摘要：本课题为研究检测在变化背景中移动物体的方法。按照特征匹配的方式，基于SIFT（尺度不变特征变换）算子和FLANN Based Match（快速近似最近邻搜索）算法为基础提出了一种Merge Scale Invariant Feature Transform(MSIFT)融合不变特征变换视频移动目标检测方法。通过融合特征的方法，改进了SIFT算子在移动物体在较大的透视角度、不同视角匹配效果不好的弱点；通过使用层次聚类方法，增强搜索算法的准确率；最后，使用单应性标注物体的位置，并可估计被遮挡的物体的整体位置。实验结果表明，由OpenCV编写的该算基本实现了基于特征匹配的移动目标检测的任务，并且，相对于使用SIFT与FLANN Based Match 算法的组合进行匹配，准确度能达到80~98%，相比单独特征匹配，效率提升了50%。另外，该方法是一种图像自学习的架构，融合之后的可以表示成一个特征描述文件，下一次对于该物体的匹配可通过读取特征描述文件。77009

毕业论文关键词： 移动物体检测；特征融合；SIFT；层次聚类；机器学习

English Title Here

Abstract: The study was to find a specific moving target under the dynamic background。 Following concept of feature matching, A method of Merge Scale Invariant Feature Transform (MSIFT) had been put forward。 It is designed based on Scale Invariant Feature Transform and Fast Approximate Nearest Neighbor Search Library。 Initially, using MSIFT algorithm, it can be improved that the target was hard to be found under its state is at large perspective change and at a different view from the camera。 On the other hand, hierarchical clustering was used to strengthen the correctness of matching among the features Descriptors。 Finally, using Homography in order to label the target, which had the ability to estimate the hidden part of the target。 The experiments showed that the program based on OpenCV have the capability of finding the moving target under dynamic background with a 50 percent’s boost of accuracy and 80%-90% matching performance。 To be more specifically, it is a kind of learning system, the merged features can be reused as a descriptor model file next time。 

Keywords: Moving target detection; merge feature; SIFT; Hierarchical Clustering; Machine Learning

目录

1 概述 1

1。1 背景及应用 1

1。2 简介 1

1。3 论文工作概述 2

2 特征匹配与目标检测 3

2。1 局部特征算子概述 3

2。2 SIFT算子 3

2。2。1 尺度空间与极值检测 3

2。2。2 关键点搜索和定位 4

2。2。3 特征点方向确定 5

2。2。4 特征点描述 6

2。3 其他局部特征算子 6

2。3。1 SURF 7

2。3。2 ASIFT SIFT特征匹配的移动目标检测方法研究:http://www.youerw.com/tongxin/lunwen_88437.html