Editor: MA Mingyang | Nov 11, 2021

In a study published in Signal Processing, Dr. MA Mingyang from Changchun Institute of Optics, Fine Mechanics and Physics (CIOMP) of Chinese Academy of Sciences (CAS) and his collaborators found that radiation intensity can be used to address the drawback of Gaussian mixture probability hypothesis density (GM-PHD) filter in close target tracking.

GM-PHD, a multi-target tracking (MTT) filter based on random finite set (RFS), is widely employed in modern search and tracking system because of its fascinating capabilities in execution efficiency. Compared with traditional tracking method, it provides an approximate solution to Bayesian framework for MTT and avoids the explicit association between measurement and target. However, due to not adopting the one-to-one principle, the mistaken estimation is prone to occur in GM-PHD for close target tracking. How to distinguish the closely spaced targets in tracking filter remains a question.

Considering the different radiation intensities of different material targets, researchers turned the problem of close target tracking into the identification of closely spaced targets. The radiation intensity was represented by signal-to-noise ratio (SNR) and integrated into measurement space as an augmented state. Further, in order to reduce the influence of environment factors on real target SNR, a SNR probability density function (PDF) was constructed. The analysis of simulation result and actual target data show that the Gaussian function can obtain a better modeling performance on the variation of target SNR in temporal dimensionality. This is the first time that  SNR PDF has been used to model the target SNR distribution. Moreover, the PHD recursion is improved by the new likelihood function of radiation intensity.

In addition, researchers also presented a labeling update scheme for managing the tab of target track. Traditional Bayesian tracking filter adopted labeling method to assign the tabs for target tracks as their identities. However, the management method was ignored. The proposed management scheme reserved the tab of target with the highest weight for each Bayesian update step. The identities of other targets will be cleared. This update scheme prevents the false propagation of tabs, and solves the falsely merged problem and fractured problem when the target tracks are cross. It is an important supplement to traditional labeling method. This is the first time that providing a track management method for multi-target Bayesian filter.

This study solved the close target tracking problem of the classic GM-PHD filter and proposed a labeling update scheme for track management.

Fig. 1 Simulation results for two filters in the scenario of two targets with crossing motion; (a) simulation scenario; (b) tracking result of GM-PHD filter; (c) tracking result of radiation intensity GM-PHD filter

Contact:

Author: Dr. MA Mingyang

Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun, Jilin, 130033. China

E-mail: mamingyang17@mails.ucas.ac.cn

Article links: https://doi.org/10.1016/j.sigpro.2021.108196