CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleDepth Recovery From CM
Thesis Advisor卢汉清
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
Keyword深度估计 同心拼图 极线平面图象 Depth Recovery Concentric Mosaic And Epipolar Plane Image
Abstract在大场景中自由平滑的漫游,这一直是图形学中一个重要的课题。近几年来兴 起了一种新的图形学绘制方法一基于图象的绘制(Image Based Rendering,以下简 称IBR),给大场景漫游的课题又注入一丝新鲜血液。 其中,同心拼图(Concentric Mosaics,以下简称CM)是一种重要的IBR绘制 方法,它通过相对简单的采集装置和比较少的数据量,可以实现用户在一个圆圈范 围内的自由平滑的漫游。当希望活动的范围更大、更自由地时候,链接的CM也就 应运而生。在这种绘制方法的支持下,用户可以在更大的范围内自由、连续、真 实、实时的漫游。同时,为了避免CM之间过渡时产生的重影现象,必须用一定的 深度信息进行校正,因此自动的从CM序列中获得深度就很重要。 本文就这个问题进行了详细的分析,提出了在CM序列图象中存在着近似的极 线平面图像(Epipolar Plane Image,以下简称EPI),而且像点在EPI图像上轨迹的 斜率和物点的深度呈近似线性关系。基于此发现,本文提出了两种从CM序列中自 动估计深度信息的方法。第一‘种方法利用EPI图像的频谱分析对场景的深度分布范 围做出估计,然后根据全光采样的原理在EPI的斜率空间均匀的分割投票箱,并对 给定窗口内频谱能量进行投票,以求得能量的最大方向,从而获得这个窗口所对应 的深度,最后组合成CM的场景深度图。第二种方法提出了深度可能性图的概念, 把深度估计的问题赋予个局平滑性约束,转换成一个最优化问题,采用动念规划的 方法进行求解,最后得到CM场景的深度图。 经过实验验证,本文的算法是有效的,并且所求得的深度图被应用在多边形深 度模型校正的链接CM系统中,得到了满意的结果。
Other AbstractOne of the ultimate goals in image-based rendering is to enable users to continuously wander around in a large environment. Image based rendering techniques, which have recently received much attention, introduce some new powerful methods for navigating in large environment. Concentric Mosaics, or CM, is one of the most important IBR methods. It provides the users nature, smooth, and real time movements in a circle, by easy capture setup and reasonable data amount. Concatenated Concentric Mosaic, or CCM, is subsequently innovated when it is needed to walk in larger scope than one CM can provide. CCM not only preserves the CM's merit by providing the users a nature, smooth and real time navigating, but also allows the user to walk in a much broader area. However, in order to prevent the annoying double image when transmitting from one CM to another, depth information has to be provided to rendering algorithm. Therefore, it becomes important to recover the depth information from CM sequences. In this paper, we claim that approximate EPI exists in CM sequence, and the directions of features' tracks on EPI have some approximate linear relationship with the features' depth. Based on this argument, we introduce two algorithms to extract depth information from CM sequences. One method uses the spectral analysis of EPI and voting technique to estimate the depth image. The other one transform the depth recovery into an optimization problem with global smoothness constraint, and solve it by dynamic program algorithm. The depth image by our algorithm has been used by the CCM system with polyhedral scene model, and it works very well.
Other Identifier570
Document Type学位论文
Recommended Citation
GB/T 7714
李寅. 从CM序列估计深度[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2000.
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
[李寅]'s Articles
Baidu academic
Similar articles in Baidu academic
[李寅]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李寅]'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.