A Realization Method of Particle Filter Based on Adaptive Kld Box Length

Abstract

The invention discloses a realization method for a particle filter based on a self-adaptive KLD (Kullback-Leibler Distance) box length. The method is realized through a series of steps. According to the method, alternate resampling and KLD sampling algorithms are provided; additional computing quantity of the particle filter is avoided; the execution efficiency of the algorithms is improved; and under the condition that the number of particles is very high, according to the method, the performance of the particle filter is greatly improved. According to the realization method of the self-adaptive KLD box length provided by the invention, the KLD box length is automatically adjusted according to the state distribution of the particles at every moment. When a distribution range of the particles is relatively wide, the KLD box length can be self-adaptively increased, the number of the particles obtained through KLD sampling is prevented from being excessive, and the computing complexity is reduced. When the distribution of the particles is dense, the KLD box length can be self-adaptively reduced, the number of the particles obtained through KLD sampling is increased, and the estimation precision is improved.

Description

technical field [0001] The invention relates to the field of particle filters, in particular to a method for realizing a particle filter based on an adaptive KLD box length. Background technique [0002] The estimation accuracy of the particle filter increases with the increase of the number of particles, so in order to improve the estimation accuracy, the particle filter needs a large number of particles, but in practice, the high dimensionality of the system state and the complex posterior density function require a large number of Particles are used to describe the state and approximate the probability distribution, so the calculation and update of a large number of particles and their weights make the complexity of the particle filter very high. Excessive complexity limits the further development and application of particle filters, especially in some high real-time application environments, so how to reduce the complexity of particle filters is a key problem that needs ...

AI Technical Summary

Problems solved by technology

Since some particles will be discarded after sampling after KLD, the resampling operation of discarded particles is redundant, so the existing KLD particle filter algorithm contains some redundant operations, which increases the computational complexity of the algorithm

[0007] In the existing KLD particle filter algorithm, the parameter value of the KLD box length is fixed, and the fixed parameter value setting will lead to some two situations: when the particle distribution is dense, the number of boxes obtained by KLD sampling Not enough; the particle distribution is scattered, and the number of boxes obtained by KLD sampling is too large

These two situations will cause the adverse consequences of the decrease of estimation accuracy and the increase of computational complexity respectively.

Images