基于MCMC无味粒子滤波的目标跟踪算法

Target tracking algorithm based on MCMC unscented particle filter

针对传统粒子滤波目标跟踪算法存在粒子退化的问题,提出了基于马尔可夫链-蒙特卡罗(Markovchain Monte Carlo,MCMC)无味粒子滤波的目标跟踪算法。该算法采用无味卡尔曼滤波(unscented Kalmanfilter,UKF)生成粒子滤波的提议分布,来代替传统粒子滤波算法采用状态转移先验概率作为粒子滤波的提议分布,以改善滤波效果,然后在无味粒子滤波的基础上融合了典型的MCMC抽样算法(Metropolis Hastings,MH),从而可以减少传统粒子滤波未考虑当前量测对状态的估计作用所带来的影响。融合后的算法将当前量测信息融入到滤波过程中,并使采样粒子更加多样化。实验结果表明,该算法较传统方法在跟踪精度方面有显著的提高。

As the problem of particles degradation exists in the traditional particle filter algorithm, a target tracking algorithm based on the Markov chain Monte Carlo （MCMC） unscented particle filter is proposed. Instead of taking a transition prior probability as proposal distribution, the unscented Kalman filter （UKF） is used to generate the proposal distribution so as to improve the filtering effect. Then the paper syncretizes the standard MCMC sampling method, Metropolis Hastings （MH）, and the unscented particle filter, which can reduce the effect that the traditional particle filter doesn＇t consider the current measurement. The syncretized algorithm takes the current measurement into the filtering process and makes the particles more diversification. Experiment results show that the algorithm has more significant advantages in tracking accuracy than other tradi- tional algorithms.