CASIA OpenIR  > 毕业生  > 博士学位论文
气动光学畸变图像复原和视觉目标跟踪方法研究
Alternative TitleResearch of Aero-Optical Distorted Image Restoration and Visual Tracking
谢源
Subtype工学博士
Thesis Advisor张文生
2013-05-28
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword气动光学效应 畸变图像复原 Bregman迭代 视觉目标跟踪 稀疏编码 Aero-optic Effect Distorted Image Restoration Bregman Iteration Visual Tracking Sparse Coding
Abstract高速飞行器的发展是一个国家国防实力的重要体现,发达国家投入大量的人力和财力开展了高速飞行器的研究。飞行器进入大气层高速飞行的过程中,光学头罩与高速气流发生剧烈的相互作用,引起气动光学效应,导致飞行器成像系统获取的图像产生了严重的模糊和畸变现象,极大的影响了飞行器对目标的检测、识别、跟踪与精确打击能力,迫切需要研究气动光学效应图像复原和目标精确跟踪的理论与方法。因此,研究畸变图像复原和视觉目标跟踪算法具有重要的理论意义和应用价值。 针对气动光学畸变图像复原,我们提出了一种基于低秩融合和时空域混合正则化的快速Bregman迭代复原算法,该算法能够对畸变图像进行高质量的复原。 针对视觉目标跟踪,我们分别提出了基于在线多示例梯度特征选择的目标跟踪算法、基于判别子空间和稀疏多视角模型的目标跟踪算法,以及基于在线判别字典学习和鲁棒性特征点配准的目标跟踪算法,上述三种算法能够很好的克服漂移问题。 本文主要工作与贡献如下: 针对气动光学效应畸变图像复原,提出了一种基于低秩融合和时空域混合正则项的快速Bregman迭代复原算法,并且在理论上证明了该算法的收敛性。算法一方面运用矩阵的快速低秩分解求解参考图像,获得的参考图像的质量较传统的均值图像有着较大的改善;另一方面用基于时域和空域混合正则项的Bregman迭代对配准后的图像进行复原,既保持了复原图像的内部结构又保证了原算法在时域上相关性。 通过与国际权威复原算法对比,实验结果表明我们的算法在图像的清晰度和形变度复原上都有明显的提升。 针对复杂场景视觉目标跟踪研究的问题,提出了三种有效的目标跟踪算法: (a)提出了一种基于在线多示例梯度特征选择的目标跟踪算法。针对穷举特征选择算法的高复杂度,运用Boosting框架下的梯度特征选择代替穷举搜索,最大限度地节约了计算资源;将多示例学习机制与在线梯度特征选择相结合,消除了分类器在线更新时的样本歧义性。 (b)提出了一种基于判别子空间和稀疏多视角模型的目标跟踪算法。针对生成模型未利用背景信息的缺陷,将基于最大间隔投影的判别子空间算法与基于增量生成子空间算法在贝叶斯框架下进行有机的结合,使得跟踪算法能够同时继承判别模型和生成模型的优势。针对最大间隔投影计算复杂度较高的缺陷,用计算复杂度较低的谱回归对投影矩阵进行快速求解,并且证明了最大间隔投影的求解过程与谱回归问题的等价性。针对目标受到遮挡导致的跟踪器性能衰退的问题,利用基于稀疏表示的多视角模型对跟踪器进行修正和增强,进一步克服漂移问题。 (c)提出了一种基于在线判别字典学习的目标跟踪算法。针对稀疏跟踪算法未利用背景信息的缺陷,定义了一种基于稀疏表示的可在线更新的判别字典,用该字典构造目标的判别模型,使得跟踪算法的区分能力显著提升;针对判别字典仅利用目标全局特征的缺陷,用基于稀疏表示的鲁棒性特征点匹配对判别字典进行局部信息的补充,使得跟踪器对目标运动估计更为精确。 上述三种算法均与国际权威跟踪算法在基准数据集上进行对比实验,实验结果表明本文...
Other AbstractThe development of high-speed aircraft is an important reflection of the national defense capability. Therefore, the developed countries have invested a large amount of materials and financial resources to conduct the research on high-speed aircraft. During the aircraft flying in the atmosphere with high speed, drastic friction occurs between the aircraft's optical hood and the high-speed airflow, thereby causing the aero-optical effects. As a result, the images received by the camera on the aircraft are blurring, distortion or shifting, which seriously affect the capability of the aircraft for target detection, recognition, tracking and precise attacking. Consequently, it is urgently need to study the theory and methods of the image restoration and visual target tracking. In the aspect of distorted image restoration, we propose a restoration method based on low rank fusion and fast Bregman iteration using the spatial-temporal mixed regularizer. In the aspect of the object target tracking, we have proposed three efficient and stable tracker. The one is a robust visual tracker based on online multiple instance gradient feature selection. The other one is a robust visual tracker based on discriminative subspace learning and sparsely represented View-Based model. The last one is a robust tracker based on sparse representation and online discriminative dictionary learning. All the methods mentioned above can handle the drifting very well. The main work and contributions are as follows: In order to restore the distorted image, we propose a restoration method based on low rank fusion and fast Bregman iteration using the spatial-temporal mixed regularizer, and giving the convergence proof of the algorithm. Firstly, low rank fusion is used to construct the reference image whose quality is superior to the mean image. Secondly, the spatial-temporal regularizer used in the Bregman iteration not only can preserve the local structure of the image but also can assure the smooth pattern between different restoration iterations. Experimental results demonstrate that the proposed method can effectively alleviate distortions and recover details of the scene. In order to track the object target in clustering background, we propose three efficient and stable trackers: (a) A robust visual tracker based on online multiple instance gradient feature selection has been proposed. Firstly, we employ a non-exhaustive feature selection approach based on the gradient d...
shelfnumXWLW1946
Other Identifier201018014629094
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6532
Collection毕业生_博士学位论文
Recommended Citation
GB/T 7714
谢源. 气动光学畸变图像复原和视觉目标跟踪方法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2013.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20101801462909(15343KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[谢源]'s Articles
Baidu academic
Similar articles in Baidu academic
[谢源]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[谢源]'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.