中图分类号：TP391.7论文编号：102870515-S193学科分类号：082503
硕士学位论文
基于双目立体视觉裸眼3D
显示技术研究
研究生姓名李超
学科、专业航空宇航制造工程
研究方向数字化设计与制造
指导教师沈建新教授
南京航空航天大学
研究生院机电学院
二О一五年三月
Nanjing University of Aeronautics and Astronautics
The Graduate School
College of Mechanical and Electrical Engineering
Research on Technology of Naked-eye 3D Display Based on Binocular Stereo Vision
A Thesis in
Aeronautical and Astronautics Science and Technology
by
Li Chao
Advised by
Prof. Shen Jianxin
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Master of Engineering
March, 2015
承诺书
本人郑重声明：所呈交的学位论文，是本人在导师指导下，独立进行研究工作所取得的成果。尽我所知，除文中已经注明引用的内容外，本学位论文的研究成果不包含任何他人享有著作权的内容。对本论文所涉及的研究工作做出贡献的其他个人和集体，均已在文中以明确方式标明。
本人授权南京航空航天大学可以有权保留送交论文的复印件，允许论文被查阅和借阅,可以将学位论文的全部或部分内容编入有关数据库进行检索，可以采用影印、缩印或其他复制手段保存论文。
(保密的学位论文在解密后适用本承诺书)
作者签名：
日期：
南京航空航天大学硕士学位论文
摘要
立体显示技术很好地解决了传统的二维显示技术缺少场景三维信息这一问题，能让人眼感受到更加全面和真实的立体影像，现已广泛的应用于医学、教育、军事、商业以及娱乐等众多领域，对基于双目立体视觉的立体显示技术研究具有十分重要的价值和现实意义。
双目立体视觉技术是计算机视觉领域的一个重要研究方向，它基于视差原理能精确的获取空间物体表面的三维信息。通过两个不同位置的摄像机拍摄同一场景，并从中恢复深度信息来模拟人类双眼对三维信息的测量。在立体显示技术中，非裸眼的立体成像技术都需要佩带如偏振眼镜、快门式眼镜、互补色眼镜等辅助工具，不能完全解放人眼，因此这种方式在许多场合并不适用。选择基于双目立体视觉原理的裸眼3D显示方法是一个最佳选择，它不再需要佩带眼镜，观察者在观看的屏幕同时不受任何限制，灵活性更强，而且三维场景信息相对准确。
从平面图像到提取包含深度信息的视差图像，以及转化到人眼能够感受到具有纵深感的立体图像，综合了摄像机标定校正、立体匹配、DIBR技术、裸眼3D显示技术等一系列新型技术。论文重点进行了双目立体视觉技术中摄像机标定校正、立体匹配算法研究以及DIBR技术的仿真等，主要研究内容包括：
1、双目立体视觉技术及平面3D技术概述。包括双目立体视觉技术、平面3D显示技术的原理、理论、研究现状以及3D技术在医疗手术领域的应用。
2、双目立体视觉系统模型分析。结合现有立体显示技术，介绍3D显示器的光学原理以及人眼获得具有真实立体感的立体视觉成像原理。引入视差概念，构造了双目立体视觉的几何模型，建立场景中三维信息的数学关系。
3、摄像机标定校正技术研究。分析了摄像机的成像模型及其特点，研究了摄像机标定方法的坐标系转换、数学原理。主要研究了基于OpenCV的标定和校正方法，采用张正友经典标定方法进行摄像机的单目及双目标定，获得摄像机的内外参数，最后通过Bouguet算法对图像进行校正。
4、基于小波变换与噪声可见度函数的立体匹配算法研究。针对现有匹配方法进行了对比分析，引入了小波变换与噪声可见度函数，对基于图像分割的立体匹配算法进行改进，基于小波变换与噪声可见度函数的立体匹配算法将图像分解为多个子带，并对各个子带图像进行阈值处理，提高图像的匹配质量，最终利用人类视觉系统模型中的噪声可见度函数融合各个子带的匹配结果获得最终的视差图像。仿真结果表明基于小波变换与噪声可见度函数的立体匹配算法，与其他算法相比，能获得精度更高的视差图像。
在上述研究的基础上，进行了基于小波变换与噪声可见度函数立体匹配算法及DIBR技术
i
基于双目立体视觉裸眼3D显示技术研究
的仿真实验，并基于VC平台和OpenCV 视觉函数库来完成相关软件设计；利用自己搭建的双目立体视觉系统硬件平台完成图像采集、摄像机标定校正、提取深度图像及图像合成等相关实验，实验结果表明，所获得的深度图像误差较小能有效反应场景信息。
关键词：双目立体视觉，平面3D显示，裸眼3D，摄像机标定，立体匹配，离散小波变换，噪声可见度函数，图像分割，OpenCV，DIBR
ii
南京航空航天大学硕士学位论文
Abstract
The traditional two-dimensional display technology lacks depth information, and this problem is solved well by 3d display technology, it can make us feel more comprehensive and true stereo feelings. The 3d display technology has been widely used in medicine, education, military, commercial, entertainment and many other fields. The research on binocular stereo vision has important value and real significant.
