Research On Maneuvering Extended Target Tracking Method Based On Poisson Multi-Bernoulli Filter

In recent years,with the advancement of modern hardware and software technology and the expansion of application scope,target tracking technology has played an irreplaceable role in both civilian and military fields.It has shown great development prospects and research space in the direction of take-away drone delivery,missile defense and underwater tracking.Therefore,target tracking technology has become an attractive research topic in various related fields.In the case of high-resolution sensor monitoring,small target generates only one measurement,called point targets,while targets that generate multiple measurements each timestep are called extended targets.Furthermore,in complex scenarios,there will coexist maneuvering extended and point targets.In this paper,from the perspective of accuracy and efficiency,we mainly focus on the tracking performance degradation problem of Poisson Multi-Bernoulli Mixture filter algorithm for maneuvering irregularly shaped extended target.We further investigate for the multi-target closing or overlap problem.In addition,since it’s hard to obtain the complete trajectory of each maneuvering extended target,we carry out some related research work.The main research work in this paper is as follows:Firstly,we consider the problem of maneuvering target tracking and shape estimation of extended target.The Poisson multi-Bernoulli Mixture filter algorithm performs prediction step based on the prior probability density for parameters of the target motion model.However,when the target takes maneuvering movement,such as changes in direction or speed acceleration.The filter algorithm can neither predict the target state under maneuvering motion model,nor match the true situation,which results in reduced tracking accuracy.Moreover,irregular shaped extended targets may have shape changing problems with different observation angles or sensor distances.Therefore,we propose a multi-Model Poisson multi-Bernoulli Mixture filter algorithm for multiple extended targets.In the algorithm,the extended target is represented by a time-varying multiple sub-targets which follows gamma Gaussian inverse Wishart distribution.Consequently,we can effectively estimate the shape of extended target.Simulation results show that the proposed algorithm has higher accuracy for tracking coexisting maneuvering extended targets and point targets.Secondly,we further consider the improvement of detecting and tracking performance in multiple maneuvering target closing and overlap scenarios.The data association problems are supposed to be solved in multi-extended target filter algorithms to ensure a correspondence between measurement random finite set and target set.The existing methods mainly focus on the two-step data association method of clustering and assignment,which works well when the targets are dispersed.When the targets get closing or overlap,the performance degrades significantly.Therefore,we achieve the PMB approximation through obtaining the marginal association probability and Bernoulli density in the filter update step,and then we truncate the low-weight global hypothesis,and reduce the computational effort of data association.Finally,the fuzzy processing method is further proposed to achieve multi-target tracking in the scenario where point targets and maneuvering extended targets are close to each other.This method is verified to have better performance in simulation experiments.Finally,we consider the practical application value of the algorithm in the target trajectory estimating problem.we combine the proposed multi-model PBP-PMBM algorithm with the trajectory-based PMB approximation method,which makes the proposed algorithm able to maintain complete target trajectory information.In practical applications,the low detection probability value of sensors or maneuvering motion model of target can easily affect the estimation.Therefore,we further integrate a backward smoothing method to reduce the tracking error,making the estimation of target trajectory closer to the true trajectory.Simulation results show that the IMM-PBP-TPMB algorithm based on trajectory set and backward smoothing method performs better in target state estimation and track maintenance,which improves the practical application value.