《视频监控系统中运动检测与跟踪技术的研究》由会员分享，可在线阅读，更多相关《视频监控系统中运动检测与跟踪技术的研究（70页珍藏版）》请在金锄头文库上搜索。
1、上海交通大学 硕士学位论文 视频监控系统中运动检测与跟踪技术的研究 姓名：朱碧婷 申请学位级别：硕士 专业：通信及信息系统 指导教师：郑世宝 20090101 上海交通大学工学硕士学位论文 摘要 - III - 视频监控系统中运动检测与跟踪技术的研究 摘 视频监控系统中运动检测与跟踪技术的研究 摘 要要 随着人工智能和计算机技术的发展，机器视觉技术研究取得了长足 的进步，并引起了越来越多的关注。目前，根据机器视觉完成的运动检 测、识别、跟踪广泛地应用于国防建设、航空航海、医药卫生、安全监 控等国民经济的各个领域。运动检测与跟踪则是其中一个重要的研究方 向。运动检测的目的是检测某监控区域内是否出
2、现目标物体并确定目标 出现的位置、大小等信息；目标跟踪的目的是对选中的目标进行轨迹的 描述，涉及到目标特征的提取、特征描述、目标匹配等多个方面。 本文研究的主要目标是仿真实现对视频序列中人体运动目标的实时 监控。系统主要由检测与跟踪两个模块构成，文中分别对检测和跟踪的 算法进行了分析和研究，并提出相应的改进依据和改进方法，最后将两 者结合，实现了在静止背景与微弱动态背景下对人体目标的全自动准实 时的跟踪。 本文首先提出了一种适合静止背景与微弱动态背景视频序列的运动 目标检测方法。该方法首先对象素做混合高斯建模(Gaussian Mixture Model，简称 GMM)，在判断像素是背景还是前
3、景点时，传统 GMM 采用 了参考相同位置上像素样本的期望和方差参数的策略。然而像素邻域间 的信息对像素的分类至关重要。基于此，在判决过程中，提出对前景点 集做空间域二次判决的策略。空间域二次判决在一次判决的基础上，利 用邻域像素的历史统计值信息进行重判决，可筛选出“伪前景点”,解决 混合高斯模型分离法无法正确检测背景物体棱角、边缘处像素的问题。 上海交通大学工学硕士学位论文 摘要 - IV - 运动检测的另一个难点是阴影消除。本文结合背景像素亮度特性，对前 景像素的归一化亮度进行空间域二次判决，同样取得良好的去除阴影效 果。最后，经过对 GMM 背景分离法的多次试验分析，对部分参数进行 了优
4、化：将建模空间从一维推广到三维，并重新定义了方差收敛函数。 实验证明基于空间域二次判决策略的 GMM 分离法可降低背景像素的误 检率和有效抑制运动物体的阴影。 本文的跟踪算法主要基于 Mean-Shift 算法和粒子滤波算法。在 Mean-Shift 理论的分析基础上， 首先归纳了 Mean-Shift 算法在跟踪应用上 的具体步骤。其次，从图像处理学角度上定义了目标可见度，并给出相 应的数学计算式。通过目标可见度和 Mean-Shift 核函数带宽作用的研究， 阐述了在目标遭遇遮挡和动态复杂背景情况下跟踪效果不佳的原因。针 对目标大面积遮挡的情况，本文提出多项式拟合运动曲线的方法来估算 目标
5、处于遮挡状态时的位置；针对动态复杂背景，本文结合粒子滤波和 Mean-shift 各自的优点，设计了基于粒子滤波器的 Mean-shift 跟踪算法。 跟踪算法选举图像核直方图作为目标模板的特征描述，在跟踪过程中， 能自适应调整标准模板，并可自动调节跟踪区域的大小。 本文最后在 PC 平台上实现了一个基于 EasyVideo 数字视频处理软件 开发包的运动物体检测跟踪系统。实验结果表明，该系统对运动物体检 测效果良好，能有效抑制阴影，并能在目标处于遮挡的情况下和动态背 景下正确跟踪目标。 关键字：混合高斯模型，Mean-Shift,粒子滤波，阴影消除，遮挡处理 上海交通大学工学硕士学位论文 摘
6、要 - V - Research of Moving Detection and Tracking Algorithm for Intelligent Video Surveillance ABSTRACT With the development of artificial intelligence and computer science, Machine Vision Technology has made considerable progress, and attracted more and more attention. At present, objects detection
7、, identification and tracking according to machine vision technology, have been widely used in various economic aspects, such as national defense, aviation and navigation, medical treatment and health care, security monitoring, and so on. Face detection and tracking is one of the most important appl
8、ications. The purpose of face detection is to detect if there are the faces occurred in the surveillance area. If that, then determine the information of location and size of the faces. Object tracking is also an important application branch of Computer Vision. The goal of tracking is to recognize t
9、arget objects from the background and extract features, then to depict the moving locus of them. How to track the objects precisely and fast is the key problem of objects tracking system. In this paper, the main objective is to design a real-time monitoring system for the object interested. This sur
10、veillance system mainly combines two sub-systems, moving detection and object tracking. By analyzing the algorithm of detecting and tracking, this paper brings out reasons and methods to promote the performance. In this paper, we propose a moving detection algorithm applied for static background vid
11、eo sequences, which can solve common problems in moving detection. We firstly establish Gaussian Mixture Model (GMM) for each pixel 上海交通大学工学硕士学位论文 摘要 - VI - and then extend the classical GMM detection from the pixel-time domain into pixel-spatial domain. For the spatial domain detection, it can re-d
12、etect the pixel which selected out as the foreground elements using the classical method. By this way, we can pick out those pixels falsely detected at first time. Besides, our method can applied for the shadow detection, as the bright feature are unique for the shadow pixels, we can combine this fe
13、ature with our spatial-domain method and substitute unitary bright for pixels gray value as the input parameter of our method. Experiments show our method can improve the detection acutance and remove objects shadows. As for object tracking, Mean-shift and Particle filter are mainly used in this pap
14、er. We analyze the theoretical shortcoming of Mean-shift in order to explain the poor object tracking when object encounters bad occlusion and in dynamic and complex background. Besides, we propose the polynomial imitation to approach objects movement trochoid and predict the location in the occlusi
15、on. For the dynamic and complex background, our method combines the advantages of both Particle Filter and Mean-shift in which the particle filter used for predicting the location with most probability and the Mean-shift algorithm converges to the final position of object. The research was carried o
16、ut on PC platform with the software development kits developed by Intel named EasyVideo, which is used for digital video and image processing. The result shows that this system is fast and precise, and basically meets the requirement of real-time surveillance. KEYWORDS: Gaussian Mixture Model, Mean-shift, Particle Filter, Shadow Removal, Occlusion Processing 上海交通大学工学硕士学位论文 - 64 - 上海交通大学工学硕士学位论文 - 65 - 上海交通大学工学硕士学位论文 第一章 绪论 - 1 - 第第 1 章章 绪论绪论 1.1 研究背景及意义 1.1 研究背景及意义 近年
