SIA OpenIR  > 光电信息技术研究室
基于直线的单目视觉位姿计算方法研究
Alternative TitleResearch on Monocular Pose Estimation Methods Based on Line Features
刘昶1,2
Department光电信息技术研究室
Thesis Advisor朱枫
ClassificationTP391.41
Keyword单目视觉 位姿定位 线特征 点特征 闭式解
Call NumberTP391.41/L71/2012
Pages122页
Degree Discipline模式识别与智能系统
Degree Name博士
2012-05-29
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract直线特征是现实世界中广泛存在的一类基本特征,利用单目视觉进行基于直线特征的视觉定位是计算机视觉领域一个重要的研究方向,同时也具有广泛的应用价值。本文针对基于直线特征的位置与姿态参数的计算方法问题进行研究,在理论研究方面,包含了对该领域前沿问题的研究,取得的成果是对现有的单目视觉定位理论的完善,在应用研究方面,解决了无人机相对于机场跑道的位姿参数估计问题,给出了最优估计和工程化算法。论文具体研究内容包括以下几个方面:    首先,研究了应用相交于两点、具有平行或垂直关系的三条直线的三线定位问题(perspective-three-line problem),证明了在此特定情况下问题解的数量与相机光心与已知直线组的相对位置有关,找到了问题存在唯一解时光心所在的区域,同时给出了问题求解的闭式方法。该成果对实际应用中的特征选择、合作目标设计等都具有重要的指导意义;闭式求解方法避免了复杂非线性方程组的求解过程,简化了计算过程,具有很好的实时性。    其次,研究了应用点、线混合特征进行单目视觉位姿定位的问题。分别针对共面的两点一线特征、非共面的两点一线特征、两线一点特征等情况研究了位姿计算问题,得出每种情况下的闭式解法、解的数量特性等结论。由于点、线混合特征在现实世界中普遍存在,该部分研究结果为实际应用中合理选择点、线特征进行单目视觉定位提供了理论依据,使特征的可选取范围进一步扩大。    最后,研究了具备冗余信息条件下的位姿计算方法,针对无人机着陆阶段相对于机场跑道的位姿参数估计问题,提出一种后验均值估计方法。目前无人机上均能安装机载的位姿传感器,这种方法考虑机载的位姿传感器的测量值均存在误差,并将这些传感器在某一时刻的测量值作为真实参数的一次采样值,利用从机载相机同时拍摄到的机场图像中提取出跑道的两条边缘线,得到真实参数必然满足参数空间中的视觉约束。当传感器的测量参数的误差分布已知时,计算在当前采样值下参数的后验均值作为位姿参数的估计值,由贝叶斯估计理论,该估计是在后验风险取平方损失函数条件下的最优估计。仿真实验表明,该方法相比于一般的“基于投影关系的方法”具有明显的优势,可显著地提高估计精度。后验均值估计方法是一种综合了传感器信息和视觉信息的数据融合方法,为位姿参数估计工程化算法的性能评价提供了标准,但具有实时性差的缺点。为适应实际应用,又提出了一种通过计算采样点到视觉约束曲面加权最短距离点进行参数估计的方法,同时应用粒子群随机搜索算法进行最短距离点的搜索,在保证精度的前提下有效地减少了计算量,能够达到工程应用对算法实时性的要求。
Other AbstractLine feature is one of features widely existed in the real world. Monocular pose estimation based on line features is one of important research fields in machine vision, and can be widely applied to many real applications. In this thesis, some pose estimation methods based on line features are studied. It includes hot and frontier theoretical problems research in this field. It also proposed some applicable methods for UAV (Unmanned Aerial Vehicle) pose estimation problem from airfield runways. The main contents and results obtained are as follows:    Firstly, this thesis studied P3L (Perspective-three-Line) problem in some special cases that three lines intersect at two points and have orthogonal or parallel relationship. It is proved that under these conditions, the problem has closed form solutions, and the solution number of the problem depends on the geometric relationship between the camera’s optical center and the three target lines. If the camera’s optical center locates in a special area, the problem has unique solution. The unique solution areas are described according to the relationship of the three given lines. The results are valuable for selecting or designing cooperative targets. In contrast to traditional complicated nonlinear equation system solver, the closed form method is superior in solving procedure and real time property.    Secondly, this thesis studied pose estimation using point and line mixed features. In following three cases: coplanar two point and one line features, non-planar two point and one line features, one point and two coplanar line features, it gave the closed form solution and solution number properties of the pose estimation problem using according features. For the sake of wide existence of point and line mixed features in the real world, the results expand the feature selection range and are useful in selecting reasonable features in real vision applications.    At last, this thesis studied the pose parameter estimation problem under the condition that redundant information exists. For the vision assistant UAV landing problem, it presented a posterior estimation method for UAV pose parameter estimation from two edge lines of the airfield runway. Generally some onboard pose sensors have been mounted on UAVs. It is supposed that all these sensors have independent measurement errors and the prior probability density of each parameter’s error is given, then the measured values from sensors could be considered as a sample of the real ones. At the same time of sampling, an image of the runway is acquired from an onboard calibrated camera. Suppose the two edge lines of the runway can be detected exactly. From the perspective view of two parallel lines, the real parameters must locate on a constraint surface in the parameter space. Then the posterior mean estimation of the parameters can be obtained. From Bayesian estimation theory, the posterior mean estimation is the optimal estimation under the condition that the posterior risk function is the squared loss. Simulation results show that the method is effective. This method is a date fusion method which combines the vision information and sensors information. It can also provide a reference criterion for other methods’ performance evaluation, but its disadvantage is its real time property is not good. For real time application of UAV landing, the thesis proposed another approach whose main idea is to minimize the weighted distance from sample point to the constraint surface in the parameter space, and use the shortest distance point on the constraint surface to estimate the pose parameters. By using PSO (Particle Swarm Optimization) algorithm to the shortest distance point searching, the calculating time is greatly reduced and simulation results show this approach can satisfy the time and exact requests of real time application
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/9241
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
刘昶. 基于直线的单目视觉位姿计算方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2012.
Files in This Item:
File Name/Size DocType Version Access License
基于直线的单目视觉位姿计算方法研究.pd(1382KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘昶]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘昶]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘昶]'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.