| Multi-target tracking technology is widely used in popular military and civil fields such as,air traffic control,precision guidance,and automatic driving.Especially in the military field,it directly affects the national initiative in information warfare and intelligent warfare.In the early 21 th century,multi-target tracking algorithm based on random finite set(RFS)came into being with unique theoretical advantages and engineering applicability.In order to break through the uncertainty of target state,sensor measurement and their correlation,this paper uses the joint prediction and update of Generalized Label Multi-Bernoulli(JGLMB)filter as the basis to conduct an in-depth study on the multi-target tracking method:Firstly,the RFS-based multi-target tracking algorithm needs to be approximated by Gaussian mixture or particle filter,and the research is carried out for nonlinear systems.The traditional particle filter does not consider the current measurement information,which causes a large number of particles to be depleted and degraded.The square-root volume Kalman filter is introduced to approximate the importance density function to improve the accuracy of numerical calculations.Based on the more compact structure of the JGLMB filter,the Gibbs sampling algorithm is used for efficient pruning of the transmitted components.Combined with the square-root cubature particle filter(SCPF)algorithm,a universally applicable SCPF-JGLMB robust algorithm is proposed,and the simulation results show that the effectiveness of the SCPF-JGLMB algorithm.Secondly,considering the problems of noise outlier interference,strong clutter density and sudden change of target motion state in the tracking scene similar to multiple "low-slow-small" targets,a tracking method suitable for but not limited to multiple "low-slow-small" targets is proposed.The student t distribution with thick tail is introduced to model the state and measurement noise,so as to reduce the influence of noise outliers on multi-target tracking accuracy.Combined with the advantages of high tolerance of multi-model method for target state transition,the matching degree with the actual target motion state is improved to avoid the problem of missing heel caused by state mutation.For the target echo measurement is easily disturbed by high-density clutter,an adaptive measurement merging method based on distance weighting is proposed to improve the associated value of real measurement.Simulation results show that the proposed method improves the computational efficiency and can be applied to the tracking of multiple maneuvering targets in complex backgrounds.Finally,for the tracking problem of multiple extended targets,the Gamma Gaussian Inverse-Wishart(Gamma Gaussian Inverse-Wishart)distribution is introduced into the JGLMB filter to realize the comprehensive estimation and trajectory tracking of the target centroid motion state and shape;The K-means++,Mean Shift main flow measurement division method has many problems such as the number of measurement divisions and the unstable clutter removal effect.A dual measurement division method based on space-time combination is proposed,and the clutter is removed through the Gaussian kernel function Processing,adaptively determine whether a secondary division is needed to improve the accuracy of the division.Simulation results verify the proposed method can accurately estimate and track multiple extended targets when the target trajectories are parallel,adjacent and crossed,and the environmental clutter density is high. |
