Maneuvering extended object tracking based on constrained expectation maximization

This paper addresses the problem of maneuvering extended object tracking, in which the object extension (OE) and the turn rate are identified simultaneously. Due to its non-linearity with respect to the turn rate, the state transition is converted into a linear form of a newly defined hyper-parametric vector by parameter substitution. For the hyper-parametric equality constraints (HECs) introduced in the transformation, a constrained expectation conditional maximization algorithm is designed. The HECs are projected onto the conditional expectation function for regularization, so as to realize the iterative identification of multi-parameter (i.e., OE and turn rate). The transformation and CECM optimization bring the advantage of avoiding nonlinear model approximation, which is important for the convergence and accuracy of state estimation. Finally, simulation results demonstrate the superiority of the proposed method in terms of both estimation accuracy and identification effectiveness.