Research On 3D Multiple Extended Target Tracking Algorithm Based On Poisson Multi-Bernoulli Mixture Filter

With the continuous development of sensor technology,the detection accuracy of sensors is getting higher and higher,so that the research of target tracking is no longer satisfied with only tracking the state of the target's center,but also needs to fit the shape of the target at the same time.Multi-extended target tracking technology came into being.Among them,random set theory has gradually become a prevalent direction in the field of multiple extended target tracking due to its high filtering accuracy and avoiding complex data association.Based on the extended target Poisson Multi-Bernoulli Mixture(PMBM)filter of random set theory,this thesis mainly studied the extended target tracking in threedimensional space.The specific research content is as follows:First,aiming at the problem of multiple extended target tracking in three-dimensional space,an ellipsoid threshold is first proposed to achieve measurement partition in threedimensional space,and then the random matrix model is used to roughly fit the extended shape of the target to an ellipsoid.In addition,in order to track three-dimensional extended targets of arbitrary shape,the Gaussian Process Regression is combined with the PMBM filter,and the 3D-GPR-GGM-PMBM extended target tracking algorithm in threedimensional space is proposed.Relying on the good probability inference ability of Gaussian process regression,the joint estimation of the target center state and the extended shape can be accurately realized.The simulation results verify the effectiveness of the two proposed algorithms in estimating the extended target state in the three-dimensional space.Then,in view of the current estimation algorithms for arbitrary shape targets,most of them use nonlinear filtering methods,which have the problems of large calculation cost and long running time.A PMBM filters based on Extension-Deformation models are proposed in twodimensional and three-dimensional spaces respectively.Firstly,PMBM is used to accurately estimate the center of the target state,secondly,the complex shape is regarded as a deformation of the simple shape to realize the estimate of the extended shape.Since the measurement model of the algorithm satisfies linearity,the Kalman filter is used for estimating center state and extended shape,which greatly reduces the running time of the algorithm.Simulation verified the tracking performance of the proposed filter in twodimensional and three-dimensional space.At last,aiming at the tracking problem of two-dimensional and three-dimensional multiextended targets in maneuvering scenarios,due to the superiority of PMBM filter for multiextended target tracking and the good performance of Interacting Mulitple Model algorithm in tracking maneuvering targets,these two methods are combined and IMM-GGIW-PMBM filter is proposed to achieve multiple extended target tracking in maneuvering scenarios.Finally,a simulation is designed to prove the effectiveness of the proposed IMM-GGIWPMBM filter in tracking maneuvering targets in two-dimensional and three-dimensional space.