Distributed cooperative localization based on Gaussian message passing on factor graph in wireless networks
Distributed cooperative localization based on Gaussian message passing on factor graph in wireless networks
复制标题
DOI：
10.1007/s11432-014-5172-y
复制
发表时间：
2015-03
期刊：
影响因子：
--
通讯作者：
N. Wu;Bin Li;Hua Wang;C. Xing;Jingming Kuang
中科院分区：
文献类型：
--
作者：
N. Wu;Bin Li;Hua Wang;C. Xing;Jingming Kuang
在传统的定位，代理可能无法达到足够数量的锚点，以获得明确的位置，特别是在稀疏网络。协作定位是在这种恶劣环境下的一种很有前途的解决方案，它使智能体能够通过交换位置信息和执行距离测量来相互协作。提出了一种基于因子图的无线网络分布式协作定位方法。高斯参数消息用于表示因子图上传递的消息。然而，由于非线性观测模型，不能得到封闭形式的解决方案。为了解决这个问题，泰勒展开被用来近似消息更新中的非线性项，这导致高斯消息在因子图上传递。因此，仅需要发送高斯分布的两个参数，并且可以显著减少用于定位的通信开销。提出的两种算法的应用程序与准确和不准确的锚，分别进行评估，通过蒙特卡罗模拟和比较SPAWN和最大似然（ML）估计。实验结果表明，本文提出的算法可以以更低的计算复杂度获得与SPAWN非常接近的性能，并且显著优于ML估计器。
In conventional localization, agents may not be able to reach sufficient number of anchors to obtain unambiguous locations, especially in sparse networks. Cooperative localization is a promising solution in that harsh environment, which enables the agents to cooperate with each other by exchanging location information and performing range measurements. In this paper, a distributed cooperative localization method based on factor graph is proposed in wireless networks. Gaussian parametric messages are used to represent the messages passed on factor graph. However, because of the nonlinear observation model, no closed-form solutions can be obtained. To solve this problem, the Taylor expansion is used to approximate the nonlinear terms in message updating, which leads to the Gaussian message passing on factor graph. Accordingly, only two parameters of the Gaussian distribution have to be transmitted and the communication overhead for localization can be significantly reduced. The two proposed algorithms for the application with accurate and inaccurate anchors, respectively, are evaluated by Monte Carlo simulations and compared with the SPAWN and maximum likelihood (ML) estimator. The results show that the proposed algorithms can perform very close to SPAWN with much lower computational complexity, and it outperforms the ML estimator significantly.