Measurement transformation algorithm for extended target tracking

Extended target tracking algorithms require the estimation of the target's shape in addition to its kinematic properties. The star-convex model can describe the details of the target's shape compared with the traditional elliptical model. The shape estimation based on the star-convex model is a complex nonlinear problem, which also contains the uncertainty of the measurement distribution on the surface of the target. Therefore, we propose a novel nonlinear measurement transformation (MT) to map measurements into a high dimensional space, so that the transformed measurements have a more linear relationship with the states. The MT also extracts the uncorrelated distance and angle information of the measurement in the local polar coordinates to improve the stability and accuracy of the shape estimation. Then a linear minimum mean square error estimator is constructed using the nonlinear uncorrelated MT, and a Bayesian filter under Gaussian assumptions is derived. Moreover, we also propose a novel mathematical metric based on the star-convex model that measures the distance between the reference shape parameters and their estimated values. Simulation results demonstrate the effectiveness of the proposed algorithm in the case of high sensor noise, low measurement rate, and dense sensor clutter. (C) 2021 Elsevier B.V. All rights reserved.