改进的交互式多模型粒子滤波跟踪算法

通常的交互多模型卡尔曼滤波(IMMKF)或交互多模型扩展卡尔曼滤波(IMMEKF)对于非高斯问题无能为力;对于非线性问题,其性能不及交互多模型粒子滤波算法(IMMPF).粒子滤波能够处理非线性/非高斯问题,其与交互式多模型结合用来获得更好的跟踪性能.然而,粒子滤波的主要问题是巨大的计算量,由于粒子滤波通常采用大量的粒子数目,将带来很大的计算负荷.该文提出了一种改进的交互多模型粒子滤波算法,其利用多模型综合使用了卡尔曼滤波和粒子滤波,与常规交互式多模型粒子滤波(IMMPF)相比,大大改善了计算效率.对于非线性/非高斯问题,其性能与IMMPF相当;对于线性问题,其性能与IMMEKF相当,并优于IMMPF的性能.

Tracking Algorithms Based on Improved Interacting Multiple Model Particle Filter

The general Interacting Multiple Model(IMM) based on Kalman Filter(IMMKF) or Extended Kalman Filter(IMMEKF) can not deal with non-Guassian problems and also does not work so well as the IMM based on the particle filter for the nonlinear problems.The particle filter can deal with nonlinear/non-Guassian problems and it has been introduced to the algorithm of IMM for higher precision.However,the disadvantage of the particle filter is heavy computational load,because a particle filter usually has a lot of particles,which will increase the computational load greatly.Here an improved interacting multiple model particle filter,which combines Kalman filter and particle filter using multiple models,is proposed to improve the computational efficiency compared with the usual Interacting Multiple Model Particle Filter(IMMPF).For nonlinear/non-Guassian problems,the new algorithm shows to possess a good performance as the IMMPF,while for linear problems it performs as well as the IMMEKF and works better than the IMMPF.