Research On Bernoulli Filter In Target Tracking

Recently,Bernoulli filter based on the random finite set theory has attracted much attention.It can avoid the data association in the process of state estimation and provide a rigorous theoretical foundation for joint detection and tracking.In this paper,target tracking under dense clutter and maneuvering are the main motivations,and various algorithms are studied based on Bernoulli filtering.The main results are as follows: 1.CBMeMBer filter aided with amplitude information: the traditional CBMeMBer filter exhibits a noticeable positive bias in cardinality estimation under high clutter density or low SNR environment,an improved CBMeMBer filter aided with amplitude information is proposed.The proposed filter introduces the standard output of radar system—amplitude information to argument the measurement space,and models the amplitude information with Rayleigh distribution.The amplitude likelihood function is respectively derived when the target SNR is known and unknown.Taking into account the practical application of the filter,the birth target density is adaptively generated using the measurements.To evaluate the effectiveness and efficiency of the proposed filter,a numerical example is tested under different clutter density and SNR,the simulation results demonstrate that the proposed filter can effectively estimate the multi-target states and improve the estimation accuracy.2.Multiple model Box CBMeMBer filter: When the actual system output quantitative measurements,the traditional MM-Particle-CBMeMBer filter needs a large amount of particles to approximate the posterior density,this will inevitably lead to real-time computation problems.Based on box particle filter and interval analysis technology,a box particle implementation of the MM-CBMeMBer filter for maneuvering target tracking is proposed.Simulation results show the high computation efficiency of the proposed filter,only a small number of boxes can guarantee the estimation performance,and the calculation time is significantly reduced than the particle implementation.3.Multiple model Bernoulli particle filter: The IMMBPF is only a simple combination of IMM and the particle implementation of BF.The iterative process of the algorithm has a lot of particle interaction.Usually,for linear,weak maneuvering model,a large number of particles are unnecessary but also bring additional computational burden,it may be appropriate to reduce the number of particles.For nonlinear,high maneuvering model,in order to get a better approximation,we can use more particles.An enhanced MMBPF filter is proposed,the number of particles for each model is given in advance according to different models and there is no interaction between the particles,which can effectively improve the execution speed of the algorithm.Finally,the estimation performance of the algorithm is verified by multiple sets of simulation conditions.