Radar-Centered Multi-Object Tracking With Moving Vehicle Platforms: An Initial Investigation

Abstract

Radar sensors mounted on moving vehicle platforms present a promising solution for maritime surveillance. However, these systems face challenges due to measurement clutter and variations in the sensing environment. This thesis presents an initial investigation into radar-centered multi-object tracking using the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter within the Random Finite Set (RFS) framework. The objective is to evaluate the GM-PHD filter as a robust and computationally efficient alternative to classical multi-target trackers in maritime radar applications. The proposed system utilizes density-based clustering algorithms, such as HDBSCAN, to aggregate radar detections into structured measurement sets suitable for PHD filtering. The GM-PHD filter recursively estimates the multi-target intensity function, inherently accounting for target births, deaths, and missed detections without requiring explicit data association. This enables reliable tracking under heavy clutter and varying detection conditions. Experimental analysis was conducted using publicly available maritime radar datasets, including the DLR MANV and DARC collections. The results show that the GM-PHD tracker achieves comparable root-mean-square error (RMSE) performance to established algorithms such as the Joint Probabilistic Data Association (JPDA) filter, without the need for initial target assumptions or data association. These findings indicate that PHD-based radar tracking provides a strong foundation for future extensions toward radar-based Simultaneous Localization and Mapping (SLAM) on moving platforms.