【作者】 郭燕；

【导师】 熊焰； 华蓓；

【作者基本信息】 中国科学技术大学 ， 计算机应用， 2008， 博士

【摘要】 无线传感器网络作为一种全新的计算模式,利用众多分布在物理环境中具有感知、计算和通信能力的微小节点近距离地观察环境,改变了人类与环境的交互方式,在环境监测、工业过程控制、战场监视、灾后救援等场合有着广泛的应用。由于可被直接置于物理环境中实现对目标的近距离监视,无线传感器网络在目标跟踪应用中有着得天独厚的优势。逃跑者-追击者问题是目标跟踪应用的一种,该问题假设在监视区域中存在一组同类目标和一个或几个追击者,要求追击者在最短时间内捕获所有的目标。该问题包括四个子问题:(1)传感器网络通过数据采集与协同计算获得各局部区域的目标信息(个数及位置);(2)追击者查询传感器网络获得全局目标信息;(3)追击者利用全局目标信息修正目标位置和预测运动轨迹;(4)追击者利用目标位置和轨迹制定追击方案。本论文研究前三个问题。多目标计数和定位是无线传感器网络的一个难题,目前这方面的研究工作还不多。典型的方法是先根据节点检测到的信号强度对节点进行分簇,尽可能将检测到相同目标的节点划分到同一个簇中,然后每个簇各自计算其覆盖范围内的目标个数和位置。不同算法的差异主要体现在节点分簇、目标个数估计和位置估计上。目前尚没有很好的办法解决这个问题,一方面,使用简单的强度传感器或二元传感器在目标接近时计数和定位精度不高;另一方面,提高精度又需要复杂的算法或高代价的传感器设备。本论文针对目标信号影响范围有限、信号强度分布在空间连续渐变的特点,提出了一个基于能量的目标计数算法EBTN。首先根据节点检测到的信号强度对节点进行分簇,然后每个簇头使用一个多项式函数拟合簇内的信号分布并计算信号总能量,除以单个目标的信号能量即得到簇内目标个数。在得到目标个数后,簇头在簇内寻找与目标个数相同、具有局部最大值的节点作为目标的估计位置。该算法具有计算复杂度小、通信代价低、分布式实现的优点,当节点空间分辨率足够时,计数和定位目标的精度较高。信息存储和查询是无线传感器网络的一个研究热点,但是目前还没有针对逃跑者-追击者应用的信息分布研究。逃跑者-追击者应用的数据类型单一、数据量少,但对信息查询和获取的实时性要求很高。目前一些以数据为中心的信息存储与查询系统适合用来管理传感器网络中大量的、多种类型的数据,并可为用户提供灵活多样的查询方式,但对节点的软硬件要求较高,而且不能保证查询的实时性。本论文首次提出了在同构网络中面向追击者-逃跑者应用的信息发布与查询问题,设计并实现了一个基于网络单覆盖的目标信息分布算法SCUS。簇头利用该算法将局部目标信息随机均匀地发布到网络中的一组节点上,保证追击者在网络中的任何一个位置都可以在自己的通信范围内高概率地得到该目标信息的一份拷贝。SCUS算法在保证快速获取目标信息的同时还尽量均匀使用节点,以最大化网络的连通寿命。目标轨迹计算与预测是目标跟踪的一个重要内容,这方面已经有了很多成熟的工作。然而,在目标机动性强、数量较多、定位精度较低的情况下,传统方法往往需要使用复杂的模型,计算复杂度大,延迟长。在前两项工作的基础上,本论文针对追击者-逃跑者问题的特点和需要提出了由追击者执行的数据关联和目标状态预测算法DASE。追击者利用从传感器网络收到的目标信息(可能不准确)计算数据和已有目标轨迹的关联概率,将数据与目标进行关联。考虑到目标追击应用中目标的机动性较强,算法仅使用较近的位置信息进行分段拟合,对目标状态进行修正和预测。相比于传统跟踪应用中的数据关联和状态预测方法,DASE实时性强,计算简单高效,可以较好地处理目标定位中的虚警和漏检现象,能够满足目标追击的信息要求。以上三个算法均在MatLab仿真平台上进行了实现与性能测试。这三个算法的有机结合可以为逃跑者-追击者应用提供准确而及时的决策信息。更多还原

