摘要本课题就视频图像序列中运动目标检测的问题进行研究。运动目标检测在许多领域都有着较高的运用价值，其应用范围涵盖了军事、医学、交通、工业等，所以一直以来备受国内外学者的关注。本文对运动目标检测技术的应用前景、研究现状等进行综述，了解了其现有的研究成果，重点掌握帧差法、背景差法、光流法三种算法，并结合matlab实现仿真，直接测试了四种算法(即三帧差分法、混合高斯背景建模、均值建模、HS光流法)的检测效果。78805

比较几种算法的处理效果，HS光流法较好得实现了检测，但计算量大，对噪声敏感，不能得到准确边缘；三帧差法一定程度上解决了普通帧差法的重影问题，但存在空洞现象，难以将一个目标作为整体检测出来，这种算法计算量很小，适合于实时处理；均值建模法的检测效果在简单场景下十分精确，但由于实际情况的多变，这种算法的局限性很大；混合高斯建模法较均值建模法实现的背景的实时更新，但这种算法的计算量很大，效果有待进一步优化。

毕业论文关键词  运动目标检测、matlab、三帧差分、混合高斯、图像分割

毕业设计说明书外文摘要

Title    Study on algorithms of moving object detection in video sequence                                               

Abstract In this paper,I do a study on moving object detection in video sequence。It has extensive application prospects in many fields,such as military，medicine，traffic，

Industry,etc。So researches on the technology of  moving object detection are carried on actively,both in China and abroad。The paper includes the content of prospect,status analysis,results and etc。With the purpose of mastering the three classic algorithms,which are temporal difference,background subtraction and optical flow,i choose four algorithms to have a test on the platform of matlab。The four algorithms are Gaussian mixture model,three-frame difference,mean background,HS optical flow。

The fours algorithms have different effects。Optical flow method has a good effect,however,it has large calculation amount and is sensitive to noise without accuracy edge;Three-frame difference method can remove ghost image in some degree。It can be used in real-time monitoring because of the small calculation amount。But holes still exist in this method,what leads to the incomplete detection;In the case of simple scence,mean background model have a nice effect,but the algorithm is limited to the changeable condition in practice;GMM can realize the real-time background update compared to mean background model,but the calculation amount is too large and the effect needs further improvement。 

Keywords  moving object detection,matlab,three-frame difference,Gaussian mixture,image segmentation

目   录

1  绪论 1

1。1  背景介绍  1

1。2  国内外发展现状 2

1。3  本文概述  2

2  运动目标检测的算法 4

2。1  算法总结 4

2。2  帧间差分法 4

2。3  背景减差法 5

2。4  光流法 7

3  技术难点及改进方案 9

3。1  技术难点 9

3。2  改进方案 9

4  算法仿真 10

4。1  仿真概述 10

4。2  基础知识说明 10

  4。3  三帧差法运动目标检测 11

4。4  HS光流法运动目标检测 15

4。5  混合高斯模型运动目标检测 21

4。6  均值法运动目标检测 26

结论  29