Polar Clustering and Stabilized Cluster Likelihoods for Extended Object Maritime Radar Tracking

Kurzfassung

Accurate maritime situational awareness is essential for safety, security, and environmental protection. While the Automatic Identification System (AIS) is widely used, it is vulnerable to jamming, spoofing, and intentional manipulation. These limitations can be mitigated through independent validation using marine X-band radars, where high-resolution coastal sensors detect extended targets with distance-dependent sparsity and raw data in polar coordinates. This paper evaluates a polar-domain clustering algorithm together with its impact on a family of extended-object multi-target tracking (MTT) frameworks of increasing complexity. A Random Matrix Model (RMM) based estimator incorporating radar intensity information is useed for a combined centroid and extension tracking. Beyond the front end, we detail the tracker mathematics required for robust operation on high-resolution clusters. The modifications include a stabilized cluster-likelihood formulation that avoids the clutter-power penalty of point-based extended-object models, a fully log-domain hypothesis-generation backbone shared by the IPDA, JIPDA, and PHD filters, and a measurement-driven birth (MDB) model that removes the one-frame latency of the classical Gaussian-mixture PHD. Performance is validated on real-world X-band radar data within selected Regions of Interest (ROI) and at different distances from the sensor. Results show that clustering in the polar domain is more robust against the distance variability of detections and benefits the overall performance of the tracking algorithms.