A novel multi-ellipsoidal approach to group target tracking using variational approximation

Group target tracking that estimates the position and shape of collective objects is a hotspot area in radar observation. Group targets, such as birds and UAVs, often exhibit complex spatial structures and highly dynamic behaviors. However, traditional multi-ellipsoidal tracking methods rely on a simple random matrix to model sub-object extension, where orientation and extension size are coupled. The initial sub-object model may fail to adapt to dynamic changes in the sub-object state, especially when the sub-object rotates. This paper introduces a modified multi-ellipsoidal tracking method, where the orientation of each sub-object is represented as an independent Gaussian random vector, separate from its extension size. The extension size is described by a diagonal symmetric positive definite matrix, with its elements following inverse Gamma distributions. A variational Bayesian approach is applied to approximate the complex posterior distribution. Real-data experiments validate the effectiveness of the proposed method.