Research On Particle Filter And Its Application In Wireless Localization And Tracking

The application of particle filter is extremely broad, such as location tracking, voice and image processing, fault detection. Although particle filter is the mainstream method to solve nonlinear non-Gaussian problems, it is still not mature enough, and there are some drawbacks. For example, the particle degradation problem, loss of diversity after resampling, and large calculating quantity. Currently, particle filter algorithm is improved mainly in two ways, the first one is selecting the appropriate proposal distribution. Another is to improve the resampling algorithm. This paper is going to improve particle filter from the two aspects, and the particle filter applied to wireless location and tracking is considered in the paper.Against particle degradation problem, the appropriate proposal distribution is designed. Taking into account the fact that particle filter are seeking to optimize the entire particle set, this paper gives an algorithm which is named as MAFSA-PF. The MAFSA-PF uses the modified version of AFSA to optimize the proposal distribution by driving particles to high likelihood area. For the nonlinear single variable economic model,when the variance of measurement noise is 10-5,Compared to UPF, the filtering accuracy of MAFSA-PF is improved by 83%, and it is close to the filtering accuracy of PSOAI-PF.Consider a case in which there is a separable Gaussian subsystem in nonlinear system, this parpar gives an algorithm which is named as MF. The idea of MF comes from MPF. With MF, MAFSA-PF and KF are used to estimate the nonlinear and linear parts of the system state respectively.For two-dimensional radar target tracking model, the localization error of MF decreases by 17.0% than MPF.In the paper, the deficiencies of Systematic resampling algorithm are analyzed, and then the solution is given. Firstly, take effective solutions to ensure medium particle weights particles preserved. Then we learn from the artificial immunization, and add mutation operator to particles, then the particles increase diversity. The matlab simulation results under different models verify that the output particle set of resampling method gived in this parper is more diversified and effective.The mobile node tracking in Wireless Sensor and Actor Network is study in this paper too. The interactive multiple model and improved particle filter based the improved resampling algorithm are combined to obtain a target tracking algorithm IMM-PF-IR. The matlab simulation results show that the performance of IMM-PF-IR is better than IMM-PF when the measurement noise is small. In addition, the simulation results based on field test data sets further evaluate the validity of the proposed algorithm. The tracking accuracy of IMM-PF-IR is slightly lower than IMM-PF when the received signal strength being used.However, the tracking accuracy of IMM-PF-IR is improved by 34.9% than IMM-PF when the estimating distance being used.