Research On Multi-target Track-before-detect Method Based On Particle Filter Algorithm

Multi-target tracking and detection in complex environment is an important research topic in modern detection systems as well as in the fields of national defense.Compared with traditional Detect-Before-Track(Detect before track,DBT),the newly proposed Track-before-detect(Track before detect,TBD)method is more efficient.Many studies have shown that TBD based on particle filter is of exceptional potency.Yet,there still exist many problems to be tackled in the field,for example,the curse of dimensionality,the interference of nearby targets and the reasonable setting of particle number,etc.This thesis mainly focuses on the problems of multi-target particle filter TBD algorithms as follows:1.To tackle the curse of dimensionality,the independent partitioned cross-sampling technique is studied,based on which an independent partition particle filter(IP-PF)is proposed.The IP method reduces the dimensions of multi-target states and improves the sampling efficiency with subparticle crossover.The IP method succeeds in making the computational complexity of multi-target tracking linearly related to the target number.2.For the nearby-target interference problem that IP-PF cannot solve,the parallel partition(PP)state-sampling PF algorithm is proposed to ameliorate the situation.While retaining the posterior independence assumption of IP method and sampling conditional on estimated states,the multi-target likelihood is inserted into the computation of the firststage weights so that the neighboring targets is properly taken into account.3.To reduce the computational burden,a particle filter with an adaptive sample-size is proposed to sample based on KL distance inference and further generalized into multitarget scenario,which adaptively determines the sample size by measuring the difference between the prior probability and the posterior probability distribution.4.The engineering implementation of multi-target particle filter TBD is finally studied to solve detection and tracking problem of passive sonar.The proposed PF-TBD algorithm is used to process two batches of offline sonar data.The performance analysis verifies the effectiveness of the proposed TBD algorithm.