【Abstract】 As a brand new computation paradigm, wireless sensor networks (WSN) employ numerous small-sized sensor nodes which are capable of sensing, computing and communication and distributed in a large area of physical world, which offer an inspiring new way for interaction between human being and the environment. WSN can be applied into various scenarios such as environment monitoring, industry process control, battlefield surveillance, after-disaster rescue, etc.Since sensor nodes can be directly deployed in the physical environment and have a close watch over targets, WSN is especially beneficial to targets tracking problems. Evader-Pursuer problem is an application based on targets tracking, in which a certain number of pursuers aim to capture a set of targets within minimum time. This problem comprises the following four sub-problems: (1) a set of sensor nodes acquire local targets information such as their number and positions through detection and collaborative computation; (2) the pursuers query the network to obtain global targets information; (3) pursuers update and predict the trajectories of targets based on the global targets information; (4) based on the result of the above steps, pursuers choose the optimum scheme to finish the capturing task. The first three sub-problems are studied in this thesis.Enumeration and localization of multi-target is by no means an easy task in wireless sensor networks. And existing researches have not delved into this problem either. A typical solution is to form nodes into uncorrelated clusters around the targets according to their measurements. Each cluster covers one or more targets, self-organizes and counts. The numeration algorithms differ mainly in the methods to form clusters and the way to compute the number of targets. The dilemma is: on the one hand, it is hard to achieve high precision with only binary or intensity-aware nodes when targets crowd; on the other hand, both multi-modal sensor nodes and more complex algorithms will either heighten the construction price of WSN or impose heavy computation and communication burden on WSN. This thesis explores energy attenuation characteristics such as limited influence range, continuous and gradual energy intensity change, etc. and proposes an energy-based target numeration algorithm EBTN. ETBN firstly clusters sensor nodes into groups according to their measurements, which depends much on the distance between sensors and targets, and elects a center node as the leader of the cluster. Then it is the leader’s task to fit the energy distribution of its cluster into a polynomial function and integrate to get the total energy of the whole cluster. Dividing the total energy by the unit energy of each target, the cluster leader will obtain the number of targets in its cluster. Based on this information, leader selects a set of nodes with local maximum measurements and uses their positions as the positions of the targets.Information storage and query has already been intensively studied in wireless sensor networks, however, information dissemination for evader-pursuer problems has not yet been investigated. Evader-pursuer problems require timely report of single typed and sketchy target information, while existing distributed data storage methods and systems aim to offer a large amount and multi-typed data, and flexible query interfaces for users, without guaranteeing real time query. Therefore these methods are not suitable for data query in evader-pursuer problems. The thesis is the first to explore the dissemination problem for evader-pursuer problems, and designs a single coverage based target information dissemination algorithms SCUS. Based on the work of EBTN, each cluster distributes the local targets information to a set of selected storage nodes, ensuring that the pursuers can get the global targets information within its communication range regardless of their position. Besides the high query success rate, SCUS makes sure that the nodes are uniformly employed as storage nodes, therefore to maximize the lifetime of network.The computation and prediction of target trajectories is a significant part of target tracking problem. And many methods have been proposed and practiced. However, when targets are of high mobility, high density and low localization precision, traditional methods would employ complex models, resulting in computation burden and long delay. With the help of EBTN and SCUS, a third algorithm DASE is proposed for pursuers to compute the target trajectory and predict new position. DASE focuses on the specific requirement of evader-pursuer problems, and computes the data association probability of imprecise target location with the existing trajectories. Since evaders are always with high mobility, DASE uses only recent history information for a linear least square fitting. Compared with traditionally used data association and state predication methods, DASE distinguishes with its simple computation, real time manner and satisfactory performance. Besides, DASE could handle false alarms and detection failures in target localization, thus meet the requirement of evader-pursuer problem.The above three algorithms have all been simulated on Matlab, and they work together to offer accurate and timely information for decision-making in evader-pursuer application.更多还原