Research On Particle Filter And Its Application In Target Tracking

With the development of modern science and technology and the needs of themilitary, higher requirements are put forward for the accuracy of target tracking.Linear gaussian model can not accurately describe the target motion model and thecorresponding kalman filter can not accurately estimate the target motion parameters.In recent years, the nonlinear/non-gaussian model has achieved great developmentand utilization, so the study of nonlinear/non-gaussian model filter method is veryimportant.Particle filter is one of the important filter methods which are applied tononlinear/non-gaussian system and has made great development and application inrecent years, but the algorithm itself is not very mature, and there are a series ofproblems in the application process, thus there have been a lot of improvedalgorithms in recent years.At first, this thesis introduces all kinds of improved algorithm in recent years,summarizes their improvement principles and points out that the improved algorithmalso exists some disadvantages. At the same time, new improved algorithms are putsforward on the basis of predecessors’ work. The first kind of improved algorithm ismainly based on the importance density function, through particle sampling fromimproving importance density function, which can make the sampling particles betterestimate the states of the target and improve the estimation precision. The secondkind of improved algorithm is mainly the improvement of the probabilitydistributions of likelihood function. Particles are chosen and given different weightsby selecting a more suitable probability distribution function, and the particles whichare close to the real value are given greater weight, otherwise less weight are given.Meanwhile, the diversity of particles is guaranteed, and the particles impoverishmentis inhibited to a certain extent. Simulation experiments show that the improved algorithm based on theimportance function can improve the tracking accuracy of the algorithm in a certainextent, but will relatively increase the computation time. The improved algorithmbased on likelihood function can not only improve the calculation precision, but alsosave the calculation time because the distribution of exponential decay choice is thefirst power, compared with the traditional gaussian distribution with the secondpower.