Multiple Model Estimation via Variable Structure With Spatiotemporal Primal-Dual Projection

This letter presents a robust multiple-model (MM) estimator based on variable structure multiple Gaussian importance filtering (VSMGIF) with spatiotemporal primal-dual projection. In VSMGIF, orthogonal-velocity-constrained maneuver models are introduced, and spatiotemporal causal constraints are incorporated within clearly defined projection zones. This framework enables approximate Rao-Blackwellization, while combining Gaussian mixture models with importance sampling to handle nonlinear, non-Gaussian hybrid estimation. Model probabilities and sample weights are adaptively corrected using all available information. Experimental results demonstrate that VSMGIF effectively mitigates coordinate coupling and Gaussian truncation errors in bearing-only tracking.