【作者】 陈勇；

【导师】 周越；

【作者基本信息】 上海交通大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 智能视频监控是利用计算机视觉技术对摄像机采集的视频数据进行分析、理解,并以此为基础对视频监控系统进行控制。由于单个摄像机都视野域有限,监控大范围场景时需要使用多摄像机系统。智能视频跟踪系统中多个摄像机的使用有利于解决遮挡问题,场景混乱、环境光照突变情况下的运动目标跟踪等问题。而且基于多摄像机的视频跟踪在高级人机交互、视频会议、医疗诊断以及基于内容的存储与检索等方面有广泛的应用前景和潜在的经济价值。本文视频监控为应用背景,主要研究了基于多摄像机的目标跟踪方法。研究内容主要有运动检测、单摄像机下的目标自适应跟踪,多摄像机间的目标映射以及多摄像机协同等方面的内容。主要工作如下:1、详细分析研究了多摄像机跟踪中的关键技术和相关算法。对当前主要的技术难点和算法做了全面的总结、分析。研究了视频前景运动目标提取原理和算法以及视频前景中目标阴影的检测与去除算法。2、为了解决目标在跟踪过程中的跟踪窗自适应变化的问题,我们提出了一种监控场景下目标自适应跟踪算法,本算法充分利用基于Mean-Shift跟踪算法和基于目标检测的跟踪算法的优点。将运动目标的检测信息,卡尔曼滤波预测性能和Mean-Shift快速收敛性跟踪性能有机结合,稳健地解决自适应跟踪和特征更新问题,使跟踪过程更具有鲁棒性。3、研究了不同摄像机下目标的匹配映射和摄像机协同方法,提出了基于目标大小和位置的多摄像机匹配和协同算法。针对多摄像机搭建中存在的小视野重叠区的情况,提出了使用临时中继摄像机的坐标映射计算方法。并提出了在小视野重叠区情况下多摄像机的协同策略。实现目标在摄像机间的准确匹配和连续跟踪。更多还原

【Abstract】 Intelligent video surveillance uses computer vision technology to analyze and understand the camera video data collected, and use it as the basis of video surveillance system. Because of the limited view of the domain, it needs to use multi-camera system to control large-scale scenes. The use of multiple cameras in intelligent video tracking system are conducive to resolving the shelter problem, chaotic scene, the environment in case of light mutation moving target tracking and other issues. And video tracking based on multi-camera have a broad application prospects and potential economic value at the senior human-computer interaction, video conferencing, medical diagnostics, as well as content-based storage and retrieval, etc.Based on the background for the use of video surveillance, this paper mainly studies multi-camera target tracking methods. The research includes motion detection, adaptive tracking under single-camera, goal mapping and collaboration between cameras and so on. The main work is as follows:1. Researching and analyzing the key technologies and related algorithms of multi-camera tracking in detail, making a comprehensive analysis and summary on the current major technical problems and algorithms. Researching on the moving video object extraction algorithm, and shadow detection and removal algorithm.2. To address the problem on adaptive changes of the tracking window, we propose an adaptive tracking algorithm in monitoring scene. This algorithm takes full advantage of the advantages of tracking algorithm based on the Mean-Shift algorithm and based on target detection algorithm. Using moving target detection information, Kalman filter forecast performance and Mean-Shift fast convergence performance, the algorithms provide a robust solution to adaptive tracking and feature updates.3. Researching the object mapping and cooperating method in different cameras, a cooperating method is proposed based on the target size and location. For the case that exist small field of vision overlap areas in multi-camera structures, a calculation method of coordinate mapping is proposed using temporary relay camera. And put forward the synergy strategy in multi-camera circumstances with small vision overlap, realizing the exact match between the camera and continuous tracking.更多还原