Improved Particle Filter Algorithm And Its Application

With the development of information technology,nonlinear systems state estimation has gradually become a very active research topic and it is currently of immense interest due to its multiple applications.With the existence of some system application of the nonlinear systems state estimation involving more complex models,filter accuracy requirements need to be improved.Some existing approaches like Kalman Filter and other traditional filter algorithms cannot meet the actual use requirements.As a nonlinear numerical filter method,particle filter can deal with nonlinear and non-Gaussian dynamic system state estimation efficiently.In the case of a more complex nonlinear model,it is not necessary to do a linear estimation of the nonlinear system considering the actual filter requirements.We study the state space filter estimation problem under the Bayesian framework.In order to solve the problem of filter estimation for nonlinear dynamic systems,several representative nonlinear filter methods have been proposed which the most used ones are the extended Kalman Filter and unscented Kalman Filter nonlinear parametric approximation methods.But they all have the constraints of the Gauss hypothesis.In order to accurately describe the state distribution of the system,it may require infinite number of estimation parameters,and it is often difficult to obtain satisfactory results.As a non-parametric random sampling and filter method,the particle filter is widely used.However,there still exist some drawbacks in the algorithm of particle filter.To improve the performance,the research work on particle filter is very meaningful both in the sense of theory and the applications.This paper addresses some critical problems in the applications of particle filter under the background of target tracking.The main work of this paper includes the following aspects:Firstly,we study the basic principles of particle filter resampling algorithm and several classical resampling algorithms.Then,in order to reduce the resampling variance and improve the efficiency of particle filter,two-time resampling algorithm is proposed.By comparing with the simulation results obtained from the conventional resampling algorithm,it can be seen that the two-resampling algorithm can effectively alleviate the problem of particle degradation and greatly improves the accuracy of the particle filter algorithm.Particle filter still has many problems to be improved such as problem related to the convergence of the particle filter algorithm,the lack of particle diversity the running speed of the algorithm,the particle degeneracy problem...In order to solve the above listed problems and to expand the algorithm for the new application field,some changes have been made on two main aspects.Those aspects include the selection of the importance distribution and the improvement of resampling algorithms which allow us to put forward an improved particle filter algorithm called Gauss based auxiliary particle filter algorithm.This algorithm introduces real-time observation into the importance probability density function,and then performs Gaussian transformation on the state of the resampled particles to enhance the diversity of the particles and adjust the weights of the transformed particles.So it improves the weight degradation and the loss of particle diversity caused by the conventional particle filter algorithm.Simulation results show that the proposed algorithm outperforms the general particle filter algorithm.Finally,this paper describes the implementation of Gauss based auxiliary particle filter tracking algorithm in the target tracking framework.Take Bearings-only target tracking as an example,the autoregressive state model and observation model of the object motion are established under the condition of the prior distribution of the known target states,and the prior predictive value of the target state is corrected and updated as the new observation information,so as to realize the estimation of object trajectory.And the simulation results show that the proposed algorithm is effective and robust.