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

Underwater target tracking technology has always been a hot spot and key field of research.With the increasing maturity of target tracking technology,people turn their attention to underwater multi-target tracking technology.Traditional multi-target tracking algorithm is mainly based on data association.Once the accuracy is not high enough,it will lead to a series of problems such as "computational explosion",which is not suitable for multi-target tracking scenarios with high real-time requirements.With the gradual development and improvement of stochastic finite set theory,Probabilistic hypothesis density filtering algorithm based on stochastic finite set framework has been gradually applied in the field of multi-target tracking because of its advantages of low complexity and good tracking effect.The research focus of this paper is also the research of multi-target tracking based on probability hypothesis density filtering algorithm,which is mainly divided into the following parts: 1.Firstly,we study the common target motion model and the common tracking filtering algorithm,and analyze the above algorithms by combining the vector hydrophone.The results show that Particle filtering is better than Unscented Kalman Fliter in terms of tracking effect.UKF,which is better than the conclusion of Extend Kalman Fliter,but at the same time,the EKF algorithm is relatively stable and the algorithm complexity is low.2.Probability Hypothesis Density Fliter,based on random finite set framework and Cardinalized Probability Hypothesis Density Fliter and its performance in multi-target tracking were simulated and compared.CPHD has higher filtering accuracy.PHD filter performs better in the estimation of the number of targets and has higher stability.The reasons why PHD filter is chosen as the research focus in this paper are elaborated in detail.3.Two ways to implement PHD filter:Gauss Mixture--Probability Hypothesis Density Fliter,Gauss Mixture--probability hypothesis Density fliter and the principle of Sequential Monte Carlo--Probability Hypothesis Density Fliter were studied.Compared with SMC-PHD,GM-PHD avoids particle degradation and computational explosion problems.Therefore,GM-PHD is chosen as the basis for subsequent research.4.In view of the relevant problems encountered by GM-PHD filter in the underwater acoustic environment,the corresponding improved algorithms are studied,respectively,by combining GM-PHD and EKF to extend to the nonlinear environment;The management of target track is realized by adding labels to Gaussian components.An improved algorithm is proposed to correct the Gaussian component weight of the missed target under low detection probability,which can retain effective information to the maximum extent and improve tracking accuracy.5.Finally,by processing the experimental data,the correctness of the multi-target tracking algorithm studied above is verified,and the results of positioning error less than 10% and OSPA value less than 25 are obtained.