SIA OpenIR  > 光电信息技术研究室
基于空间特征匹配的视觉目标跟踪技术研究
Alternative TitleImage Spacial Feature based Visual Tracking System
周全赟1,2
Department光电信息技术研究室
Thesis Advisor史泽林
ClassificationTP391.41
Keyword局部特征 计算跟踪 辐射模糊 特征稳定性 单应矩阵
Call NumberTP391.41/Z75/2013
Pages60页
Degree Discipline模式识别与智能系统
Degree Name硕士
2013-05-28
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract传统的目标跟踪算法通常选取波门内的矩形区域作为跟踪模板,采用相关匹配或迭代算法在后续帧中寻找目标的新位置。其有效性的前提之一是波门内图像区域具有足够的纹理特征能和周围区域分开来,否则很容易发生漂移或者丢失现象。针对待跟踪目标不具备足够的纹理信息以实现波门跟踪的应用需求,本文提出了计算跟踪算法。与传统波门跟踪算法不同,波门跟踪利用波门内的图像信息进行相关匹配来确定目标位置,而计算跟踪则充分利用图像所有信息以计算得到目标位置。波门内的信息通常情况指的是波门内的信息,而计算跟踪所利用的图像全局信息是指运用多种图像局部特征提取算子对图像信息进行解构,在目标跟踪的框架下对多种信息进行统筹计算以确定目标在新的图像中的准确位置。计算跟踪所涉及的两大关键技术即图像局部特征提取算法和融合多种信息条件下对目标位置的计算算法。图像的局部特征提取算法成为计算机图像和视觉相关领域的研究热点,近年来在多视几何、三维重建、目标跟踪等领域获得了广泛的应用。图像的局部特征是对灰度分布符合某一假设的图像局部区域,因为假设的不同而形成多种不同的局部特征提取算子。在面向目标跟踪的应用需求中,因为在跟踪过程中成像条件不断变化,图像局部特征在计算跟踪过程中的稳定性尤为重要。运用前后帧图像中提取的局部特征进行匹配,以计算出相邻图像之间的变换关系,从而以递推方式来确定目标的跟踪位置。逐帧递推计算往往会造成跟踪误差的发散,融合图像多种信息则可以对误差发散进行校正和抑制,从而实现计算跟踪的稳定性。 本文在对计算跟踪特殊的应用需求和基于局部特征的图像信息解构方法进行了详细的阐述后,提出了基于多特征融合的计算跟踪算法。在面向跟踪的应用需求中,特征点的稳定性是关键问题。通过对相向运动下图像辐射状模糊机理的建模分析,给出了特征点的筛选方法,以获得最高重复率的特征子集。相邻两帧图像之间的变换关系(单应阵)的计算精度和时间复杂度在面向基于递推计算的跟踪应用中尤为重要,通过对几种单应阵优化计算算法的时间复杂度和计算精度的比较,选取了综合表现最优的单应阵计算算法。最后提出了融合点特征SIFT和区域特征MSER的计算跟踪方法,两种特征之间的互补性抑制了误差的发散性,实现了稳定鲁棒的计算目标跟踪。
Other AbstractTraditional target tracking algorithms usually take the windowed area as the tracking template, adopt the correlation matching or iterated algorithm to calculate the new target position in the following frames. One of the prerequites of this method is that windowed area contains enough texture to be distinguished from its surrounding area. If not, the tracking will be drifted or losted. As to the application that the target do not have the enough texture to make the windowed tracking(WT), so we propose the extensive computational tracking(ECT) algorithm. The difference between the WT and ECT is that, WT uses correlation matching by the information in the windowed area to calculate the target position, but the ECT effectively exploits the information within the whole image to compute the target location. The information in the tracked window usually means the gray information in the windowed local patch, but the ECT adopts the multiple image feature detector to extract the useful information within the image and fuses all the information under the target tracking framework to compute the target new location. The image feature detector is research focus of the computer image and vision related field, image feature receives extensive attention and rapid development in the multi-view, 3D reconstruction and target tracking areas. The image feature is the local patch whose gray distribution agrees with the detector’s hypothesis. In the target tracking oriented application, for the imaging conditions are changing by the tracking process, so the stability of the image feature if critically important in the tracking application. The features are matched in the adjacent frames, and then the geometric transformation matrix is calculated and the target position is computed in the recursive way. One-by-One recursive calcution is always leading to the divergence of the tracking error. Multiple detectors and information fusion will suppress the divergence and realize the stability of the ECT. With the detailed description of the special requirements and the feature based image information extraction method, this dissertation proposed the multi-feature based ECT. In the tracking application, the stability of the image feature is the critical problem. With the analysis of the image radial-blurry under the opposite-direction movement, we proposed the feature sifting method to get the high-repeatability feature set. The computation precision and time complexity of the holography computation algorithm is important in the recursive tracking framework. After the comparison of the several holography calculation methods by the time complexity and computation precision, the method with the best overall performance is adopted in the ECT. Finally, we propose the ECT framework which fuses the point feature-SIFT and the region feature-MSER. The complementary of the SIFT and MSER suppress the divergence of the tracking error and realize the stable and robust ECT.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/10764
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.中国科学院大学
Recommended Citation
GB/T 7714
周全赟. 基于空间特征匹配的视觉目标跟踪技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2013.
Files in This Item:
File Name/Size DocType Version Access License
基于空间特征匹配的视觉目标跟踪技术研究.(3833KB) 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[周全赟]'s Articles
Baidu academic
Similar articles in Baidu academic
[周全赟]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[周全赟]'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.