Networking Decentralized Surveillance System based on Cellular Vision Dynamics

网络化传感器技术具有解决一些最重要的科学和社会问题的潜在能力。网络化传感器可以获取海量信息，特别是在视觉系统中。但它还有另一个方面：如果我们建立大规模的网络化传感系统，我们将面临严重的问题，即如何处理如此海量的信息，如何检索我们所需的情报。即使分布式数据处理可以减轻计算任务和网络流量，中央处理器也可能很难从分散处理获得的信息中重建和分析整个空间的信息。针对这一问题，我们提出了一种用于运动检测的网络视觉传感器信息处理的新概念。这有两个方面。1)监控：中央处理器不直接处理本地数据；它从本地代理获取的元数据中检索必要的情报。2)改变信息类别：中央处理器不直接处理图像数据；它从不同类型的物理值，即视觉细胞的运动中提取必要的智能。也就是说，原始数据被转换为不同类型的物理价值。为了便于理解我们提出的概念，我们利用VRML(虚拟现实建模语言)工具箱，使用MatLab和Simulink制作了一个人群运动识别模拟器。此外，通过以监控为重点的基本实验，证实了本文提出的概念可以用于运动检测。

The networked sensor technology has a potential capability to solve some of our most important scientific and societal problems. The networked sensors can acquire huge amount of information, especially in case of vision systems. But it has another aspect; if we build large-scale networked sensing system, we face to the serious problems, i.e., how we can handle such huge amount of information and how we can retrieve our necessary intelligence. Even if distributed data processing may alleviate the computational tasks and network traffics ; it might be very difficult for the central processor to rebuild and analyze the information of the whole space from the information gotten by decentralized processing. For this problem, we propose a new concept for information processing of networked vision sensors for motion detection. There are two aspects. 1) Supervisory Monitoring : The central processor does not treat local data directly; it retrieves necessary intelligence from Meta data acquired by local agents. 2) Changing Information Category : The central processor does not treat image data directly ; it retrieves necessary intelligence from different kind of physical value, i.e., motions of vision cells. Namely, the original data is transformed into a different type of physical value. In order to facilitate the understanding of our proposed concept, we made a simulator for recognizing movements of a crowd of people by using Matlab and Simulink with VRML(Virtual Reality Modeling Language) Toolbox. Moreover, we confirmed that the proposed concept can be utilized for motion detection by basic experiments focused on the supervisory monitoring.