Adaptive extended object tracking based on variational Bayesian with couple under unknown noise

The random matrix approach offers an efficient framework for extended object tracking (EOT) which directly facilitates the estimation. Its accuracy is inherently related to the statistical properties of noise, which typically assumes available. However, the noise prior is often difficult to acquire under environmental uncertainties. Consequently, existing methods suffer from a mismatch between the assumed and actual noise, leading to significant tracking performance degradation. Furthermore, estimating noise covariance gives rise to parameter coupling. The coupling, which current methods fail to resolve adequately, makes it impossible to directly derive a variational Bayesian (VB) solution. Therefore, a novel EOT is developed to resolve the coupling under unknown noise conditions. First, a decoupling model is developed, which enables coupled parameters to be estimated independently. Second, the process noise covariance matrix (PNCM) and measurement noise covariance matrix (MNCM) are modeled as inverse Wishart (IW) and inverse Gamma (IG) distributions, respectively. Through these models, the VB approach is utilized to derive the posterior of the kinematic states and extension. The proposed approach is evaluated in typical EOT scenarios. The results exhibit effectiveness performance and robustness of our approach under unknown noise conditions.