Research On Tracking Before Detection Algorithms Of Dim-small Targets Based On PMBM Filter

Track-Before-Detect is more suitable for detecting and tracking dim-small targets in the environment of low SNR than the traditional Detect-Before-Track technique.Random Finite Sets can effectively track dim-small targets after combining with Track-Before-Detect on account of its ability to weaken clutter correlation effects.Poisson Multi-Bernoulli Mixture filter has the advantages of high tracking accuracy and accurate estimation of the number of targets as a new achievement of Random Finite Sets.This thesis focuses on the Track-Before-Detect algorithms of dim-small targets based on Poisson Multi-Bernoulli Mixture filter.The specific research content is as follows: Firstly,an improved Poisson Multi-Bernoulli Mixture filter is proposed to solve the problem that the spawn targets are not included in the dynamic model of Poisson Multi-Bernoulli Mixture filter,which makes Poisson Multi-Bernoulli Mixture filter unable to track spawn targets.The spawn targets are included in the dynamic model as Poisson components,then the prior density and the posterior density become conjugate in the update process.The prediction and update process of the improved filter and the Gaussian mixture realization are shown in the context.The simulations show that the improved Poisson Multi-Bernoulli Mixture filter can track several spawn targets at the same time.Secondly,in order to solve the problem that Poisson Multi-Bernoulli Mixture filter can not be used in dim-small target detection and tracking,a new dim-small target Track-BeforeDetect algorithm based on Poisson Multi-Bernoulli Mixture filter is proposed.Based on the measurement characteristics in Track-Before-Detect scenes,the corresponding likelihood function is calculated and brought into the update equation.The probability generating functional of the new update equation is used to obtain the probability of transition from a Poisson component to a Bernoulli component and the update steps used in Track-Before-Detect scenes can be acquired by means of the probability.Then the proposed algorithm is implemented using Sequential Monte Carlo technology.Simulations verify the real-time and accuracy of the proposed algorithm under a variety of scenarios.Finally,to solve the problem of excessive data transmission and memory consumption in the traditional Track-Before-Detect algorithm,a threshold-measurement Track-Before-Detect algorithm based on Poisson Multi-Bernoulli Mixture filter is proposed.By setting the threshold value,the indices of pixels whose intensities exceed the threshold are gathered into an index set,and the detection probability of each pixel is calculated.The complexity of likelihood calculation is reduced by using the detection probability to calculate the measurement likelihood.The use of the index set avoids the transmission and storage of a large number of useless data and improves the utilization of limited resources.Simulations show that the algorithm can effectively track the dim-small targets while reducing the complexity of the algorithm and the use of storage space.