A novel robust MM filter for target tracking with glints

摘要

Interacting Multiple Model (IMM) filter faces significant outlier-caused peak-errors. In this paper, the Bayesian probability update in IMM is found equivalent to Dempster's Rule of Combination which cannot handle evidence conflicts caused by outliers. Furthermore, a novel robust MM (RMM) filter is proposed through introducing expert rules about mode evolvement and presenting the Likelihood Temporal Ratio (LTR) and building the Induced Combination Rule (ICR). Simulations about target tracking show the effectiveness of the proposed method.