Gradient Based Model Probability Interacting Multiple Model Algorithm

 

Zvonko Radosavljević

 

Parallel bank of non-adaptive estimators is used for adapting to target dynamics change. The discrete set of models describes all possible maneuver modes of the target. The state estimation is based on mixing/switching between filters in the set. Based on the gradient of the best model, a new adaptive model probability algorithm for the Interacting Multiple Model (Gradient based Model Probability Interacting Multiple Model - GMPIMM) is proposed in this article. The model probability coefficients are corrected in each IMM (Interacting Multiple Model) recursive loop, giving Minimum Mean-Square Error (MMSE) estimation. Experimental results show that the proposed GMPIMM gradient algorithm enhances global efficiency of the IMM algorithm in highly maneuvering target tracking applications.

Key words: target tracking, air target, interacting mode, interacting multiple model, gradient method.