目标跟踪中非线性滤波算法的研究

Research on Several Nonlinear Filtering Algorithms in Target Tracking

介绍了3种非线性估计方法。在处理目标跟踪等动态系统实时估计问题中,EKF将系统进行线性化近似时存在估计误差,从而影响目标跟踪的精度;PF对系统噪声和量测噪声的概率分布没有要求;RPF是改进的粒子滤波算法。分析了EKF、PF和RPF算法的原理,比较了3种算法的性能差异。仿真结果表明,PF滤波精度优于EKF,而RPF在精度和计算复杂度等方面均优于PF,且随着粒子数目的增加,PF和RPF的精度也不断提升。

Three nonlinear estimation methods are introduced,which are Extended Kalman Filter（EKF）,Particle Filter（PF） and Regularized Particle Filter（RPF）.We present the general principle of these algorithms and compare the differences among them.When EKF is applied to dealing with real-time estimation of dynamic system,such as target tracking,here is estimation error due to the defects of EKF in nonlinear estimation,which affects the accuracy of target tracking.PF does not require the probability distribution on the system noise and measurement noise.RPF is an improved particle filter algorithm.Experiment results show that the estimation precision of PF is better than that of EKF,but they are both poorer than RPF on estimation precision and calculation capacity.And with the increase of the number of particles,the estimation precision of PF and RPF will become better.