基于双目立体视觉的人脸三维重建关键技术研究
发布时间：2019-02-25 19:06
【摘要】：人脸是非常重要而又复杂的交流通道,在人类日常生活中起着十分重要的作用,而人脸重建提供了表现和处理人脸信息的有效手段,因此三维人脸建模是一项十分有意义的工作。双目立体视觉模拟人类双眼感知场景深度信息,从而实现对三维场景的重建。本文正是利用双目立体视觉的这一原理来完成对人脸的三维重构。具体完成了以下几方面工作:(1)以张正友标定法为理论基础,对摄像机标定技术进行了深入研究。(2)分析了立体匹配中的约束条件,并通过双目立体视觉极线几何结构对图像进行校正。针对现有的立体匹配算法进行了分析。在此基础上,重点对传统的归一化互相关匹配算法进行了改进,提出了基于递推增量的双向归一化互相关匹配算法。为进一步去除误匹配,本文通过对双目立体视觉模型的分析提出了一种双目立体视觉异面直线模型。利用该模型求得匹配点在双目系统中的距离参数,可对误匹配点对有效的甄别。(3)首先应用一种基于区域生长的视差图非法点填补方法对初始视差图进行填补,而后利用亚像素级插值算法对视差图进行平滑插值运算,最终得到完整、平滑的视差图。通过优化后的视差图得到人脸三维散乱点云信息。利用三角剖分方法实现人脸的三维重构。(4)利用HALCON与MFC的联合编程,开发一个便捷的双目立体视觉三维人脸重建软件,验证了我们所设计的人脸系统及匹配技术是可行有效的。
[Abstract]:Face is a very important and complex communication channel, and plays a very important role in human daily life. Face reconstruction provides an effective means to represent and process face information. Therefore, 3D face modeling is a very meaningful work. The binocular stereo vision simulates the depth information of the human binocular perception scene, thus realizing the reconstruction of the three-dimensional scene. In this paper, the principle of binocular stereo vision is used to reconstruct three-dimensional face. The following works are accomplished: (1) based on Zhang Zhengyou's calibration method, the camera calibration technology is deeply studied. (2) the constraint conditions in stereo matching are analyzed. The image is corrected by polar geometric structure of binocular stereo vision. The existing stereo matching algorithms are analyzed. On this basis, the traditional normalized cross-correlation matching algorithm is improved, and a two-way normalized cross-correlation matching algorithm based on recursive increment is proposed. In order to remove the mismatching, this paper presents a binocular stereo vision model based on the analysis of the Binocular Stereo Vision (Binocular Vision), which is a straight line model of binocular stereo vision. The distance parameters of matching points in binocular system can be obtained by using this model, which can be used to distinguish the mismatched points effectively. (3) A region-based disparity map illegal point filling method is first used to fill the initial disparity map. Then the sub-pixel level interpolation algorithm is used to smooth the disparity image, and finally the complete and smooth disparity map is obtained. The 3D scattered point cloud information of face is obtained by the optimized parallax map. The method of triangulation is used to realize the 3D face reconstruction. (4) A convenient 3D face reconstruction software based on binocular stereo vision is developed using the joint programming of HALCON and MFC. The results show that the face system and the matching technique are feasible and effective.
【学位授予单位】：沈阳航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
[Abstract]:Face is a very important and complex communication channel, and plays a very important role in human daily life. Face reconstruction provides an effective means to represent and process face information. Therefore, 3D face modeling is a very meaningful work. The binocular stereo vision simulates the depth information of the human binocular perception scene, thus realizing the reconstruction of the three-dimensional scene. In this paper, the principle of binocular stereo vision is used to reconstruct three-dimensional face. The following works are accomplished: (1) based on Zhang Zhengyou's calibration method, the camera calibration technology is deeply studied. (2) the constraint conditions in stereo matching are analyzed. The image is corrected by polar geometric structure of binocular stereo vision. The existing stereo matching algorithms are analyzed. On this basis, the traditional normalized cross-correlation matching algorithm is improved, and a two-way normalized cross-correlation matching algorithm based on recursive increment is proposed. In order to remove the mismatching, this paper presents a binocular stereo vision model based on the analysis of the Binocular Stereo Vision (Binocular Vision), which is a straight line model of binocular stereo vision. The distance parameters of matching points in binocular system can be obtained by using this model, which can be used to distinguish the mismatched points effectively. (3) A region-based disparity map illegal point filling method is first used to fill the initial disparity map. Then the sub-pixel level interpolation algorithm is used to smooth the disparity image, and finally the complete and smooth disparity map is obtained. The 3D scattered point cloud information of face is obtained by the optimized parallax map. The method of triangulation is used to realize the 3D face reconstruction. (4) A convenient 3D face reconstruction software based on binocular stereo vision is developed using the joint programming of HALCON and MFC. The results show that the face system and the matching technique are feasible and effective.
【学位授予单位】：沈阳航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 安妍s，
本文编号：2430446
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2430446.html
