Online Clutter Measurement Modeling for Surveillance Radar Tracking via Normalizing Flows

In modern surveillance radar systems, ground and atmospheric clutter can obscure target signals and impair the accuracy of measurement processes. Accurate estimation of clutter measurement density—a critical yet often unknown parameter—plays a vital role in enhancing radar measurement and multi-target tracking (MTT), particularly in dynamic environments. This paper proposes an online clutter measurement density estimation method based on a Normalizing Flow (NF) model. The method extracts clutter characteristics from radar measurement streams and leverages an asynchronous training framework to adaptively model changing clutter distributions in real-time. Unlike conventional approaches that rely on limited gated measurements and exhibit reduced estimation accuracy, the proposed method achieves superior performance in both static and dynamic clutter conditions. Simulation results demonstrate enhanced robustness and precision in clutter measurement density estimation, leading to improved measurement reliability and reduced false alarms in MTT. Validation using real-world radar data—from highway contour extraction and dynamic atmospheric monitoring—confirms the practical applicability of the approach in intelligent and adaptive radar measurement systems.