SIA OpenIR  > 光电信息技术研究室
基于EIV模型的点线位姿估计研究
Alternative TitleResearch on EIV Model Based Pose Estimation for Noisy Points or Lines
汪俊文1,2
Department光电信息技术研究室
Thesis Advisor朱枫
ClassificationTP391.4
Keyword位姿估计 双重四元数 Eiv 伪线性化 奇异值分解
Call NumberTP391.4/W27/2007
Pages70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位姿估计是计算机视觉的一个重要的研究内容,它在目标定位、摄像机标定、手眼系统、移动机器人及三维重建等领域有着广泛的应用。常用的确定目标位姿的视觉方法包括基于模型的单目视觉方法和双目视觉方法。目前,单目视觉方法大多采取点特征和线特征,这是因为点线特征的提取比较容易,而且具有数学上的简易性。在实现目标位姿测量时,像素量化引起的特征点的图像坐标噪声是不可避免的问题,它必将引起物体的位姿计算值的误差。同时,为了充分利用图像处理中提取的点线特征,我们希望能将这些特征在一个统一的数学模型下进行位姿估计。针对上述问题,本文取得了如下研究成果:首先,采用双重四元数同时表示旋转和平移,从而将点线特征约束统一成二次型约束的形式。这种形式利于采用伪线性化方法简化问题,进而通过迭代算法求解位姿参数。然后,为了在位姿估计中考虑到量化误差的影响,引入了EIV模型描述影响点线特征投影的量化误差。同时,给出了优化目标函数,提出了基于奇异值分解的迭代算法来估计位姿参数,并在迭代算法中通过增加一条收敛规则解决了实现过程中遇到的振荡问题,保证了算法的收敛。和传统的迭代算法相比,该算法具有以下两个特点:第一,受初始值的影响小,实验中采用随机初值依然可以收敛;第二,迭代过程不是采用传统的步长搜索,收敛速度快。此外,本文还研究了位姿参数估计值的统计特性,证明了估计值的无偏性。最后,考虑到量化误差、点线特征数目和目标相对摄像机的距离等影响位姿估计结果的几个因素,本文设计了仿真实验和实际实验分别对点特征和线特征的位姿估计进行测试。结果表明:该算法可以应用于任意三个以上点特征和线特征的情况,受初值影响小,收敛快,提高了位姿估计结果的精度和鲁棒性。
Other AbstractPose estimation appears important in computer vision. It has many applications such as object positioning, camera calibration, hand-eye system, mobile robots and 3D reconstruction. This problem is usually solved by model-based methods which include monocular and binocular ones. Currently, pose estimation mostly uses point and line correspondences because point and line features can be easily extracted and described for their simplicity in mathematics. We expect to employ both points and lines in a uniform framework for estimation with the sufficient use of the extracted features from image processing. Besides, the quantization error in pixels is an inevitable obstacle in pose estimation. Aimed at the problems above, we have done the following researches: First, we used dual number quaternions to describe both rotation and translation so that the constraints from point or line correspondences can be turned into unified quadratic forms. Then we can employ pseudo-linearization to simplify the forms and estimate the pose parameters with iterative algorithm. Second, we introduced the EIV model to describe the noise in images so that we can take account of the noise in the process of estimation. We provided the error function and presented an iterative algorithm based on singular value decomposition. However, it happens easily that the algorithm vibrates at two values because of the symmetric quadratic constraints. Thus, we introduced an additional rule for its convergence. Compared with some traditional methods, our method can be used with random initials and iterates not based on steps. As a result, it has a weaker dependence on initial solution and a fast convergence. In addition, we investigated the statistical properties of the estimator which was proved unbiased. Finally, we designed the simulation and real experiments taking account of the noise in images, the number of features and the distance between objects and the camera. The results demonstrated that our method can be used with more than three point or line features and it can converge fast with random initials and provide reliable accuracy and robustness.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/344
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
汪俊文. 基于EIV模型的点线位姿估计研究[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_20042801472803(1328KB) 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
[汪俊文]'s Articles
Baidu academic
Similar articles in Baidu academic
[汪俊文]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[汪俊文]'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.