无线传感器网络中目标跟踪的马尔可夫模型与预测方法

MRF Model and the Tracks Forecasting Method of Target Tracking in WSN

针对目标运动是一个包含许多不确定因素的非线性非高斯随机过程,提出基于马尔可夫随机场模型与粒子滤波的WSN分布式目标跟踪方法(MRF-PF)。把目标跟踪过程看作是一个马尔可夫过程,基于贝叶斯规则,建立目标状态分布函数,用粒子滤波估计目标状态,实现目标跟踪。实验结果:对于泊松白噪声,MRF-PF方法的跟踪均方根误差RMSE相比卡尔曼滤波(KF)和扩展卡尔曼滤波(EKF)方法分别降低52.6%、49.2%;对于方差σ2由0.3→3的高斯噪声,GM-PF方法的RMSE相比KF、EKF分别降低54.5%～77.2%和23.5%～54.2%。这表明MRF-PF方法在非线性非高斯噪声或高斯噪声变化较大时具有较好的抗噪能力及跟踪性能。

In view of the actual target motion is a non-linear non-Gaussian random process which contains many uncertain factors,a distributed target tracking method based on Markov Random Field（MRF）model and particle filter is proposed.First,the target tracking process is seen as a Markov Random process and the target state function is built based on Bayes rules.Then the target state is estimated by particle filter method and distributed target tracking is achieved.Experimental results show that the root mean square error（RMSE）based on MRF-PF method is reduced by 52.6 % and 49.2 % respectively compared with that of Kalman filter（KF）and extended Kalman filter（EKF）when noise is Poisson white noise.Similarly,the RMSE based on MRF-PF is reduced by 54.5 %~77.2 % and 23.5 %~54.2 % respectively when noise is Gaussian and its σ2 changes from 0.3 to 3.MRF-PF method shows better anti-noise ability and tracking performance compare with KF and EKF.