摘 要
这些年随着人们对于生命,财产保护意识的日益增强,对于视频监控系统的要求也越来越高,视频监控系统的智能化得到了人们的广泛关注和大量的研究。运动目标识别与跟踪技术做为智能视频监控系统的关键技术也得到了快速的发展。ARM嵌入式技术作为目前研究最火热的IT技术之一,以其低成本、低功耗、高性能的特点,得到了广泛的应用。将视频监控系统与ARM嵌入式技术相结合是当前智能监控系发展的趋势。本文主要围绕基于ARM嵌入式硬件平台的运动目标识别算法和运动目标跟踪算法展开了研究,通过实验选取了适合ARM嵌入式平台的运动目标识别算法和运动目标跟踪算法,并最终在ARM嵌入式开发板ARMS3C2440上实现了对运动目标的识别与跟踪。在运动目标识别算法方面,本文主要研究了卡尔曼（Kalman）滤波算法、CAMShift等算法,并结合嵌入式的开发环境在Linux+Qt平台上进行了仿真实验。在利用帧差法对运动目标进行识别的实验中对帧差法中的关键参数二值化阈值做了三组对比实验,通过分析实验效果,得出了相关结论。在利用背景建模法对运动目标进行识别的实验中对背景建模法的关键参数学习率做了三组对比实验,通过分析实验结果,得出了相关结论。在利用Surendra自动背景更新法对运动目标进行识别的实验中,通过实验验证了该算法对运动目标识别的有效性,通过分析实验结果,得出了相关结论。在运动目标跟踪算法方面,本文主要研究了Mean shift算法、Kalman滤波算法以及基于Kalman滤波的Mean shift算法三种目标跟踪算法,并结合嵌入式开发环境在Linux+Qt平台下分别对三种算法进行了仿真实验。通过对实验结果的分析,验证了每种算法对运动目标进行跟踪的有效性,并通过对实验中遇到问题的分析,对每种算法的优缺点进行了总结。在本文的最后阶段,基于对运动目标识别与跟踪算法进行了大量实验的基础上,选取了适合在嵌入式开发板ARMS3C2440上运行的目标识别算法和目标跟踪算法,并最终将两种算法衔接起来在嵌入式开发板ARMS3C2440上实现了对运动目标的识别和跟踪。
关键词 运动目标识别；运动目标跟踪；ARM；OpenCV；
Abstract
Over the years as people's lives and property protection awareness increasing, the requirements for video surveillance systems have become more sophisticated, intelligent video surveillance system has been widely attention and a lot of research. Moving target detection and tracking technology as an intelligent video surveillance system, key technology has also been rapid development. ARM embedded technology as one of the current study the hottest IT technologies, its low-cost, low-power, high-performance features, has been widely used. The video surveillance system technology combined with the ARM Embedded intelligent monitoring system is the current trend of development. This paper mainly focus on moving target detection algorithm and moving target tracking algorithm embedded hardware platform A study carried out by selecting a moving target experiments and moving target detection algorithm for tracking algorithm embedded ARM platform, and eventually in the ARM Embedded Development On board ARMS3C2440 realize the moving target detection and tracking. In terms of moving target detection algorithm, this paper studies the frame difference, background modeling method, Surendra automatic background updating method of the three algorithms, combined with an embedded development environment on Linux + Qt platform simulation experiments. In moving target detection experiments on frame difference in key parameters binarization threshold do three sets of comparative experiments, by analysis of the experimental results, the conclusions drawn using the frame difference method. Using background modeling method for detecting moving target experiments background modeling method for key parameters learning rate three comparative experiments done by analyzing the experimental results, the conclusions drawn. Surendra using automatic background updating method for moving target detection experiments, experiments verify the effectiveness of the algorithm to detect moving objects by analyzing the experimental results, the conclusions drawn. In terms of moving target tracking algorithm, the paper studies the Mean shift method, Kalman filtering method and the method of three Mean shift tracking algorithm based on Kalman filtering, combined with embedded development environment for the three algorithms were under Linux + Qt platform simulation experiments. Through the analysis of experimental results verify the effectiveness of each algorithm to track moving targets, and problems encountered by the experimental analysis of the advantages and disadvantages of each method are summarized. In the final stage of this article, based on a moving target detection and tracking algorithms are a lot of basic www.biyezuopin.cc
experiments, the selected target detection algorithms and target tracking algorithm suitable for running on embedded development board ARMS3C2440, and eventually link up the two algorithms in the embedded development board ARMS3C2440 realize the moving target detection and tracking.
Key words Moving Target Detection；Moving Target Racking；ARM； OpenCV
目 录
摘 要 I
Abstract III
第1章 绪 论 1
1.1 课题的背景与研究意义 1
1.2 国内外研究现状 2
1.2.1 目标识别与跟踪技术的发展 2
1.2.2 ARM的发展 3
1.3 研究的主要内容 4
1.4 论文的章节安排 4
第2章 系统总体功能分析 7
2.1 系统功能分析 7
2.2 系统的总体设计 7
2.2.1 系统总体结构框架 7
2.2.2 系统硬件的总体设计 8
2.2.3 嵌入式处理器的选择 8
2.2.4 操作系统的选择 8
2.3 本章小结 14
第3章 图像预处理 15
3.1 颜色空间 15
3.1.1 RGB颜色空间 15
3.1.2 HSV颜色空间 16
3.2 图像滤波 18
3.2.1 中值滤波 18
3.2.2 高斯滤波 19
3.2.3 图像的膨胀和腐蚀 21
3.2.4 图像二值化 23
3.3 本章小结 24
第4章 运动目标识别与跟踪算法 25
4.1 引言 25
4.2 常用运动目标跟踪算法分析 25
4.2.1 卡尔曼滤波算法 25
4.2.2 CAMShift算法 28
4.2.3 Surendra算法 29
4.3 基于几何特征的卡尔曼滤波和直方图匹配算法 32
4.3.1 几何特征计算 32
4.3.2 卡尔曼滤波 34
4.3.3 灰度直方图匹配 35
4.3.4 多目标跟踪问题 36
4.4 跟踪算法流程及实验结果 37
4.4.1 跟踪算法流程 37
4.4.2 跟踪算法实验结果 38
第5章 系统软硬件平台的搭建 45
5.1 系统硬件开发平台 45
5.1.1 ARM开发板 45
5.1.2 视频采集设备 45
5.2 系统软件平台 46
5.2.1 构建交叉编译环境 46
5.2.2 QT嵌入式平台的移植 46
5.2.3 OpenCV函数库嵌入式平台的移植 48
5.3 嵌入式平台开发流程 52
第6章 嵌入式运动目标识别与跟踪系统实现 55
6.1 对运动目标识别算法的选择 55
6.2 对运动目标跟踪算法的选择 57
6.3 基于ARM嵌入式的运动目标识别与跟踪系统的实现 59
结 论 61
参考文献 63
攻读硕士学位期间发表论文及所做工作 67
致 谢 69
