Research On Multi-target Tracking Algorithm Based On Random Finite Set

Multi-Target Tracking(MTT)technology refers to a method to continuously estimate the number and motion status of targets using sensor measurement data.It is playing an increasingly important role in signal processing with the continuous improvement of sensor performance.Aiming at the problem of multi-target tracking under the condition of single platform passive detection,the Probability Hypothesis Density(PHD)multi-target tracking algorithm under the framework of Random Finite Set(RFS)theory is studied in this paper.This paper first introduces the related theory of random finite sets and the multi-objective bayesian filter,and then focuses on the realization of the first-order moment approximation method of multi-objective bayesian filter under linear Gaussian conditions——Gaussian Mixture Probability Hypothetical Density(GM-PHD)Filter,and proposes related improvement methods for its deficiency in engineering applications.The standard GM-PHD algorithm models the PHD of the new target as a Gaussian mixture distribution,which can only track the target that appears at a specific position,but is not suitable for tracking scenario where the target appears randomly.To solve this problem,this paper models the new target PHD as a partially uniform distribution and propose a measurement set division strategy for this model.By filtering out the clutter at each moment,the interference of the clutter on the target state estimation is reduced and the computational efficiency of the filter is improved.Smoothing algorithm based on GM-PHD filtering is also studied is this paper,including forward-backward smoother and Rauch-Tung-Striebel(RTS)smoother.Aiming at the abnormal smoothing problem of the forward-backward smoothing algorithm when the target is missed and disappeared,an improved algorithm is proposed.By comparing the filtering and smoothing results,the weight of the Gaussian component of the missing and dead terms is updated to improve the accuracy of the estimate of the number of targets.In addition,in order to solve the problems of target label confusion and low state estimation accuracy in the cross-target scenario,this paper designed a weight correction scheme based on the assumption of one-to-one relationship between target and measurement.By optimizing the PHD distribution,the improved algorithm can correctly divide the different target and have a high estimation accuracy of cross-target.