15. Photo-consistency measure
?? Scene space
–?シーン中の点を各カメラに投影しPhoto-
consistencyを計算
–?Photo-consistencyはSSDやNCCで計算するこ
とが一般的
15
538 Computer Vision: Algorithms and Applications (Septemb
p
x1
x0
(R,t)
p∞
e1e0c0 c1
epipolar plane
p
(R,t)
c0
epipolar
lines
x
0
e0 e1
l0
16. Photo-consistency measure
?? Image space
–?カメラ画像をシーンに投影してPhoto-
consistencyを計算
16
11.1 Epipolar geometry 541
Virtual camera
d
x
y
Input image k
u
v
Homography:
u = H x
x
y
k
d
k
(a) (b)
20. Initialization requirements
?? 初期化の要件
–?Rough Bounding Box or Volume
?? Space carving
?? Level-set (質の高い初期値が必要)
–?Foreground/background segmentation
?? silhouette
–?Range of disparity or depth values
?? Image-space algorithm
20
21. Benchmark Datasets
21
bird dogs
lti-view datasets with laser-scanned 3D models.
317 camera positions and orientations for the temple
gaps are due to shadows. The 47 cameras correspond-
g dataset are shown in blue and red, and the 16 sparse
only in red.
at serves as an initial estimate of scene geom-
31,47,48].
tion of 640 × 480 pixels att
arm. At this resolution, a pix
0.25mm on the surface of th
10cm × 16cm × 8cm, and th
The system was calibrated
tion grid from 68 viewpoints
[61] to compute intrinsic an
these parameters, we compu
and rotational offset relative t
abling us to determine the cam
as a function of any desired a
The target object sits on
center of the gantry sphere an
lights. Because the gantry c
certain viewpoints, we double
two different arm con?gurat
images. After shadowed im
we obtained roughly 80% cov
resulting images, we created
corresponding to a full hemis
temple temple model
temple
dino
カメラ配置??47視点
カメラ解像度?640x480
Temple
10x16x8 cm
Dino
7x9x7 cm
26. 最先端のMVS研究例
?? "Towards Internet-scale Multi-view Stereo "
–? CVPR 2010
–? 法線付きの点群として3次元復元
–? 大規模MVS
26
cale Multi-view Stereo
Steven M. Seitz1,2
Richard Szeliski3
Washington 3
Microsoft Research
Figure 1. Our dense reconstruction of Piazza San Marco (Venice)
Pizza San Marco
(Venice)
視点数 : 13,703
点群数 : 27,707,825