基于多幅深度图像的三维重构技术  

Depth images-based approach for 3D reconstruction

作　　者：章勤[1] 虢伟[1] 

机构地区：[1]华中科技大学计算机科学与技术学院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2008年第2期88-91,共4页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家高技术研究发展计划资助项目(2006AA02Z347)

摘　　要：提出基于多幅深度图像的三维重构方法.采用6幅固定视点采样图像,分别代表三维空间中立方体包围盒的6个面;通过对这些二维的采样集合进行三维信号还原,生成一个空间采样样本,可以在视点任意改变的情况下,输出近似的重构图像.该方法由于采样集合固定,经过采样集筛选,背景剔除筛选和棱台视锥体裁减后,可以极大减少待变换的样本集合,采用z-buffer算法解决图像折叠问题.当视点在采样点附近任意变换时,可以重构出比较满意的结果,而且该方法与场景复杂度无关.在纯软件模式下在普通PC机上达到20帧/s的速度,在通过硬件加速情况下可以达到30帧/s.Depth images-based approach for 3D reconstruction which employs six pictures with the fixed viewer is put forward in this essay. They stand for the faces of the cube model respectively. For converting 2D sample to 3D signal, the image of any viewing can be reconstructed. Through the fixed sampling of scene, it can get 20fps in a common pc after a series of filtration by software and can get up to 30fps by hardware. The z-buffer is used to solve the image fold. The better outcome can be get when the viewing is not very difference to the sample viewing. And it is independent to the complexity of scene.

关 键 词：多幅深度图像 三维重构 立方体固定采样 筛选 场景复杂度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]