CASIA OpenIR  > 毕业生  > 硕士学位论文
Thesis Advisor马颂德
Degree Grantor中国科学院自动化研究所
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
Abstract在计算机视觉领域里,立体视觉是一种广泛用于物体深度信息检 测的方法。目前已有很多相关方法见于文献中,它们大致可分为两类: 基于特征的立体视觉方法和基于光流的立体视觉方法。对于基于光流的 方法,其固有的大运算负荷严重限制了它的实际应用,并且光流计算本 身就是个不适定问题,需要人为加上一些启发性约束才能得到实际的唯 一解。而对于基于特征的方法,如何提取适当的特征,及如何进行特征 匹配,很久以来就是个难以解决的问题。 本文中,我们引入一种基于主动视觉的立体视觉方法,它把确定 物体深度的问题转化为一个容易得多的计算直线斜率的问题。其创新之 处在于该方法不涉及任何特征匹配问题。该法可以简单描述如下: 首先通过控制主动视觉平台沿x轴运动,并均匀摄取若干㈣幅图 象Ik(i,j),i,j=0,1…,N,k=1,2,…M,其中M是图象大小。然后从这M幅图象 中生成虚拟图象Io(i,j,k)。可以看到,对应于同一空间点的图象点应处 在某一图象平面Io(i,j=jo,k)的一条直线上,而且该空间点的深度正比 于直线斜率。 因此, 这种方法把传统立体视觉问题转化成直线提取问 题,而后者显然要容易得多。 此外,本文进行了系统的误差分析,如主动视觉平台的俯仰角、 侧摆角造成的影响,深度分辨率分析等等。最后,我们设计了一些实验 来验证该方法的正确性和实用性,实验结果令人满意。
Other AbstractStereo vision is a widely used technique to get depth information of object in computer vision field. There are many relevant techniques in the literature which can be roughly divided into two classes, feature based stereo vision and optical flow based stereo vision. For optical flow based stereo vision, its real applicability is severely limited due to its inherent heavy computational load. In addition, optical flow determination itself is in essence an ill-posed problem and some heuristic constraints must be added to obtain a unique solution in practice. For feature based stereo vision, the problem of how to extract appropriate features and how to determine the feature correspondence has long been a challenging one, and is nothing easy to be solved for. Hence, how to find out a robust stereo vision method is of constant interests in computer vision community. In this work, we introduce an active vision based stereo vision method which converts the problem of the determination of depth information into a much easier one of calculation of the slope value of a straight line. The major novelty of the method is that it does not involve any feature correspondence problem. The method can be briefly described as follows: At first, multiple images, say M images, Ik(i,j),i,j =0,1,.. ,N,k = 1,2,...,M were taken at evenly located M points on x-axis by controlling an active vision platform to move along x -axis, where N is the size of the image. Then a virtual image Io(i,j,k) was formed by orderly ranging such M images. It was shown that the image points which correspond to a same object point must lie on a straight line on a image plane I0 (i,j = j0,k) and that the depth information of this object point is proportional to the slope value of the straight line, where Jo is a constant. Hence, the proposed method converted the traditional stereo vision problem into a line extraction one, and the later is much easier to be done. In addition, a systematic error analysis was carried out in the thesis, such as the influences of pan angle, tilt angle of the active vision platform, accuracy analysis of obtained depth information etc. Finally, some experiments were designed to test the correctness as well as the applicability of the proposed method, and the obtained results are satisfactory.
Other Identifier342
Document Type学位论文
Recommended Citation
GB/T 7714
王策. 基于主动视觉的立体视觉方法研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,1995.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王策]'s Articles
Baidu academic
Similar articles in Baidu academic
[王策]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王策]'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.