Online Robust Low-Rank Enhancement and Structure-Aware Tracking for PPI Images from Marine Radar Displays

This is the authors' abstract. We don't add key points for this paper.

Marine radar plan position indicator (PPI) images may be the only observation source available for image-domain post-processing, yet weak target-like echoes are often obscured by sea clutter and persistent structural returns. Missed detections can fragment tracks, whereas permissive front-end processing can generate excessive candidates. This study presents an online robust low-rank enhancement and structure-aware tracking method. A finite history window supports background estimation by Huber iteratively reweighted projection, and raw intensity, local contrast, energy, and positive sparse-residual evidence are fused to generate ordered-statistics constant false alarm rate (OS-CFAR)-type candidates. Structural occupancy, compact-echo protection, multi-evidence aggregation, and multi-window motion cues then suppress structure-induced pseudo-motion. Active-track-first association, two-stage reassociation, and constrained candidate recovery are used to preserve track continuity. Across three PPI sequences acquired at different times at one radar site, structure-aware processing reduced the mean candidate load from 61.16 to 19.01 per frame. Against frame-by-frame dense visual references, the complete method achieved an F1 score of 88.24% and a multiple object tracking accuracy (MOTA) of 76.09%. The results show that, under the evaluated image-domain conditions, the method balances weak-echo retention, suppression of structure-induced false activations, and track continuity. Generalization across devices, sites, and sea states remains to be established.