Application Of Multi-Bernoulli Filter In Target Tracking

Multi-Bernoulli filter is a kind of representative method in the framework of stochastic finite set system with scientific and rigorous theoretical foundation.In recent years,This algorithm has attracted much attention for its high performance of efficiency and stability in target tracking.It can avoid the process of data association and solve sorts of uncertainty problems under the complex application background.In view of this,this paper focus on the application research Based on MB filter in target tracking.Specifically,in consideration of filtering algorithm improving and sensor control strategy,some novel solutions have been proposed aiming at solving the problems and shortcomings in the existing research,which make contributions to improving the performance of target tracking and extending applicability of those researches.The main contents of this paper are as follows:Firstly,a study on how to improve the tracking performance of multi-model Bernoulli filter(MMBF)algorithm is carried out.The particle realization of regular MMBF algorithm allocates the number of particles according to the probability of different models.In this way the statistical characteristics of models with small probability can not be fully represented,which leads to a large peak error and slow convergence speed When target moving mode changes.In this paper,an improved MMBF implementation method is proposed.By pre-setting the number of particles of each model and updating the model posteriori probability with the model likelihood function,the target state estimation and the model probability estimation are carried out separately.It effectively overcomes the problem of model switching delays caused by small number of particles.At the same time,in order to solve the problem of particle degradation in resampling,particles with large weight are used to optimize particle with small weight.The simulation results show that the proposed method can reduce the peak error and result in a more stable tracking performance.Secondly,aiming at the high computational complexity of MB filter in sensor control application,a novel scheme based on Cauchy Schwartz(CS)distance is proposed.In this method,the predicted particle set is resampled by using an auxiliary variable and the sampled particles are used to approximate the density function of multi-targets.We chose the CS distance which is easy to calculate as the criterion to measure the difference between the prior density and the posterior density.Then accomplish the proposed control strategy Based on the principle of gain maximization.Simulation results verify the effectiveness of the proposed method.Finally,a sensor control scheme which takes advantage of labeled Multi-Bernoulli filter(LMB)is proposed.This solution is more applicable for multi-sensor background compared to those control methods based on MB filter because the latter ones have difficulty to output target tracking trajectory.We still use an auxiliary variable to resample predicted particles and choose the CS divergence between prior information and posteriori information as the evaluation criterion to study sensor control strategy.In addition,a control method implementation based on posterior expected error of cardinality and state(PEECS)is given.Simulation results show that the two types of sensor strategies based on LMB are effective and the performance of multi-target tracking is more accurate and stable.