A Framework for Batch Processing Tracklets in the GM-PHD Filter

Dalton Durant

doi:10.1016/j.asr.2025.11.124

Return

A Framework for Batch Processing Tracklets in the GM-PHD Filter

2025-12-03

PRE

DOI：10.1016/j.asr.2025.11.124

Original

Original request for help

Save

Abstract

En 中文

The objective of this article is to propose a framework for which tracklet data can be batch processed and used within the update step of the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter. A tracklet represents a sequence of sensor measurements that are associated with the same target over a limited time interval, where if there is high confidence that a tracklet is associated with a true target, it can serve as valuable a priori information, helping to reduce the impact of clutter and improve multi-target filtering (MTF) performance. However, traditionally, the GM-PHD filter is not outfitted to handle tracklets within its update step and can only incorporate sensor measurements individually. Therefore, instead of updating recursively with individual sensor measurements, the proposed approach updates the GM-PHD filter’s posterior intensity function using tracklets. This article achieves this by pre-processing tracklets into a batch solution of the state space, where the batch solution is then treated as a noisy measurement of the multi-target state. This article uses two different methods for batch processing: Least Squares (LS) for mildly nonlinear and non-Gaussian systems and Markov Chain Monte Carlo (MCMC) with Kernel Density Estimation (KDE) for more complex systems. In the presented radar crossing and cislunar debris tracking examples, the results show that the proposed framework for batch processing tracklets within the GM-PHD filter improves MTF accuracy, demonstrating that they provide more informative updates that reduce the impact of clutter.