Maneuvering extended object tracking using extension-deformation approach with multiple reference extensions

This paper proposes a new approach for tracking of a maneuvering extended object with its kinematic state and extension jointly estimated. This approach is developed based on the extension-deformation framework, in which an object extension is represented as a deformation from a simple reference one. This deformation is achieved by relocating specific control points from the reference to their actual positions in the object extension. Consequently, given a reference, a complex object extension can be effectively captured by these control points. This formulation reduces the modeling and estimation of an object extension to the state estimation of its control points. Notably, the estimation accuracy improves when the chosen reference extension closely matches the true extension. However, during maneuvers, the object extension may vary over time, necessitating dynamic adaptation of the reference. To address maneuvering extended object tracking (EOT), this paper employs a hybrid system that incorporates multiple reference extensions to model time-varying object extensions. A multiple-model approach, based on the extension-deformation framework, is subsequently developed. Simulation results for maneuvering EOT demonstrate the effectiveness of the proposed approach.