没有合适的资源?快使用搜索试试~ 我知道了~
elas论文1
需积分: 0 0 下载量 45 浏览量
2022-08-03
12:33:08
上传
评论
收藏 3.9MB PDF 举报
温馨提示
试读
14页
elas论文1
资源详情
资源评论
资源推荐
Efficient Large-Scale Stereo Matching
Andreas Geiger
1
, Martin Roser
1
, and Raquel Urtasun
2
1
Dep. of Measurement and Control, Karlsruhe Institute of Technology
2
Toyota Technological Institute at Chicago
Abstract. In this paper we propose a novel approach to binocular stereo
for fast matching of high-resolution images. Our approach builds a prior
on the disparities by forming a triangulation on a set of support points
which can be robustly matched, reducing the matching ambiguities of
the remaining points. This allows for efficient exploitation of the dispar-
ity search space, yielding accurate dense reconstruction without the need
for global optimization. Moreover, our method automatically determines
the disparity range and can be easily parallelized. We demonstrate the
effectiveness of our approach on the large-scale Middlebury benchmark,
and show that state-of-the-art performance can be achieved with signif-
icant speedups. Computing the left and right disparity maps for a one
Megapixel image pair takes about one second on a single CPU core.
1 Introduction
Estimating depth from binocular imagery is a core subject in low-level vision as
it is an important building block in many domains such as multi-view reconstruc-
tion. In order to be of practical use for applications such as autonomous driving,
disparity estimation methods should run at speeds similar to other low-level vi-
sual processing techniques, e.g. edge extraction or interest point detection. Since
depth errors increase quadratically with the distance [1], high-resolution images
are needed to obtain accurate 3D representations. While the benefits of high
resolution imagery are already exploited exhaustively in structure-from-motion,
object recognition and scene classification, only few binocular stereo methods
deal efficiently with large images.
Stereo algorithms based on local correspondences [2, 3] are typically fast,
but require an adequate choice of window size. As illustrated in Fig. 1 this
leads to a trade-off between low matching ratios for small window sizes and
border bleeding artifacts for larger ones. As a consequence, poorly-textured and
ambiguous surfaces cannot be matched consistently.
Algorithms based on global correspondences [4–9] overcome some of the afore-
mentioned problems by imposing smoothness constraints on the disparities in the
form of regularized energy functions. Since optimizing such MRF-based energy
functions is in general NP-hard, a variety of approximation algorithms have been
proposed, e.g., graph cuts [4, 5] or belief propagation [6]. However, even on low-
resolution imagery, they generally require large computational efforts and high
2 Andreas Geiger, Martin Roser, Raquel Urtasun
Fig. 1. Low-textured areas often pose problems to stereo algorithms. Using local meth-
ods one faces the trade-off between low matching ratios (top-right, window size 5 × 5)
and border bleeding effects (bottom-left, window size 25 × 25). Our method is able to
combine small window sizes with high matching ratios (bottom-right).
memory capacities. For example, storing all messages of a one Megapixel image
pair requires more than 3 GB of RAM [10]. In these approaches, the disparity
range usually has to be known in advance, and a good choice of the regularization
parameters is crucial. Furthermore, when increasing image resolution, the widely
used priors based on binary potentials fail to reconstruct poorly-textured and
slanted surfaces, as they favor fronto-parallel planes. Recently developed meth-
ods based on higher-order cliques [7] overcome these problems, but are even more
computationally demanding.
In this paper we propose a generative probabilistic model for stereo matching,
called ELAS (Efficient LArge-scale Stereo)
3
, which allows for dense matching
with small aggregation windows by reducing ambiguities on the correspondences.
Our approach builds a prior over the disparity space by forming a triangulation
on a set of robustly matched correspondences, named ‘support points’. Since our
prior is piecewise linear, we do not suffer in the presence of poorly-textured and
slanted surfaces. This results in an efficient algorithm that reduces the search
space and can be easily parallelized. As demonstrated in our experiments, our
method is able to achieve state-of-the-art performance with significant speedups
of up to three orders of magnitude when compared to prevalent approaches; we
obtain 300 MDE/s (million disparity evaluations per second) on a single CPU
core.
