没有合适的资源?快使用搜索试试~ 我知道了~
(Colored ICP算法)ICCV2017_Colored Point Cloud Registration Revisit
需积分: 0 0 下载量 173 浏览量
2022-08-03
15:23:49
上传
评论
收藏 4.1MB PDF 举报
温馨提示
试读
10页
(Colored ICP算法)ICCV2017_Colored Point Cloud Registration Revisited1
资源推荐
资源详情
资源评论
Colored Point Cloud Registration Revisited
Jaesik Park Qian-Yi Zhou Vladlen Koltun
Intel Labs
Abstract
We present an algorithm for aligning two colored point
clouds. The key idea is to optimize a joint photometric
and geometric objective that locks the alignment along both
the normal direction and the tangent plane. We extend a
photometric objective for aligning RGB-D images to point
clouds, by locally parameterizing the point cloud with a vir-
tual camera. Experiments demonstrate that our algorithm
is more accurate and more robust than prior point cloud
registration algorithms, including those that utilize color
information. We use the presented algorithms to enhance
a state-of-the-art scene reconstruction system. The preci-
sion of the resulting system is demonstrated on real-world
scenes with accurate ground-truth models.
1. Introduction
We are concerned with the following problem: given two
roughly aligned three-dimensional point clouds, compute a
tight alignment between them. This is a well-known prob-
lem in computer vision, computer graphics, and robotics.
The problem is typically addressed with variants of the ICP
algorithm [
1, 3, 31]. The algorithm alternates between find-
ing correspondences and optimizing an objective function
that minimizes distances between corresponding points. A
common failure mode of ICP is instability in the presence of
smooth surfaces [
14, 46]. The alignment can slip when ge-
ometric features do not sufficiently constrain the optimiza-
tion.
This ambiguity can be alleviated if the points are asso-
ciated with color. This is often the case. Modern depth
cameras commonly produce pairs of depth and color im-
ages. Many industrial 3D scanners are also equipped with
synchronized color cameras and provide software that as-
sociates color information with the 3D scans. Multi-view
stereo pipelines reconstruct colored point clouds from im-
age collections [
8, 13, 39]. Considering color along with
the geometry can increase the accuracy of point cloud reg-
istration.
The standard formulation for integrating color into geo-
metric registration algorithms is to lift the alignment into
a higher-dimensional space, parameterized by both posi-
tion and color. Typically, correspondences are established
in a four- or six-dimensional space rather than the physical
three-dimensional space [
21, 22, 27, 28]. This is an ele-
gant approach, but it is liable to introducing erroneous cor-
respondences between points that are distant but have sim-
ilar color. These correspondences can pull away from the
correct solution and prevent the method from establishing a
maximally tight alignment.
In this work, we develop a different approach to aligning
colored point clouds. Our approach establishes correspon-
dences in the physical three-dimensional space, but defines
a joint optimization objective that integrates both geometric
and photometric terms. A key challenge is that color is only
defined on discrete points in the three-dimensional space.
To optimize a continuous joint objective, we need to define
a continuous and differentiable photometric term, the gradi-
ent of which indicates how color varies as a function of posi-
tion. This is challenging because unstructured point clouds
do not provide a natural parameterization domain. We build
on dense and direct formulations for RGB-D image align-
ment, which use the two-dimensional image plane as the
parameterization domain [
35, 25, 44, 40]. To define a pho-
tometric objective for point cloud alignment, we introduce
a virtual image on the tangent plane of every point, which
provides a local approximation to the implicit color vari-
ation. Using this construct, we generalize the photometric
objectives used for RGB-D image alignment to unstructured
point cloud alignment. The resulting photometric objective
is integrated with a geometric objective defined using the
same virtual image planes. This enables efficient joint pho-
tometric and geometric optimization for point cloud align-
ment. Our formulation unifies RGB-D image registration
and colored point cloud registration. We show that our al-
gorithm achieves tighter alignment than state-of-the-art reg-
istration algorithms, including those that use color informa-
tion.
Our primary contribution is a new approach to colored
point cloud registration. Beyond this, we make two sup-
porting contributions. Since point cloud registration plays
a central role in high-fidelity scene reconstruction, we have
used the presented algorithms to enhance a state-of-the-art
1
143
资源评论
蔓誅裟華
- 粉丝: 19
- 资源: 304
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功