Interactive multi-model maneuvering extended target tracking algorithm with variational Bayes inference

Abstract

To enhance the tracking accuracy of multi-ellipse extended target tracking for maneuvering targets, we propose a Variational Bayesian-based Interacting Multiple Model Maneuvering Extended Target Tracking Algorithm. The algorithm leverages multiple kinematic models within an Interacting Multiple Model framework to predict the target state, enabling better adaptation to various target kinematic modes during the time update. Following the prediction step, variational Bayesian inference, which includes inference of unknown measurement noise, is applied independently within each model to estimate the target’s position and shape. Then, model probabilities are updated based on the likelihood functions, and the model transition probabilities are adaptively learned. Concurrently, an ellipsoid fusion method is employed to combine the shape estimates from different models, while kinematic states are fused using model weights, yielding a robust overall target state estimate. The feasibility and performance of the proposed algorithm are validated by comparing its target tracking simulation results with those of other algorithms under different scenarios.