Particle Filter And Its Application To Target Tracking

Particle filter can effectively solve the strong nonlinear and non-gaussian state valueestimation problems, which will be faced in the target tracking. The research of particle filteringand its application in target tracking is discussed in this paper. The concrete research content arereflected in the following aspects:1. Basic theory of target tracking is analyzed and discussed. Different state-space modelsare applied to different target systems and different filtering algorithms are applied to differentstate-space models. With respect to linear and nonlinear system, the basic principles of classicrecursive filter and standard particle filter are introduced and their characteristics and shortagesin practical application are analyzed respectively.2. A variable structure control plan which is based on the particle filter is designed andapplied to motor servo control system. If particle filter is applied in motor servo control systemto estimate the random signals, white noise interference and uncertainty interference will berestrained. As a result, system chattering problem which is caused by sliding mode variablestructure control can be solved. System’s tracking accuracy is improved and dynamicperformance of the system to movement is enhanced by the index reaching sliding modecontroller which is based of particle filter.3. An improved particle filter algorithm which has adjustment factor and adaptive factor isproposed. Estimation error will exist and the precision of the estimation will be declined whenthe statistical properties of system noise and measurement noise are inaccurate. In order to solvethis problem, the improved system noise variance adjustment method is introduced into theparticle filtering and then an improved particle filter algorithm which has adjustment factor andadaptive factor is obtained. The improved the system noise variance adjustment method ispromoted based on Lee’s theory which is combined with NUPF algorithm later in this paper.Estimation precision will be declined when the system noise statistical characteristics are notaccurate. Such problem can be effectively solved by the algorithm and the increased calculationand memory space is not too much. Furthermore, the real-time performance of the algorithm is satisfying.