【作者】 陶丹；

【导师】 马华东；

【作者基本信息】 北京邮电大学 ， 计算机科学与技术， 2007， 博士

【摘要】 覆盖控制和协作处理是以视频传感器网络为基础的目标监测应用的两个关键问题和研究的热点。在传统传感器网络中，覆盖控制和协作处理技术已经积累了较多的研究成果。近年来，随着新型视频传感器网络的出现，其节点方向性感知和媒体大数据量等显著特点，现有方法不能有效适用。这就迫切需要我们设计出一系列新的覆盖控制和协作处理方法。本论文围绕着视频传感器网络中节点感知模型、覆盖控制及协作处理等关键问题开展深入研究，侧重点于视频传感器网络的区域覆盖增强、路径覆盖增强、层次性协作模型以及基于视觉相关性的协作图像处理等几个方面。针对上述问题，提出了相应的新模型和算法，并给出了分析和仿真。本论文的主要创新点如下：(1)在现有的有向感知模型基础上，设计一种新型的方向可调感知模型，以有效刻画视频传感器节点的感知能力；并抽象了视频传感器网络协作信息处理的层次协作模型。(2)研究利用位置固定但传感方向可调的视频传感器节点实现传感器网络覆盖增强问题。基于图论和计算几何方法，提出一种集中式的区域覆盖增强算法。该算法采用“分而治之”思想，将对整个监测区域覆盖增强问题划分为若干个子区域覆盖增强问题，大大降低算法计算复杂度，加速网络区域覆盖增强进程。另外，区域覆盖增强方案一经汇聚节点确定，网络中所有视频传感器节点并发地进行传感方向的一次性重置。仿真实验表明：该算法能以较小的代价获得整个视频传感器网络区域覆盖性能的改善。(3)基于虚拟势场方法，提出一种分布式的视频传感器区域覆盖增强算法。通过引入“质心”的概念，将待解决问题转化为相应质心点均匀分布问题，以质心点围绕节点作圆周运动来代替视频传感器节点传感方向的转动。众多质心点在虚拟力作用下作扩散运动，逐步消除视频传感器网络中感知重叠区和盲区。仿真实验表明：该算法能显著地增强视频传感器网络区域覆盖性能，提高网络对目标的探知能力。(4)针对视频传感器网络中目标跟踪问题，提出一种基于虚拟势场方法的路径覆盖增强算法。通过引入“质心”和“运动轨迹点”的概念，将待解决问题转化为质心点-运动轨迹点，质心点-质心点间虚拟力作用问题。该算法在保证路径最大化覆盖的同时，也兼顾相邻视频传感器节点间覆盖重叠区域尽可能小，进而实现路径的充分高效覆盖。仿真实验表明：该算法可以显著提高对目标路径的覆盖性能，改善网络的目标跟踪质量。(5)基于视觉相关性，研究适用于视频传感器网络目标监测应用的协作处理方法。考虑到视频传感器节点的有限资源和相邻节点间冗余视觉信息，采用协作思想将同一场景的视觉监测任务分配到相关度较大的多个视频传感器节点上。特别地，基于立体视觉中极线约束性质，提出了一种协作图像处理方法，在簇头节点上实现对多源图像的融合处理，最终实现所监测目标视觉信息的重建。仿真实验证明：该方法简单易行，既可大大减少网络节点传输量，节约网络能量，又可实现目标视觉信息的有效监测。更多还原

【Abstract】 Coverage control and copperative processing are two kernels and hot topics in the research field of the target detecting application based on video sensor networks. The two issues in conventional sensor networks have been studied and analyzed intensively. Recently, with the appearance of novel video sensor networks, their distinct characteristics with directional sensing ability and a large amount of image/video data, make that many methods for conventional sensor networks are not suitable for video sensor networks. Thus, video sensor networks demand a series of innovative solutions, especially for coverage control and copperative processing. The thesis studies some fundamental issues in video sensor networks, such as sensing model, coverage control and copperative processing, especially focuses on area coverage enhancement, path coverage enhancement, layered copperation model and cooperative image processing based on visual correlation. Aiming at the problems mentioned above, we propose a series of new models and algorithms, carry out performance evaluation and simulation analysis. The main contributions of this thesis are as follows:(1) Based on the existing directional sensing model, we design a novel rotatable directional sensing model to describe the sensing ability of video sensors, moreover, we propose a layered cooperation model to abstract cooperative information processing for video sensor networks.(2) We study the issue of area coverage enhancement in video sensor networks with the assumption that video sensors have fixed locations and adjustable sensing directions. We present a centralized area coverage-enhancing algorithm by using graph theroy and computational geometry. We divide and conquer a video sensor network into several sub-areas to decrease the time conplexity of algorithm and quicken the area coverage-enhancing process. In addition, once the area covrage-enhancing scheme is calculated by the sink node, all of video sensors in the network will rotate their sensing directions at one time. Simulation results show that this centralized alogorithm can improve the area coverage performance of video sensor networks with small cost.(3) We present a distributed potential field based coverage enhancement algorithm for video sensor networks. By introduing the concept of "centroid", we translate the pending problem into centroid uniform distribution problem by moving centroid points around the video node instead of rotating sensing directions. Multiple centroid points repel each other to eliminate the sensing overlapping regions and coverage holes. Simulation results show that this distributed algorithm can significantly improve area coverage performance and enhance the target detecting ability.(4) For object tracking application using video sensor networks, we propose a potential field based path coverage-enhancing algorithm. By introducing the concepts of "centroid" and "trackpoint", we construct virtual potential field to study on the force laws to govern the interaction between centroid-centroid and centroid-trackpoint. This algorithm maximizes path coverage along with target trajectory while minimizing the overlapping sensing area among multiple neighboring video sensors thus achieve adequate and efficient path coverage. Extensive simulation results show that this algorithm can effectively enhance path coverage performance with small cost, thus improve the object tracking qaulity provided by video sensor networks.(5) Based on visual correlations, we study a cooperative information processing method for video sensor networks. Given the severe resource constraints on individual video sensors and the redundant visual information among video sensors, we adopt a divide-and-conquer method to partition a sensing task among highly correlated ones. In particular, we propose a cooperative image proceesing method according to the epipolar line constraint. Sink node fuses the received multiple partial images, thus reconstruct a complete visual scene. Experimental results show that our method is more efficient than the non-cooperative ones in reducing transmission workload, saving network energy and performing visual monitoring task.更多还原