS) vector to zero. It is shown that the structure of the model, especially when used in an IMM estimator, allows for an intuitive and robust description of target maneuvers in 3D space with significant improvement in tracking accuracy and with destination estimation capability. To strengthen the capability of tracking systems that operate using multiple independent trackers estimating the same target\u27s state, a data fusion method dedicated to information coming from IMM tracks is developed. The Fusion Center (FC) receives local IMM ``inside information\u27\u27: mode-conditioned estimates and mode probabilities representing the Gaussian mixture density information at the local trackers. The method is capable of on-demand fusion, where the fusion is performed without memory of past fused estimates. It is derived using Bayesian principles and proposes a first-order Taylor series approximation of the evolution of data computed by independent IMM estimators. Additionally, the solution includes the novel log-ratio (Log-R) transformation of mode probabilities that allows for a Gaussian approximation and Bayesian fusion. Simulations show that this fusion estimator has accuracy that is slightly better than fusion using outside information (local moment-matched IMM outputs) and that the consistency of the fused covariance matches the ideal consistency of the Centralized Measurement Fusion (CMF) covariance because of the computation and inclusion of crosscovariances in the fusion