An Improved Particle Filter Method Based on GPU Architecture

Abstract

The invention belongs to the technical field of particle filter and particularly relates to a modified particle filter method based on a GPU (Graphic Processing Unit) architecture. The modified particle filter method based on the GPU architecture comprises the following steps of: S1, setting the number of particles and an observation moment k at a CPU (Central Processing Unit) end, and initializing the particles at a GPU end; S2, transmitting an observation vector to a GPU video memory, and executing the step S3 when k is equal to 1; S3, carrying out importance sampling at the GPU end; S4, carrying out double sampling at the GPU end so as to obtain maximum likelihood sampling particles at the moment k; S5, obtaining the probability of acceptance of each maximum likelihood sampling particle at the moment k by utilizing the GPU; S6, figuring up an estimated value at the GPU end at the moment k; S7, figuring up a resampling index of each maximum likelihood sampling particle at the CPU end at the moment k, obtaining resampling particles at the GPU end at the moment k according to the resampling index, and serving as initial particles at the next moment; S8, repeatedly executing the step S3 to the step S7 for M times so as to obtain estimated values at M moments.

Description

technical field [0001] The invention belongs to the technical field of particle filtering, in particular to an improved particle filtering method based on GPU architecture. Background technique [0002] Nonlinear filtering problems widely exist in many fields such as signal processing, data communication, radar detection, target tracking, satellite navigation, etc. Such problems can be summarized as state estimation problems of nonlinear systems in the presence of observation noise. The particle filter method is a more general filtering method, which uses a series of sample points with weights to approximate the posterior probability distribution of the state. This method is essentially based on state search. The computational complexity of this method is high due to the large number of particles used in the state approximation. Moreover, there are two main problems in the particle filter: first, when the particle sampling is inaccurate, such as the sampled particles are lo...

AI Technical Summary

Problems solved by technology

At the same time, in view of the high computational complexity of the particle filter method, an implementation method based on the GPU architecture is proposed, which greatly reduces the computing time, improves the real-time performance of the algorithm processing, and can meet the needs of real-time processing.

Images