基于单张低精度深度图的实时精确三维曲面重建

项目名称： 基于单张低精度深度图的实时精确三维曲面重建

项目编号： No.61202229

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 范涵奇

作者单位： 北方工业大学

项目金额： 23万元

中文摘要： 投影增强现实系统目前已经得到了广泛的应用，但是一些重要的应用仍然需要使用昂贵的电子传感器。使用电子传感器的方法存在诸多缺陷，并且无法解决物体自身发生形变时的投影问题。解决这些问题的关键是实时、精确地重建物体表面形状。虽然曲面重建是计算机图形学的基础研究内容之一，但是目前没有方法能够实时精确地重建三维曲面。本项研究拟开发一个基于单个深度相机和几个彩色相机的实时精确的曲面重建算法来解决该问题，并预备开展三方面的研究：首先，研究深度数据的噪声概率分布，利用数据的冗余信息消除部分噪声；其次，研究利用粗糙的深度值提高MVS配对像素查找速度、提升配对的准确性，再结合MVS的结果去除深度图中的噪声、提高深度值的精度； 再次，研究低分辨率的深度图上采样到高分辨率；最后，研究点云的高速三角化问题。

中文关键词： 曲面重建；深度图；去噪；配准；上采样

英文摘要： Projector Augmenting Reality System has now been widely used, but some important applications still need to use expensive electronic sensors. Such method of using electronic sensors has many weaknesses, and it cannot solve the problem of projecting image to the deforming object surface. The key to solve these two problems is reconstructing the object surface real-time and accurately. Surface reconstruction is one of the basic research topics of computer graphics, however, there is currently no method to real-time accurately reconstruct a three dimensional surface. This research is intended to provide a real-time accurate surface reconstruction algorithm based on single depth camera and several color cameras, and prepares researching three areas: Firstly, research the probability distribution of noise in depth data, and denoise the data by exploiting the redundancy information; Secondly, accelerate searching matching pixel and improve the accuracy of MVS algorithm by taking advantage of the coarse noisy depth values, and then eliminate the noise in the depth map and improve the precision of depth values; Thirdly, research low resolution depth map up-sampling algorithm; Lastly, research fast triangularization algorithm on point cloud.

英文关键词： surface reconstruction；depth map；denoising；calibration；upsampling