多视图立体重建算法的比较与评估

需积分: 50 144 浏览量更新于2024-09-08 收藏 1.71MB PDF 举报

"这篇文章是关于多视图立体重建算法的比较和评估，是MVS领域的权威综述，详细探讨了近年来MVS技术的发展。作者包括Steven M. Seitz、Brian Curless、James Diebel、Daniel Scharstein和Richard Szeliski等知名学者，来自华盛顿大学、斯坦福大学和微软研究院。" 在计算机视觉领域，多视图立体重建（Multi-View Stereo，简称MVS）是一种关键技术，用于从多个不同视角的图像中恢复场景的三维几何信息。这篇论文提供了一个定量的比较，对几种不同的MVS重建算法进行了深入分析。在此之前，由于缺乏合适的、带有已知真实三维模型的多视图图像数据集，直接的算法比较变得困难。论文首先概述了MVS算法，并利用一种分类法，根据它们的关键特性对这些算法进行了定性比较。接着，作者详细介绍了他们获取和校准高精度多视图图像数据集的过程，这些数据集带有精确的地面实况（3D形状模型）。此外，他们还提出了一种评估方法，旨在为MVS算法的性能提供公正的标准。最后，论文展示了在六个基准数据集上对最新先进MVS重建算法进行定量比较的结果。这些基准数据集、评估细节以及提交新模型的指南都可以在线获取，网址为http://vision.middlebury.edu/mview。多视图立体重建算法的比较通常关注以下几个关键方面： 1. **匹配质量**：算法如何在不同视图间寻找对应点，这直接影响到三维重建的准确性。 2. **稀疏到密集**：从初始的特征匹配到构建稠密深度图的过程，算法的效率和精度是评价的重点。 3. **优化策略**：如何处理遮挡、光照变化和噪声，以提高重建的鲁棒性。 4. **计算效率**：在保持高精度的同时，算法运行时间和内存消耗也是衡量其实用性的关键因素。 5. **后处理**：包括深度图融合、去噪和空洞填充等，以提升最终的3D模型质量。通过对这些算法的全面比较，研究者和开发者可以更好地理解各种方法的优势和局限性，从而推动MVS技术的进一步发展。这项工作对于研究者选择合适的算法或改进现有算法具有重要指导意义，同时也为未来算法设计提供了基准和挑战。

A Comparison and Evaluation of Multi-View Stereo Reconstruction Algorithms

Steven M. Seitz Brian Curless

University of Washington

James Diebel

Stanford University

Daniel Scharstein

Middlebury College

Richard Szeliski

Microsoft Research

Abstract

This paper presents a quantitative comparison of several

multi-view stereo reconstruction algorithms. Until now, the

lack of suitable calibrated multi-view image datasets with

known ground truth (3D shape models) has prevented such

direct comparisons. In this paper, we ﬁrst survey multi-view

stereo algorithms and compare them qualitatively using a

taxonomy that differentiates their key properties. We then

describe our process for acquiring and calibrating multi-

view image datasets with high-accuracy ground truth and

introduce our evaluation methodology. Finally, we present

the results of our quantitative comparison of state-of-the-art

multi-view stereo reconstruction algorithms on six bench-

mark datasets. The datasets, evaluation details, and in-

structions for submitting new models are available online

at http://vision.middlebury.edu/mview.

1. Introduction

The goal of multi-view stereo is to reconstruct a com-

plete 3D object model from a collection of images taken

from known camera viewpoints. Over the last few years,

a number of high-quality algorithms have been developed,

and the state of the art is improving rapidly. Unfortunately,

the lack of benchmark datasets makes it difﬁcult to quan-

titatively compare the performance of these algorithms and

to therefore focus research on the most needed areas of de-

velopment.

The situation in binocular stereo, where the goal is to

produce a dense depth map from a pair of images, was until

recently similar. Here, however, a database of images with

ground-truth results has made the comparison of algorithms

possible and hence stimulated an even faster increase in al-

gorithm performance [1].

In this paper, we aim to rectify this imbalance by pro-

viding, for the ﬁrst time, a collection of high-quality cal-

ibrated multi-view stereo images registered with ground-

truth 3D models and an evaluation methodology for com-

paring multi-view algorithms.

Our paper’s contributions include a taxonomy of multi-

view stereo reconstruction algorithms inspired by [1] (Sec-

tion 2), the acquisition and dissemination of a set of

calibrated multi-view image datasets with high-accuracy

ground-truth 3D surface models (Section 3), an evalua-

tion methodology that measures reconstruction accuracy

and completeness (Section 4), and a quantitative evaluation

of some of the currently best-performing algorithms (Sec-

tion 5). While the current evaluation only includes meth-

ods whose authors were able to provide us their results by

CVPR ﬁnal submission time, our datasets and evaluation

results are publicly available [2] and open to the general

community. We plan to regularly update the results, and

publish a more comprehensive comparative evaluation as a

full-length journal publication.

We limit the scope of this paper to algorithms that re-

construct dense object models from calibrated views. Our

evaluation therefore does not include traditional binocular,

trinocular, and multi-baseline stereo methods, which seek

to reconstruct a single depth map, or structure-from-motion

and sparse stereo methods that compute a sparse set of fea-

ture points. Furthermore, we restrict the current evaluation

to objects that are nearly Lambertian, which is assumed by

most algorithms. However, we also captured and plan to

provide datasets of specular scenes and plan to extend our

study to include such scenes in the future.

This paper is not the ﬁrst to survey multi-view stereo

algorithms; we refer readers to nice surveys by Dyer [3]

and Slabaugh et al. [4] of algorithms up to 2001. How-

ever, the state of the art has changed dramatically in the last

ﬁve years, warranting a new overview of the ﬁeld. In addi-

tion, this paper provides the ﬁrst quantitative evaluation of

a broad range of multi-view stereo algorithms.

2. A multi-view stereo taxonomy

One of the challenges in comparing and evaluating

multi-view stereo algorithms is that existing techniques

vary signiﬁcantly in their underlying assumptions, operat-

ing ranges, and behavior. Similar in spirit to the binoc-

ular stereo taxonomy [1], we categorize existing meth-

ods according to six fundamental properties that differen-

tiate the major algorithms: the scene representation, photo-

consistency measure, visibility model, shape prior, recon-

struction algorithm, and initialization requirements.

下载后可阅读完整内容，剩余7页未读，立即下载

小玄玄

粉丝: 14
资源: 3

多视图立体重建算法的比较与评估

MVS算法比较与评估

Multi-View Stereo.pdf

Research on Establishment and Comparison of Blue-tongue-virus Multi-clone Antibody C-ELISA (2014年)

Fusion and comparison of multi-granulation rough sets

Comparison Theorems of the multi-dimensional BDSDEs and Applications

Comparison-of-Disparity-Estimation-Algorithms:实现简单的块匹配、动态规划的块匹配和使用置信传播算法的立体匹配-matlab开发

Comparison of high-energy multi-pass Ti:sapphire amplifiers with a different Ti-dopant concentration

An all-optical comparison scheme between two multi-bit data with optical nonlinear material

Comparison-of-modin-and-pandas-df

Comparison of end-pumped and multi-point pumped Yb^3^+-doped gain guided and index antiguided fiber laser

最新资源