Understanding multiperspective distortions: Theory and applications.

详细信息   

• 作者：Ding ; Yuanyuan.
• 学历：Doctor
• 年：2010
• 导师：Yu, Jingyi,eadvisorYu, Jingyiecommittee memberKambhamettu, Chandraecommittee memberRasmussen, Christopherecommittee memberYang, Ruigangecommittee member
• 毕业院校：University of Delaware
• Department：Department of Computer and Information Sciences
• ISBN：9781124240701
• CBH：3423318
• Country：USA
• 语种：English
• FileSize：13742035
• Pages：191

文摘

A multi-perspective camera combines what is seen from several viewpoints into a single image. It provides a potentially advantageous imaging system for understanding the structure of observed scenes. However, images in a multi-perspective camera also exhibit unique distortions such as the curving of lines, apparent stretching and shrinking, and duplicated projections of a single point. In this dissertation, I investigate how to systematically analyze multi-perspective distortions and more importantly, how to effectively use these distortions to solve traditionally challenging computer vision problems. My solution is to use the recently proposed General Linear Cameras GLCs) as the primitive imaging model for analyzing arbitrary multi-perspective cameras. To characterize the line curving distortions, I derive a curved-linear image CLI) theory to characterizes all possible curves that would be observed in the GLC. I show that a 3D line will appear as a conic in any GLCs. For a fixed GLC, the conic type is invariant to the position and orientation of the line and is determined by the GLC parameters. Furthermore, CLIs under single reflection/refraction can only be lines or hyperbolas. Based on our new theory, I develop an efficient algorithm to use multiple CLIs to recover the GLC intrinsic parameters. I also develop an epsilon stereo theory for stereo matching and fusion. An epsilon stereo pair consists of two GLC images with a slight vertical parallax. I show many multi-perspective camera pairs that do not satisfy the stereo constraint can still form epsilon stereo pairs. I develop an epsilon stereo graph-cut algorithm for generating a pseudo depth map. Finally, I derive a novel multi-perspective defocus theory to characterize defocus distortions in general multi-perspective cameras. I introduce a Thin Lens Operator to study how ray geometry transform through the thin lens and couple it with the aperture operator to derive the closed-form multi-perspective point-spread-function PSF). Our theoretical components lead to a variety of novel computer vision and graphics applications including distortion corrections, 3D specular surface reconstructions, and a new class of imaging and projection systems. Specifically, I develop an interactive distortion correction system, a novel shape-from-distortion framework for recovering nearflat specular surface, a spherical mirror array for light field acquisition and multi-view reconstruction, and a catadioptric projection system that can offer an unprecedented level of flexibility in aspect ratio, size, and field of view. These new applications and systems have the potential to fundamentally change the way people think about images, cameras, projectors, and etc.