为在分布式目标跟踪中交换局部似然函数的信息,研究常见的分布式目标跟踪方法,提出一种基于信念传播的分布式粒子滤波方法(DPF-BP).在有限次的迭代中,计算图的最大直径.为避免网络评估的分歧性,在计算评估之前运用一致性最大化,将节点及迭代次数表示成函数形式,经过标准化和估值计算后重采样替换.仿真实验结果表明,与标准信念一致方法、随机流言方法和都市信念一致方法(MBC)相比,在相同配置下,DPF-BP方法的均方根误差指标较优,在环形网络中运用DPF-MBC方法较好,而在树状网络中运用DPF-BP方法最佳.%In order to exchange information of partial likelihood function in a distributed target tracking,several common distributed target tracking methods are studied,and a Distributed Particle Filter method based on Belief Propagation(DPF-BP)is proposed.The maximum diameter of the graph is calculated in a limited number of iterations.In order to avoid difference in network assessment,consistency maximization is used before assessing and nodes and the number of iterations are expressed as a function.After standardization and valuation calculations,the replacement is re-sampled.Simulation experimental results show that,compared with Standard Belief Consensus(SBC),Randomized Gossip(RG)and Metropolis Belief Consensus(MBC),under the condition of the same configuration,DPF-BP is excellent at RootMean Square Error(RMSE).In addition,DPF-MBC is best in the circular network,and DPF-BP is best in the tree network.
展开▼
