Research On Box-particle Filter Based Labeled Random Finite Sets Multi-target Tracking

Multiple target tracking as a key technology in the field of information fusion has broad applications in military defense and civil engineering,and been received wide attention of scholars from all over the world.With the in-depth study of random finite sets filtering theory and labeled random finite sets filtering theory,multi-target tracking technology has been developed effectively.Box particle filter is a kind of nonlinear filtering algorithm,combined with the interval analysis and sequential Monte Carlo method,which has the advantage of less number of particles,high computational efficiency,and convenience of dealing with unconventional measurements.Based on box particle filter and labeled random finite sets filtering theory,this thesis makes an in-depth study on multi-target tracking.The main work is as follows:Firstly,a novel approach of multi-target tracking called box-particle ?-generalized labeled multi-Bernoulli(BP-?-GLMB)based on the labeled random finite sets filtering theory and box particle filter is proposed.The algorithm can not only estimate the multi-target states and number,but also effectively estimate the target trajectories and especially the algorithm shows low computational complexity.Simulation results show that BP-?-GLMB maintains the same tracking precision as SMC-?-GLMB with less number of particle,shorter run time and higher efficiency.Compared with BP-CBMe MBer,BP-?-GLMB can estimate the target trajectories.Moreover BP-?-GLMB provides a more accurate estimate of the target number and states than BP-CBMe MBer in the case of high clutter rate.Secondly,based on the combination of box particle filter and labeled multi-Bernoulli filter,box particle labeled multi-Bernoulli(BP-LMB)is proposed to address the problem that ?-GLMB has high computation cost.The computational efficiency of multi-target tracking is improved by BP-LMB which is regarded as an efficient approximation of BP-?-GLMB.Simulation results show that compared with BP-?-GLMB and SMC-LMB,although the accuracy of the proposed algorithm is slightly lost,the run time is short and the computational efficiency is high.Finally,combined with the application of interval analysis method in the extended target tracking,multiple extended target BP-?-GLMB and multiple extended target BP-LMB are proposed.These two kinds of algorithm inherits the advantages of box particle filter and labeled random finite sets filtering.The proposed algorithms are able to reduce the computational complexity of extended target tracking and guarantee good filtering performance and trajectories estimation performance at the same time.The simulation results show that the proposed algorithms can effectively achieve multiple extended target states and trajectories.