分布式传感器多目标跟踪改进算法

利用分布式传感器网络进行目标跟踪，能够有效增加传感器的覆盖范围，提高对运动目标的检测和跟踪能力，但如何充分利用相邻传感器之间的信息进行有效的融合，仍然是一个难点问题。本文在多伯努利滤波框架下，提出了一种改进的分布式融合跟踪算法用于目标数未知且变化的多目标跟踪。提出算法包含三种精度提升策略，即特征级融合反馈、决策级融合输出及交互反馈；其中，决策级融合输出策略可以提取更加准确的估计状态，特征级融合反馈策略可以降低错误融合结果对后续滤波过程的不良影响，交互反馈策略可以避免单传感器因漏检而导致的滤波失败。实验结果表明，提出算法的跟踪精度明显要优于传统的基于GCI分布式融合算法以及粒子多伯努利跟踪算法，具有较好的跟踪性能。 

An improved algorithm of distributed multi-sensor multi-target tracking

Distributed multi-sensor (DMS) network can effectively increase the coverage of the sensors and improve the ability of detection and tracking for moving targets. However, the Generalized Covariance Intersection (GCI) based fusion algorithm is suffer from the problem that the tracking performance will be deteriorated under complex environment. In this paper, we proposed an improved distributed fusion algorithm under the multi-Bernoulli (MB) filter framework for improving the tracking performance under complex environment. First, a decision-level fusion strategy is proposed to extract more accurate estimation states, and then a feature-level fusion feedback strategy is proposed to reduce the negative influences of the inaccurate fusion results for the subsequent filtering process. Moreover, an interactive feedback strategy is proposed to avoid the miss tracking of each single sensor. The experimental results show that the proposed algorithm has a better tracking accuracy than the traditional GCI-based distributed fusion algorithm and the traditional particle filter MB (PF-MB) tracking algorithm, with a good multi-target tracking ability in complex environments.