Research Of Multi-target Tracking Algorithm Based On Random Finite Set

With the development of sensor and information technology,the multiple target tracking(MTT)technology raises rapidly,and also plays an important role in military defense field.Under a complicated battlefield environment with many uncertain factors,the traditional algorithm whose calculation will dramatically increase as the number of target goes up in the tracking process causes the accuracy becomes lower and lower.Therefore,the Random Finite Set(RFS)theory,a new method which provides a new idea for multi-target tracking,is created to solve this problem by building separate dataset on both target state and measurement and implementing the recursive filter algorithm based on the significance of set,which avoids the process of data association and solve the problem of the changing target number.This paper focuses on the multi-target tracking method based on RFS theory(1)Research on basic algorithm of multi-target tracking based on RFS theory.This paper introduces the basic filtering algorithm of multi-target tracking technology,and focuses on its basic theory of RFS and application in multi-target tracking.Two filtering algorithms are studied carefully which is the probability hypothesis density(PHD)filter and the cardinality probability hypothesis density(CPHD)Filter.And the Gaussian mixture implementation steps of PHD and CPHD are deduced,and their advantages and tracking performance are compared and verified by simulation result.(2)Research on improved Gaussian mixture probability hypothesis density(GM-PHD)filtering of multi-target tracking algorithm.The engineering application of this algorithm is limited due to the missing the target in the strong clutter background of GM-PHD algorithm as well as the low efficiency in traditional Gaussian mixture implementation of PHD and CPHD caused by the uncertain position of the new target and the factor that the intensity function needs to appear in the whole monitoring area when the prior probabilities are initialized.Therefore an adaptive target birth intensity GMPHD algorithm(ATBI-GMPHD)is proposed for solving the problems above by adding an identity to the original algorithm to distinguish between new targets and survival targets,and dividing the filtering process of these two targets into two channels and using the obtained measurement of each scan to generate the new intensity adaptively based on the measured values,which avoids the assumption of prior probabilities and reduces the amount of calculation effectively.The simulation results show that the improved algorithm has good performance.(3)Research on improved extended target Gaussian mixture probability hypothesis density(ET-GMPHD)algorithm for extended maneuvering multi-target tracking.With the application of high-resolution sensors in recent years,the extended target(ET)has become a hotspot in the field of multi-target tracking.The RFS theory is applied to solve the extended multi-target tracking.To solve the problem that the ET-GMPHD filtering algorithm cannot track the extended maneuvering multi-target accurately,the interactive multi-model(IMM)method is introduced into this paper,and the IMM-ET-GMPHD filtering algorithm,an interactive multiple model multi-target which extended maneuvering Gaussian mixed probability hypothesis density algorithm,is proposed to reach a higher tracking accuracy on extended maneuvering multi-target in complex scenes.Also,Gaussian component identification method is added to solve the problem that ET-GMPHD algorithm cannot obtain the complete target of each target and realize the track management of maneuvering extended multi-target.Finally,the traditional ET-GMPHD algorithm and the improved algorithm are compared.The simulation results show that the improved algorithm has excellent performance.