Gaussian Mixture Model-Based Variational Bayesian Approach for Extended Target Tracking

Abstract

This article presents a new Gaussian mixture model-based variational Bayesian approach (VBSDD-ETT) for solving the problem of skew-dense distribution (SDD) of measurement points in the extended target tracking (ETT). Random matrix-based ETT often presumes that the measurement points are evenly dispersed across the entirety of the ellipse object. However, its performance will significantly decrease when measurement points are SDD. The proposed VBSDD-ETT approach not only solves this issue, but also obtains recursive estimation within a Bayesian framework. Specifically, a new Gaussian mixture model that uses translations, scaling, and rotations of subellipses is proposed to address the problem of SDD. Second, a VBSDD-ETT approach based on the Gaussian mixture model is presented to derive the posterior distribution in an analytical form. Also, we propose a VBSDD-ETT-based information-theoretic interacting multiple model (ITIMM-VBSDD) algorithm to tackle the problem of model uncertainty caused by the maneuvering of the target. The ITIMM-VBSDD algorithm can obtain more accurate estimation results of the kinematic state and extension. Finally, the performances of VBSDD-ETT and ITIMM-VBSDD are evaluated in the simulation and real scenarios. The results demonstrate the effectiveness of the proposed approach compared with existing random matrix methods.