论文总字数：26324字

摘 要

前景目标检测即运动目标检测是对视频序列中目标的状态进行初始化的过程，广泛地应用于智能视频监控、高级人机交互以及军事目标打击等众多领域中，具有切实的应用价值与现实意义。尽管运动目标检测的发展已经迈过30个年头，仍然面临着诸如遮挡、光照等影响带来的巨大挑战。如何对视频或图像序列中的运动目标进行自动检测，是计算机视觉领域中的一项基础但又非常重要的研究，具有十分重要的意义。混合高斯模型与码本模型是最常用的背景建模方法，已广泛用于运动目标自动检测。

本文通过对运动目标的自动检测方法的调研，指出目前最常用的背景建模方法是混合高斯模型和码本模型。并在此基础上，分别对两类模型的原理、方法以及实现步骤进行了介绍和详细阐述。同时，针对传统的GMM模型和Codebook提出了一定的改进方案，在一定程度上抑制了光照变化等的影响，使它们能够适用于光照变化下的检测。

最后，本文在Visual Studio 2008环境下结合OpenCV函数库，实现了自动检测运动目标的实验程序，对相应的实验结果进行了说明与分析。本文所述的方法经过结果的证明能够较好地处理遮挡以及光照变化等影响，实现对运动目标的自动检测。

关键词：运动目标检测，背景建模，混合高斯模型，码本模型，光照变化

A STUDY ON FOREGROUND DETECTION ALGORITHM IN VIDEO SEQUENCES

ABSTRACT

Moving object detection is the process of state of the video sequence of the target will be initialized, widely used in intelligent video surveillance, advanced human-computer interaction, and many other fields like the fight against military objectives, with practical application value and practical significance. Despite the development of moving target detection has crossed 30 years, still faces enormous challenges, such as shelter, light and other influence. How the video or image sequence automatically detect moving objects is a very important research field of computer vision is a basic but has very important significance. Gaussian mixture model and code in the model is the most common background modeling method has been widely used for automatic detection of moving targets.

Based on the method for automatic detection of moving targets research, pointed out that the most common background modeling method is GMM and codebook model. And on this basis, respectively, principles, methods and implementation steps of the two models are introduced and elaborated. Meanwhile, the traditional model of GMM and Codebook made certain improvement program, to a certain extent, inhibited the effect of light changes so that they can be applied to detect changes under illumination.

Finally, this paper under the Visual Studio 2008 environment OpenCV library of motion detection performed automatic detection experiments, and the experimental results were analyzed and the corresponding details. Experimental results show that the method used in this paper can better handle the impact of shading and lighting changes, etc., to achieve automatic detection of moving targets.

Key words：Moving object tracking, Background modeling, Gaussian mixture model, Code book, illumination changes

目 录

摘 要 I

ABSTRACT II

1 绪论 1

1.1 研究背景及意义

1.2 运动目标检测的研究现状

1.3 难点与问题

1.4 论文结构安排

2 运动目标检测方法 5

2.1 帧间差分法 5

2.1.1 算法概述 5

2.1.2 方法评价 6

2.2 背景减除法 6

2.2.1 算法概述 7

2.2.2 方法评价 10

2.3 小结 10

3 基于GMM的运动目标检测 11

3.1 单高斯背景模型 11

3.2 混合高斯背景模型 12

3.2.1 参数初始化 13

3.2.2 模型更新 13

3.2.3 模型生成 14

3.3 GMM改进方法 15

3.3.1 模型更新改进 15

3.3.2 模型生成改进 15

3.4 小结 15

4 基于Codebook的运动目标检测 16

4.1 Codebook原理简介 16

4.2 Codebook背景建模 18

4.3 Codebook前景检测 20

4.4 Codebook改进方法 21

4.5 本章小结 23

5 运动目标检测实验 24

5.1 仿真环境 24

5.1.1 Visual Studio 2008 24

5.1.2 OpenCV 24

5.2 实验结果 25

5.2.1 实验数据 25

5.2.2 基于GMM的目标检测实验 25

5.2.3 基于Codebook的目标检测实验 27

5.3 实验对比分析 29

6 结论与展望 31

6.1 结论 31

6.2 展望 31

致 谢 33

参考文献 34

1 绪论

剩余内容已隐藏，请支付后下载全文，论文总字数：26324字