Multi-Ellipse PMBM Filter Using Maximum-Linear-Correlation-Based Measurement Partitioning for Non-Ellipsoidal Extended Object Tracking

The Poisson multi-Bernoulli mixture (PMBM) filter has proven to be effective for multiple extended object tracking. However, PMBM-based extended object tracking methods commonly represent the object extent using an ellipse, which cannot accurately describe the actual shapes of non-ellipsoidal extended objects. When multiple objects are closely spaced, the resulting loss of shape information also makes their trajectories more difficult to distinguish and increases the ambiguity of data association. To address these issues, this paper proposes a maximum-linear-correlation-based multi-ellipse modeling method within the PMBM framework. The measurements are partitioned according to a maximum linear correlation criterion, and the resulting subsets are used to construct multiple Gamma Gaussian Inverse Wishart (GGIW) components that jointly represent the non-ellipsoidal extent. Based on this model, the corresponding likelihood and the prediction and update processes of the NEO-PMBM filter are presented. Its GGIW implementation is denoted as NEO-GGIW-PMBM. Three simulation scenarios compare the proposed filter with the conventional GGIW-PMBM and NEOT-PMBM filters. The results show that NEO-GGIW-PMBM achieves a faster reduction in GOSPA error and yields lower overall GOSPA and extent errors, particularly when multiple objects are closely spaced or intersect.