【作者】 周昌；

【导师】 陈耀武；

【作者基本信息】 浙江大学 ， 测试计量技术及仪器， 2007， 博士

【摘要】 随着大规模安防系统在各种公共场所中的应用,通过智能视频监控系统实现预防恐怖袭击和公共治安等突发事件的需求日益增长。目前基于中心处理的智能视频监控系统由于计算能力和通讯带宽等因素限制,无法应用在大规模视频监控应用中,实施分布式智能视频监控系统是实现大规模智能视频监控应用的基础。具有场景状态感知能力的嵌入式智能摄像机网络是分布式智能视频监控系统的重要组成部分,研究嵌入式智能摄像机网络的相关问题是实施分布式智能视频监控的关键和核心问题。针对嵌入式智能摄像机网络这一研究课题,本文对以下若干关键问题进行了相关的研究。1.高性能的嵌入式智能摄像机设计本文在分析嵌入式智能摄像机特点的基础上,提出一种新型的高性能嵌入式智能摄像机设计方案。该方案采用双核DSP处理器作为硬件系统核心,并设计了一个适应复杂智能视频监控任务的软件框架。2.面向嵌入式智能摄像机的视觉分析算法设计与研究本文针对嵌入式智能摄像机这一特定的计算环境,对在嵌入式智能摄像机上如何实现智能视觉分析进行相应的研究,包括:运动目标检测、阴影消除、运动目标分类以及运动目标跟踪。提出了一种基于局部图像描述的运动目标跟踪算法,该方法利用图像关键点邻域的局部图像描述作为目标定位的搜索空间,在这个空间内进行全局匹配,实现跟踪目标定位。该方法能够很好的适应跟踪目标尺度和外观的变化。3.嵌入式智能摄像机网络组织架构研究本文提出了一种摄像机间快速重叠场景探测算法,以及一种以马尔可夫随机场模型为基础的应用信念传播算法的相关摄像机组划分方法。在这两个算法的基础上,设计一个以相关摄像机组为核心的嵌入式智能摄像机网络组织架构,包括相关摄像机组的状态机、通讯协议以及组织流程。更多还原

【Abstract】 As video surveillance system is proliferating worldwide, the application through intelligent video surveillance (IVS) system to achieve interrupting or preventing acts of crime or terrorism is becoming more and more important. Because of the limitation of processing and communication, the center-based IVS cannot adapt to the large-scale applications. Implementation of the distributed intelligent video surveillance (DIVS) is a solution to larger-scale video surveillance application. The embedded smart cameras network with the ability to provide an automatic interpretation of scenes is the primary component of DIVS. The research of embedded smart camera network is critical in implementing DIVS system. The thesis is organized as follows:1. The design of high performance embedded smart cameraA novel high performance embedded smart camera is proposed in this thesis. The design of the embedded smart camera is accomplished based on a dual-core DSP processor and a software framework for multi IVS task is also discussed.2. The visual analysis algorithms based on embedded smart cameraUnder a particular computing environment, the implementation of several visual analysis algorithms with the embedded smart camera is discussed. It includes moving object detection, shadow removal and moving object classification. A novel moving object tracking algorithm based on local image descriptor is presented. The tracking task is accomplished by locating the target in the search space of the local image descriptor which is created by the key points of image. The method is stable even there are scale and appearance changes to the tracked target.3. The organization architecture of the embedded smart networkIn the thesis the issues about camera group are investigated thoroughly. A fast view matching based method is proposed to detect the overlapped areas between cameras. A camera grouping approach based on Markov random filled is also proposed, the grouping is accomplished by the belief propagation algorithm. Combing these two algorithms, a camera-group based organization-architecture of embedded smart network including a state machine and camera communicating protocol is present in the thesis.更多还原