摘要
首先确立单相机加单平面镜的体系结构,然后研究在该体系下实现三维重构的基本理论和基本方法,具体探讨了以下4个关键问题:(1)手边沿的提取;(2)对应关系的获取;(3)3D重构的基本方法;(4)校准算法.通过揭示出空间物点在像平面上的投影、该物点的对称点在同一像平面上的投影、镜面以及该物点本身这四者之间的关系,得到三维重构的新方法,既便于理论分析,又便于程序设计;既使校准过程简单易行,又保证了三维重构的精度.
Rendering and displaying of a computer operator's hands in real time based on computer vision technique is one of the kernel problems in human-computer systems and virtual reality systems, which has extensive and wide applications in virtual making systems and virtual museum systems and so on. In this paper, authors adopt an effective mainframe with a camera and a mirror, under which some basic theories of a 3-D hand reconstruction are studied. It is focused on the following problems (1) how to abstract silhouettes of hand gestures from 2-D images, attaining feature points; (2) how to obtain the correspondences from feature points, (3) how to reconstruct a operator's hand using the correspondences, and (4) how to calibrate the camera, obtaining the necessary parameters. By analyzing the inherent relationship among the image plane, the mirror plane and the correspondences, this paper puts forward a novel way for 3D hand reconstruction, which not only facilitates theoretically analyses for further research, but also make it easy to reconstruct a 3D object with high accuracy. Several experimental results are presented in the last section, illustrating the satisfactory features of the above algorithm.
出处
《计算机学报》
EI
CSCD
北大核心
2006年第3期448-456,共9页
Chinese Journal of Computers
基金
国家自然科学基金(60473103
60473127)资助.
关键词
对应关系
人手重构
人机交互
三维重构
correspondences
hand reconstruction
human-computer interaction
reconstruction