SIA OpenIR  > 光电信息技术研究室
Alternative TitleResearch on Application Theory of Model-based Monocular Vision
Thesis Advisor谈大龙
KeywordP3p 位姿测量 鲁棒性 误差分析
Call NumberTP391.41/Z81/2010
Degree Discipline模式识别与智能系统
Degree Name博士
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract在实际工程应用中,经常需要自动完成两个物体间位置与姿态的测量,如自主交会对接、机械手的抓取操作、目标跟踪等。在需求的牵引下,出现了多种采用不同原理的测量传感器,如激光、微波、GPS、视觉测量等。其中视觉测量由于具有被动、非接触、灵活等特点,是一种备受关注的测量方法。视觉测量理论的研究已经持续了很长时间,取得了一大批重要的成果,但仍有未完全解决的理论问题,它目前仍然是计算机视觉领域的一个研究热点。与此同时,依托已经取得的理论成果,视觉测量已经作为有效的测量方式在很多领域得到了实际应用。视觉测量的基本工作方式及原理是:以摄像机为传感器,对目标物体成像,通过目标物体的特征在图像中的成像位置完成测量计算。视觉测量在不同的条件与配置下,可以测量出目标物体的方位角、距离、相对位置、相对姿态等。其中,基于模型的单目视觉是测量位置与姿态的一种重要方法,本文结合实际工程应用背景,对基于模型的单目视觉在实际应用中遇到理论问题进行研究,主要内容包括:1) 基于模型的单目视觉的计算可以归结为求解高次方程,研究该方程的求解有很重要的理论意义,但由于其计算的时间复杂度较高,实际应用时有一定困难。本文针对这一问题,在一定的约束条件下,以实际工程应用为背景,研究数值解法,提出了一种基于几何意义的两分迭代算法,算法具有收敛快、精度高的特点。2) 物体特征在图像中的成像坐标是视觉测量计算的依据,由于特征提取算法与量化误差的影响,检测出的特征图像坐标一定含有误差,这一误差成为影响测量精度的重要因素,误差分析是视觉测量研究的重要研究内容。当目标与摄像机的相对位置不同时,同样的坐标误差所引起的测量误差也不一样,本文面向实际工程应用,用多种方法分析了等腰三角型这类目标与像机间俯仰角的变化对测量精度的影响,得到了光轴与合作目标中3个特征点所构成的平面平行时测量精度最高的重要结论,对实际测量系统的设计具有重要指导意义。3) 在具有双相机系统中,立体视觉计算是常用的测量方法之一,本文在一定条件下,比较了两单相机融合算法与基于立体视觉算法的测量精度,结果表明,在位置计算时,两种方法的精度基本相当;姿态计算时,前一种方法精度明显高于后一种方法,这一结论不仅具有理论意义,更对实际应用系统中算法的选择具有重要指导意义。4) 分别以水下机器人、移动机器人与机械手的定位操作为例,进行了本文研究的算法在实际系统中的应用研究,结果表明,本文的方法可以满足许多实际系统对测量的要求。本文的研究成果既是对视觉理论的丰富与补充,又有实际的应用价值。
Other AbstractIn practical engineering applications, the position and pose between two objects need to be measured automatically, especially in areas such as automatically rendezvous and docking, snatching operation by a manipulator, object tracking, and so on. Under such demands, kinds of measurement-sensors based on different principles have been invented, for instance, laser sensor, microwave sensor, GPS and vision estimation. Nowadays, vision estimation has recieved considerable publicity for its advantages such as passivity, non-contact and agility. Although study on theory of vision estimation has been conducted for a long time, and substantial results have come out, there are still some problems not solved, which make it a hotspot in computer vision. Meanwhile, based on current theoretical achievements, vision estimation has been widely applied in many fields as an effective measurement method. The main function ways and principles of vision estimation are as follows: take cameras as imaging sensors for the objects, and then carry out the measurement using imaging positions of objects’ features. Under different conditions and configurations, we can get information such as the azimuth, distance, relative position and pose of an object. Model-based monocular vision is an important method for position and pose estimation. In this paper, we mainly studied the model-based monocular vision by combining some theoretical problems with  engineering applications. Main contents are as follows: 1) Calculations for model-based monocular vision can be concluded as solving high-order equations, which has vital significance. But the complicated computations make it difficult for practical use. In this paper, numerical solutions for practical application have been studied under certain constrains, and an iterative optimization algorithm of dichotomy based on geometric meaning is proposed. The algorithm can converge quickly with high precision. 2) The imaging coordinates of an object’s features are the basis of vision estimation. However, feature extracting algorithm and quantization errors cause some errors in the detected image coordinates, and these errors are main factors that affect the measurement precision. Thus error analysis is an important part of vision estimation. While the relative position between an object and the camera are different, the same imaging coordinate error may lead to different measurement errors. In this paper, we aim at practical applications and analyze the effect of changes of pitching angle between objects and camers within an isosceles triangle on estimation precision. Finally, we have concluded that optimal precision is gained when the axis parallels with the plane formed by the 3 character points in the object, whis may have important guidance significance in designing a measurement system. 3) When the system has two cameras, stereo vision measurement is a commonly used method. In this paper, we have a comparison of measurement precision between the fusion algorithm of two monocular cameras and the algorithm based upon stereo vision, under certain conditions, It comes out that the two algorithms have equivalent precision in position estimation, while the monocular vision based algorithm has much higher performance in pose estimation. This conclusion not only has theoretically significance, but also has guidance meanings in algorithm selection for application systems. 4) Taking underwater-robot, motion-robot and manipulator for instance, we have studied the application of our algorithms in practical systems. The results show that the precision of thses methods can satisfy most practical systems. The achievements of this paper not only supplement the vision theory, but have value of practical application.
Contribution Rank1
Document Type学位论文
Recommended Citation
GB/T 7714
朱枫. 基于模型的单目视觉应用理论研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2009.
Files in This Item:
File Name/Size DocType Version Access License
基于模型的单目视觉应用理论研究.pdf(2367KB) 开放获取LicenseApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[朱枫]'s Articles
Baidu academic
Similar articles in Baidu academic
[朱枫]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[朱枫]'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.