Binocular stereo vision technology is one of the important fields in computer vision technology, It can obtain 3d(three-dimensional) information of the object in the scene accurately based on parallax principle. Binocular stereo vision directly simulate man’s eyes processing information manners to compute the three-dimensional information, that is, the use of two cameras shooting the same scene from different locations, and recover the three-dimensional information. In non-naked-eye 3d display technology we need wear polarized glasses, shutter glasses, complementary color glasses or other auxiliary tools, therefore it cannot liberate our eyes, so this way is not applicable in many occasions. The naked-eye 3d technology based on binocular stereo vision is one of the best choices, it no longer need to wear glasses, observers can watch the screen without any restrictions, this technology is more flexible, and it can provide observers with accurate scene information.
Extracting disparity images which contain depth information from planar images and transforming depth images to stereo images, it needs a series of new technologies, such as camera calibration, stereo matching, DIBR technology and autostereoscopic display and so on. This paper mainly focuses on camera calibration and correction technology, stereo matching algorithm and simulation of DIBR technology, the main research contents are listed as follows:
1 The overview of binocular stereo vision technology and planar 3d display technology. This part includes principles and theories of binocular stereo vision and planar 3d display technology, their domestic and foreign research at present and past, and their applications in medical surgery.
2 The analysis of the models of binocular stereo vision system. In this part we express the optical principle of 3d display and the imaging theory of stereo vision with the reality feelings, combined with the introduction of the different 3d display technologies, introduce the concept of disparity, structure the geometric model of binocular stereo vision and the mathematical relationship of the scene information.
iii
基于双目立体视觉裸眼3D显示技术研究
iv 3 The research of the camera calibration and correction. Analyze the features of the imaging
models of camera, the main research includes the coordinate transformation and mathematical principles of the traditional camera calibration methods. Calibrate the binocular cameras by Zhangzhengyou method based on OpenCV platform, obtain the intrinsic and external parameters of the cameras, and on the basis of calibration we correct the images by BOUGUET algorithm.
4 The research on stereo matching algorithm based on wavelet transform and noise visibility function. Compare and analyze the present matching methods, found the stereo matching based on image segmentation has great advantages, this paper proposes a novel method of stereo matching, which decomposes object images into a set of sub-bands by discrete wavelet transform, then modifies each sub-band image, and improves the matching quality of each sub-band image by implementing discrete inverse wavelet transform. On this basis we estimate disparity of each modified image, and merge different sub-band disparity maps into a final disparity map using HVS model. Experimental results demonstrate the proposed algorithm achieve a high-precision than the existing algorithm.
Accomplished the simulation of the stereo matching algorithm based on wavelet transform and noise visibility function and DIBR technology according to the above research results. The software was designed on VC platform by OpenCV vision library. And at last accomplished the experiments of images capturing, the camera calibration and correction, depth image generating and stereo images synthesizing on the binocular stereo vision hardware platform, the results show that the error of the depth image is small and it can reveal the scene information effectively.
