Research On Multiple Target Tracking Method Based On Random Finite Set And Software Implementation

The problem of multi-target tracking in complex environment has brought great challenges to radar system.The current mainstream multi-target tracking methods are all based on the theoretical framework of Bayesian filtering.Thereinto,the Gaussian Mixture Probability Hypothesis Density(GM-PHD)filter algorithm based on the Random Finite Set(RFS)shows good tracking performance in clutter environments.Compared with the traditional multi-target tracking algorithm based on data association,it has the advantages of no data association,low complexity,great tracking capabilities for an unknown and time-varying number of targets,etc.,which makes it attract wide attention.However,GM-PHD filter requires the priori knowledge of the PHD function of the newborn target containing the target initial position information and is inapplicable to the nonlinear systems due to the linear Gaussian assumption,which causes its application scenarios to be severely limited.In addition,the GM-PHD filter algorithm has the characteristics of estimation result of RFS and no data association,which makes it not be directly applied to track extraction and management.The corresponding improvement methods for the above problems are studied,and the track management method based on data association in radar system is realized.The main contents are as follows:1.For the problem that GM-PHD filter requires the priori knowledge of the PHD function of the newborn target and is inapplicable to the nonlinear systems,the measurement-driven birth(MDB)PHD model and the partial uniform birth(Partially Uniform Birth,PUB)PHD model is combined with Cubature Kalman Filter(CKF)respectively,and two improved methods are studied: MDB-CK-GM-PHD filter and PUB-CK-GM-PHD filter.Both algorithms use cubature numerical integration rule to achieve nonlinear filtering,where the MDB-CK-GM-PHD filter uses the set of measurements to adaptively estimate the PHD function of the newborn target,while the PUB-CK-GM-PHD filter approximates the observed part of the newborn target PHD function to a uniform distribution to avoid setting it.The simulation experiment verifies the effectiveness of the two algorithms in the nonlinear system of the unknown newborn target PHD,and the performance of PUB-CK-GM-PHD filter is equivalent to that of GM-PHD nonlinear filter of known newborn target PHD.2.For the problem that the GM-PHD filter is unable to directly extract and manage the tracks,a track management method based on the tag GM-PHD filter is studied.This method introduces tags containing target ID,confirmation and missed count variables,and realizes track management based on ID matching and data association.Simulation experiments show that this method realizes the track extraction and management,and improves the estimation accuracy in the scene of target cross-motion.In addition,For the practical problems of track management in radar system,a track management module based on C/C++ was designed and implemented.This module takes three-point fast track initiation as the key and data association as the core.Software experiments show that the module enables the radar system software to have the function of track management and has certain practical value.