SIA OpenIR  > 光电信息技术研究室
小波多尺度跟踪技术研究
Alternative TitleResearch of Target Tracking Based on Multi-scale Wavelet Transform
惠颖1,2
Department光电信息技术研究室
Thesis Advisor罗海波
ClassificationTN911.7
Keyword目标跟踪 小波变换 目标分割 特征检测
Call NumberTN911.7/H87/2007
Pages73页
Degree Discipline模式识别与智能系统
Degree Name硕士
2007-06-01
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract本论文研究的主要内容为基于小波多尺度特性的序列图像目标跟踪技术。目标跟踪作为一个在军事、工业和科学研究方面有着广泛应用背景的研究领域,一直以来吸引了大批国内外学者。由于小波变换具有多分辨率分析的特点,而且在时频两域都具有表征信号局部特征的能力,使得基于小波变换的目标跟踪算法具有传统算法无法比拟的优势。针对目标跟踪技术的研究现状和存在问题,本文着重从目标分割和特征检测与匹配两个角度对基于小波变换的几种新的目标跟踪方法进行了研究。 1. 采用基于多尺度Gabor小波的特征点检测算法对序列图像进行跟踪。借助图像的金字塔变换得到多尺度的Gabor小波特征图像,并对特征图像进行特征点检测,提取对图像变换具有鲁棒性的特征。针对两种特征检测方案,提出不同的特征匹配准则,按照分层匹配的策略由粗到精逐步定位目标的准确位置,具有较快的搜索速度。 2. 采用多尺度小波函数所提取的相位一致性特征进行基于目标分割和基于角点特征的跟踪。对目标图像进行相位一致性检测,得到一个具有光照不变性的无量纲特征量—相位一致系数。利用相位一致性检测的这种特性,针对孤立目标的情况,提出了两种自适应目标分割和跟踪的算法。基于区域增长的目标分割算法利用从相位一致图像中找到的对比度最大点及其法线方向两边的灰度分布确定目标和背景的种子像素,进行自适应目标分割。基于相位一致性检测的目标分割算法只需确定一个阈值即可利用相位一致特征图像的方向性,依据目标在不同方向响应的不同将目标和背景区分开,适应于复杂纹理背景中的目标分割。最后,分别将两种算法所得的分割结果向水平和垂直方向投影即可确定各自的质心位置,实现自适应的质心跟踪。进一步提取相位一致性图像的最小矩特征就能得到目标的角点信息。文中用实验验证了此方法检测到角点的综合性能。在此基础上,提出了利用单演相位差进行角点匹配跟踪的算法,并将其同基于灰度相关的匹配算法进行了对比,证明了本算法能够检测出更多准确匹配的角点、减少误匹配,同时具有较小的匹配运算量。对以上提出的几种目标跟踪算法进行了大量的仿真实验,实验结果表明,这几种方法均取得了较好的跟踪效果,能够实现稳定、精确的跟踪。
Other AbstractThe thesis is focused on the technique of multi-scale wavelet transform based target tracking in image sequences. As a widely used technology in military, industry and scientific research, tracking of target in image sequences has always been attractive to many experts. Wavelet transform has multi-resolution property, and it is highly localized in both time and frequency domains, which lead to wavelet transform based tracking method performs much better than other traditional tracking method. Aimed to the main problems existed in object tracking, 3 new wavelet based target tracking methods are present in this thesis. 1. Tracking interest object in image sequence using feature points extracted by multi-scale Gabor wavelet transform. Implement pyramid transform and Gabor wavelet transform to get multi-scale Gabor feature images. Then, extract feature points which are robust to image transformation from feature images. Furthermore, two different matching criteria are established aimed to two feature detection strategies. At last, we fix target position according to a coarse-to-fine matching scheme with proposed strategy, which is proved has rapid searching speed. 2. Tracking interest object using Phase congruency (PC) extracted by multi-scale wavelet. First, calculating Phase congruency of the target image to get a dimensionless measure, phase congruency value, that is also invariant to variations in image illumination and contrast. Then, we proposed two adaptive target segmentation and tracking methods aimed to track isolated object by using such properties. Region-growing based segmentation finds the point with max intensity contrast through PC image firstly, and then fixes the seeds of target region and background region according to the intensity distribution along its perpendicular direction. Then segmentation can be realized by using the seeds. Phase congruency based segmentation makes use of multi-orientation PC images, and distinguishes the target from background through different response of 6 orientations with a fixed threshold. This method can be used to tracking target in complex texture background. At last, we project the two segmentation results, binary images, to horizontal and vertical direction separately, and use the projection results to find target and realize adaptive tracking. Furthermore, more information can be got through calculating principle moments of phase congruency, where minimum moment indicates corners. The thesis validates the good performance of this corner detector. In order to use corners to tracking, we proposed a feature points matching strategy by looking for points that have minimal differences in monogenic phase. And our method performs rather well relative to grayscale correlation based matcher. There are more putative matches found and fewer outliers; besides, the matching stage is much faster as the fewer matching computational cost. Large numbers of simulation have been implemented to test our algorithms. Experimental results show that all these methods perform well in most applications, the stable and accuracy tracking is realized.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/348
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
惠颖. 小波多尺度跟踪技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2007.
Files in This Item:
File Name/Size DocType Version Access License
10001_20042801472804(7171KB) 开放获取--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[惠颖]'s Articles
Baidu academic
Similar articles in Baidu academic
[惠颖]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[惠颖]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.