Research And Application Of Particle Filter Algorithm Based On Innovation Error

Bayesian theory provides a solution based on the probability distribution form for the problem of nonlinear and non-Gaussian state estimation.Particle filter is a sequential Monte Carlo algorithm based on the recursive Bayesian estimation theory.It is an effective method to deal with nonlinear and non-Gaussian problems and is widely used in many fields such as visual tracking,target positioning,communication,and signal processing.However,the particle filter algorithm has problems such as particle weight degeneracy,loss of particle diversity,curse of dimensionality,and high computational cost in the recursive calculation process.To address these problems in particle filter,an improved particle filter algorithm based on innovation error was proposed in this thesis and applies the improved algorithm to target tracking,simultaneous localization and mapping of mobile robots.The main work in this thesis is listed as follows:Firstly,based on Bayesian theory,this thesis introduces the theoretical derivation of the particle filter and particle flow filter,analyzes the reasons of the particle weight degeneracy.Secondly,refer to the latest observation information,an innovation error structure is introduced into the process of particle flow,the feasibility of this structure is theoretically proved and the update of each particle is independent.In order to get the“velocity field” solution,the Galerkin finite element method is used to obtain the numerical solution,which avoid the numerical instability problem that may be caused by fitting sample prior.Thirdly,considering the possible stiff problem of solving ordinary differential equations,the use of different numerical solutions is discussed and a particle redrawing method is proposed to optimize the improved particle filter algorithm,this method is designed by designing a fitness function,in this way there is no need to assume the distribution of particles before redrawing.Finally,the simulation experiment of the algorithm is implemented,the improved particle filter algorithm and other comparison algorithms are applied to the simulation of nonlinear systems and Fast SLAM models,which verifies that the proposed algorithm has better estimation accuracy and higher calculation efficiency in multi-dimensional situation.