分布式多目标伯努利滤波器的网络共识技术

本文研究了基于网络共识的分布式多目标伯努利(Multi-Bernoulli, MB)滤波器的目标跟踪技术。网络共识技术是实现传感器网络中分布式计算的一个强大工具，但同时对传感器间公共信息被“重复计算”问题尤为敏感。为解决该问题，本文首先在基于广义协方差交集(Generalized Covariance Intersection, GCI)准则的分布式MB (GCI-MB) 滤波器的基础上，通过采用序贯信息交互-本地融合的策略，提出网络共识(Consensus)-GCI-MB融合算法，简称C-GCI-MB融合；然后，通过数学理论分析了C-GCI-MB融合可以有效的避免“重复计算”问题；最后给出了C-GCI-MB融合算法的混合高斯(Gaussian Mixture)实现方法，并通过典型场景仿真验证了该算法的有效性及性能优势。 

Consensus for Distributed Multi-Bernoulli Filter

This work investigates the consensus based distributed multi-object tracking with multi-Bernoulli (MB) filters. Consensus has emerged as a powerful tool for distributed computing, however, this technique suffers from the problem of the susceptibility to the double counting of common information. In this paper, based on our previous work, namely, distributed MB filter in the framework of Generalized Covariance Intersection (GCI), we exploit the consensus based distributed MB filter, referred to as C-GCI-MB fusion. Then we analyze how the C-GCI-MB fusion can avoid the double counting problem. Finally, we present the Gaussian mixture implementation of the proposed C-GCI-MB fusion. The performance advantages of the proposed fusion algorithm are demonstrated in a challenge multi-target tracking scenario.