2 Related work
In the past few years much progress has been made towards solving the stereo
problem, as evidenced by the excellent overview of Scharstein et al. [2]. Local
methods typically aggregate image statistics in a small window, thus imposing
3
C++ source code, Matlab wrappers and videos online at http://www.cvlibs.net
Efficient Large-Scale Stereo Matching 3
smoothness implicitly. Optimization is usually performed using a winner-takes-
all strategy, which selects for each pixel the disparity with the smallest value
under some distance metric [2]. Weber et al. [3] achieved real-time performance
using the Census transform and a GPU implementation. However, as illustrated
by Fig. 1, traditional local methods [11] often suffer from border bleeding effects
or struggle with correspondence ambiguities. Approaches based on adaptive sup-
port windows [12, 13] adjust the window size or adapt the pixel weighting within
a fixed-size window to improve performance, especially close to border discon-
tinuities. Unfortunately, since for each pixel many weight factors have to be
computed, these methods are much slower than fixed-window ones [13].
Dense and accurate matching can be obtained by global methods, which en-
force smoothness explicitly by minimizing an MRF-based energy function which
can be decomposed as the sum of a data fitting term and a regularization term.
Since for most energies of practical use such an optimization is NP-hard, approx-
imate algorithms have been proposed, e.g. graph-cuts [4, 5], belief propagation
[6]. Klaus et al. [14] extend global methods to use mean-shift color segmentation,
followed by belief propagation on super-pixels. In [15], a parallel VLSI hardware
design for belief propagation that achieves real time performance on VGA im-
agery was proposed . The application of global methods to high-resolution images
is, however, limited by their high computational and memory requirements, es-
pecially in the presence of large disparity ranges. Furthermore, models based
on binary potentials between pixels favor fronto-parallel surfaces which leads to
errors in low-textured slanted surfaces. Higher order cliques can overcome these
problems [7], but they are even more computationally demanding.
Hirschm¨uller proposed semi-global matching [16], an approach which extends
polynomial time 1D scan-line methods to propagate information along 16 orien-
tations. While reducing streaking artifacts and improving accuracy compared to
traditional methods based on dynamic programming, computational complex-
ity increases with the number of computed paths. ‘ground control points’ are
used in [17] to improve the occlusion cost sensitivity of dynamic programming
algorithms. In [18, 19] disparities are ‘grown’ from a small set of initial corre-
spondence seeds. Though these methods produce accurate results and can be
faster than global approaches, they do not provide dense matching and strug-
gle with textureless and distorted image areas. Approaches to reduce the search
space have been investigated for global stereo methods [10, 20]. However, they
mainly focus on memory requirements and start with a full search using local
methods first. Furthermore, the use of graph-cuts imposes high computational
costs particularly for large-scale imagery.
In contrast, in this paper we propose a Bayesian approach to stereo matching
that is able to compute accurate disparity maps of high resolution images at
frame rates close to real time without the need for global optimization. The
remainder of this paper is structured as follows: In Section 3 we describe our
approach to efficient large-scale stereo matching. Experimental results on real-
world datasets and comparisons to a variety of other methods on large-scale
剩余13页未读,继续阅读
坐在地心看宇宙
- 粉丝: 24
- 资源: 330
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
最新资源
- 基于Python+OpenCV对多张图片进行全景图像拼接,消除鬼影,消除裂缝+源代码+文档说明+界面截图(高分课程设计)
- 基于C++实现的全景图像拼接源码(课程设计)
- 基于SIFT特征点提取和RASIC算法实现全景图像拼接python源码+文档说明+界面截图+详细注释(95分以上课程大作业)
- 基于matlab实现眼部判别的疲劳检测系统+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
- 基于Matlab的异常姿势识别系统+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
- 隐藏文件展示工具,用来展示被病毒隐藏的文件
- 基于Matlab的图像分割系统+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
- 基于MATLAB指纹门禁GUI设计源码+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
- 基于Matlab的仪表指数识别系统霍夫曼变换+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
- 基于MATLAB 的霍夫曼变换答题卡识别带GUI界面+源代码+全部数据+文档说明+详细注释+使用说明+截图(高分课程设计)
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
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
评论0