双摄像机智能视频监控系统设计与实现

【作者】 王一鸣；

【导师】 茅耀斌； 项文波；

【作者基本信息】 南京理工大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 视频监控系统以其直观、方便、信息内容丰富的特点已成为各行各业安全防范系统的重要组成部分。本文针对目前视频监控系统对智能化、可靠性越来越高的要求,设计并构建了由一个全景摄像机和一个跟踪摄像机构成的双摄像机智能视频监控系统,并主要针对视频监控中常见的摄像机配准技术和运动目标跟踪技术两个方面进行了研究。在摄像机配准方面,本文采用了基于图像匹配的方法。由于双摄像机观测到的图像之间存在着平移、旋转、尺度等变化对图像配准造成的困难,本文提出了使用基于尺度不变特征的SIFT(Scale Invariant Feature Transform)算法来进行摄像机配准的方法。在详细研究了算法的原理及实现过程后,通过实验寻找到匹配过程中的合适参数设置并验证了其在双摄像机配准上的可行性。在视频目标跟踪方面,针对目标在场景中易被遮挡、丢失,监控系统的实时性、动背景等常见的问题,在详细研究了帧差混合算法以及基于颜色直方图的粒子滤波算法的原理及实现过程后,本文将帧差混合算法和粒子滤波算法结合使用,提出了一种稳健的视频目标跟踪方法并通过了实验的验证。最后本文使用了VC++、OpenCV库和DirectShow的相关技术并结合软件工程设计中常用的白盒测试和黑盒测试法,完成了整个双摄像机视频监控系统平台的构建。更多还原

【Abstract】 Video surveillance has become an important component in the Security fields because of its distinctive feature such as visualization, convenience and rich information. In view of the increasing requirements of intelligence and reliability in the current video surveillance, this paper designs and constructs a dual-camera intelligent video surveillance system which composed by a full-view-camera and a tracking-camera. A research has been made mainly focus on the two aspects of video surveillance, which are camera registration and moving-target tracking technology.In the aspect of camera registration, this paper uses an image matching method, because of the existence of the translation, rotation, scale and other changes, the thesis proposes a method of camera registration based on the use of Scale-Invariant Features Transform(SIFT) algorithm. After a detailed research in the principle and process of the algorithm, suitable parameter settings are found in the match process through the experiments and its feasibility on the dual-camera registration is verified.In the aspect of video tracking, in view of common problems, such as the target in the scene easily obscured, lost, the real-time demand of surveillance system, moving background, after a detailed research in the conterminous frames differencing algorithm and an adaptive color-based Particle Filter tracking algorithm, the thesis proposes a method by combining the two algorithms. The simulation results show that the method is rather robust.Finally, a framework of video surveillance system has been developed on the platform of Visual C++, with OpenCV library and DirectShow technology. In the process of constructing system, White-Box testing and Black-Box testing were used to debug the program.更多还原