[其它] Fusion of Depth Maps with Multiple Scales

[复制链接]

晃晃

1023 主题	3 听众	359 积分

设计实习生

Rank: 2

纳金币: 335582
精华: 0

电梯直达

楼主

发表于 2011-12-28 09:10:16 |只看该作者 |倒序浏览

Fusion of Depth Maps with Multiple Scales

Simon Fuhrmann Michael Goesele

TU Darmstadt TU Darmstadt

Abstract

Multi-view stereo systems can produce depthmaps with large varia-

tions in viewing parameters, yielding vastly different sampling rates

of the observed surface. We present a new method for surface re-

cons***ction by integrating a set of registered depth maps with dra-

matically varying sampling rate. The method is based on the con-

s***ction of a hierarchical signed distance ﬁeld represented in an in-

complete primal octree by incrementally adding triangulated depth

maps. Due to the adaptive data s***cture, our algorithm is able to

handle depth maps with varying scale and to consistently represent

coarse, low-resolution regions as well as small details contained in

high-resolution depth maps. A ﬁnal surface mesh is extracted from

the distance ﬁeld by cons***ction of a tetrahedral complex from the

scattered signed distance values and applying the Marching Tetra-

hedra algorithm on the partition. The output is an adaptive triangle

mesh that seamlessly connects coarse and highly detailed regions

while avoiding ﬁlling areas without suitable input data.

CR Categories: I.3.5 [Computer Graphics]: Computational Ge-

ometry and Object Modeling—Surface Recons***ction I.3.6 [Com-

puter Graphics]: Methodology and Techniques—Graphics data

s***ctures and data types

Keywords: multi-scale depth map fusion, multi-view stereo depth

maps, depth map integration, hierarchical signed distance ﬁeld, sur-

face recons***ction, marching tetrahedra

1 Introduction

Surface recons***ction is an important problem with huge practi-

cal applications and a long history in computer graphics. The goal

is to build high quality 3D surface representations from captured

real-word data. Important applications include the preservation of

cultural heritage, model reverse engineering, and prototyping in the

multi-media industry. Typical inputs to surface recons***ction algo-

rithms are either unorganized points or more s***ctured data such as

depth maps. In this work we will focus on the latter kind of data,

which is produced by range scanners and some multi-view stereo

algorithms. To fully capture an object of interest, multiple overlap-

ping depth maps are necessary, each covering parts of the object

surface. In a general acquisition framework, these depth maps need

to be aligned into a common coordinate system and fused into a

single, non-redundant surface representation. This process is called

the integration or fusion of depth maps.

One source of depth maps are multi-view stereo (MVS) systems,

which recently attained renewed interest [Seitz et al. 2006]. These

algorithms recons***ct the scene geometry from photographs of the

scene by regaining the 3D information lost during capture. Cur-

rent s***cture-from-motion systems [Snavely et al. 2008] are able

to recover the camera parameters of thousands of photographs un-

der very uncontrolled conditions. This enables modern MVS algo-

rithms to make use of the massive amount of Internet imagery for

geometry recons***ction [Goesele et al. 2007; Agarwal et al. 2009;

Frahm et al. 2010].

全文请下载附件：