Research On Ground Multi-target Tracking Method Based On Random Finite Set

The ground multi-target tracking technology refers to estimating the number,states and trajectories of multi targets by observations,which can be collected by the single sensor or multi sensors.Actually,affected by factors such as complex terrain,the states(position,velocity,acceleration,etc.)and number of the ground targets are time varying.Meanwhile,the observations and their number are also randomly changed with time,due to the obstruction of tall buildings,undetected cases,and clutters.Thus,the associations between states and observations are rather hard to be constructed.Therefore,the tracking performance of the conventional data associated algorithms cannot be effective.In recent years,algorithms based on random finite set(RFS)has been proposed to tracking multi targets.As for these algorithms,both of states and observations are modeled by random finite set.Using such a model,the associations between states and observations can be neglected.Thus,the RFS based algorithms can avoid the restriction of the association based algorithms.It provides an effective solution for tracking the ground targets.The ?-generalized labeled multi-Bernoulli filter(?-GLMB)is one of the most commonly used algorithms of the random finite set filtering.It can directly estimate the trajectories of the moving targets with robustness.Therefore,in this dissertation,we apply the ?-GLMB algorithm to tracking multi ground targets.Considering the characteristics of strong nonlinearity and strong clutters of the ground target tracking scenarios,we conduct our research with two aspects(1)we integrate the Square-rooted cubature Kalman(SCKF)into the ?-GLMB Gaussian mixture implementation algorithm.Attribute to the significant tracking performance of nonlinear target tracking scenarios,the ?-GLMB Gaussian mixture implementation algorithm can be applied for tracking targets with nonlinear moving models.This algorithm can be applied under environment of remote suburbs with small number of clutters.(2)In the urban environment,due to the influence of factors,such as scattering of tall buildings and high clutter density,the computational complexity of the tracking filter algorithm may be very large.In order to make compromise between accuracy and time,we uses cubature Kalman(CKF)to predict and update the Gaussian parameters of the improved GM-?-GLMB filtering algorithm by sacrificing part of the accuracy.For further reducing the computational complexity,an observation combining strategy based on distance weighting is proposed to improve the CKFGM-?-GLMB algorithm.Thus,our algorithm can be used to track targets in the scene of strong clutter density.We analyze the presented algorithms in this dissertation by simulation.First,we implement the proposed SCKF-GM-?-GLMB algorithm,EKF-GM-?-GLMB and UKF-GM-?-GLMB algorithms at the cases of fixed clutter density and different clutter density.The optimal subpattern assignment distance and estimation error of target number are adopted as the metric.The simulation results show that the SCKF-GM-?-GLMB algorithm improves tracking accuracy of the GM-?-GLMB algorithm.Second,we compare the tracking performance of our CKF-GM-?-GLMB,GM-?-GLMB and SCKF-GM-?-GLMB algorithms.The simulation results show that the GM-?-GLMB has the largest tracking errors.Although the estimated results of CKF-GM-?-GLMB has larger errors than the SCKF-GM-?-GLMB algorithm,the tracking time of our CKF-GM-?-GLMB is smaller than the SCKF-GM-?-GLMB algorithm.In other word,our CKF-GM-?-GLMB can make comprise between tracking errors and computing time.