视频传感器网络中基于一致性的分布式目标跟踪研究

发布时间：2018-06-29 11:16

本文选题：视频传感器网络 + 一致性　；参考：《湘潭大学》2017年硕士论文

【摘要】：基于无线传感器网络的目标跟踪目前已应用于建筑物监控观测、动态定位与跟踪、城市管理、抢险救灾、国防等各个领域,是国内外研究的热点。在基于状态估计实现目标跟踪的方法中,分布式估计方法通常依赖于传感器节点之间的点对点通信,具有稳定性高、容错能力强、计算代价小等优点。在传感器网络中,传感器节点可以获得关于目标状态的多个测量值,分布式估计的目的是利用网络中的所有测量值来获得关于目标状态的精确估计,而并不需要中心节点作为信息融合中心。论文对基于视频传感器网络的分布式目标跟踪技术进行了深入研究,并提出了提高跟踪精度的方法。本文提出了基于一致性算法的信息加权一致性滤波(Information-weighted Consensus Filter,ICF)算法,并将其扩展应用到非线性观测模型。论文的主要工作如下:首先,大量查阅与研究专业相关的国内外文献,简单阐述了基于无线传感器网络的目标跟踪的研究意义及研究现状。对几种视频传感器网络中的目标跟踪方法的优缺点进行了概括。其次,分析了现有的两种分布式目标跟踪方法。本文首先简要描述了平均一致性算法的基本原理。然后阐述了目前应用较广的两种基于视频传感器网络的分布式目标跟踪方法,即卡尔曼一致性滤波(Kalman Consensus Filter,KCF)方法和广义卡尔曼一致性滤波(Generalized Kalman Consensus Filter,GKCF)方法。重点分析了目标状态估计实现过程,并从理论和仿真两方面进行了对比分析。然后,针对视频传感器节点具有有限视场,传感器网络中存在无法获得关于目标测量信息的朴素节点这一问题,提出了信息加权一致性滤波算法。首先介绍了视频传感器网络的特性,然后基于集中式最大后验估计推导出目标的集中式状态估计的信息形式,接着基于一致性算法推导出目标状态估计方程的分布式实现,最后结合目标状态动态模型,提出信息加权一致性滤波方法。本文还从理论上对比分析了ICF算法相较其他算法的优势。仿真表明,相较传统的KCF算法和GKCF算法,ICF算法的跟踪精度更高且趋于集中式目标跟踪算法。最后,提出了基于非线性观测模型的扩展信息加权一致性滤波(Extended Information-weighted Consensus Filter,EICF)算法。首先描述了视频传感器网络的非线性观测模型,然后根据扩展卡尔曼滤波算法和一致性算法,得到目标状态估计和信息矩阵在非线性模型下的分布式形式,最后结合目标状态动态模型提出了EICF方法。仿真结果验证了该算法的有效性。
[Abstract]:The target tracking based on wireless sensor network has been applied to many fields, such as building monitoring and observation, dynamic location and tracking, urban management, rescue and disaster relief, national defense and so on. In the method of target tracking based on state estimation, distributed estimation usually depends on point-to-point communication between sensor nodes, which has the advantages of high stability, strong fault tolerance and low computational cost. In sensor networks, sensor nodes can obtain multiple measurements of the target state. The purpose of distributed estimation is to obtain accurate estimation of the target state using all the measurements in the network. The central node is not needed as the information fusion center. In this paper, the distributed target tracking technology based on video sensor network (VSNs) is studied, and a method to improve the tracking accuracy is proposed. In this paper, an information weighted consensus filter (ICF) algorithm based on consistency algorithm is proposed and extended to nonlinear observation model. The main work of this paper is as follows: firstly, the research significance and research status of target tracking based on wireless sensor networks are briefly described by consulting a large number of domestic and foreign literatures related to the research major. The advantages and disadvantages of several target tracking methods in video sensor networks are summarized. Secondly, two existing distributed target tracking methods are analyzed. In this paper, the basic principle of average consistency algorithm is briefly described. Then, two widely used distributed target tracking methods based on video sensor networks, Kalman consensus filter (KCF) and Generalized Kalman consensus filter (GKCF), are described. The realization process of target state estimation is analyzed, and the theoretical and simulation results are compared. Then, aiming at the problem that video sensor nodes have limited field of view and the naive nodes in sensor networks can not obtain information about target measurement, a weighted consistency filtering algorithm is proposed. This paper first introduces the characteristics of video sensor networks, then derives the information form of the centralized state estimation based on the centralized maximum a posteriori estimation, and then deduces the distributed realization of the target state estimation equation based on the consistency algorithm. Finally, combining the target state dynamic model, the information weighted consistency filtering method is proposed. The advantages of ICF algorithm compared with other algorithms are also analyzed theoretically. Simulation results show that the tracking accuracy of ICF algorithm is higher than that of traditional KCF algorithm and GKCF algorithm. Finally, an extended Information-weighted consensus filter (EICF) algorithm based on nonlinear observation model is proposed. Firstly, the nonlinear observation model of video sensor network is described, and then the distributed form of target state estimation and information matrix under nonlinear model is obtained according to extended Kalman filter algorithm and consistency algorithm. Finally, an EICF method is proposed based on the target state dynamic model. Simulation results verify the effectiveness of the algorithm.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212

【参考文献】