Information Technology Reference
In-Depth Information
Fig. 7
Sample interpolation results for
Flower Garden
sequence. From left to right, top row:
frames 318, 320 and 322 of the original sequence, middle row: corresponding interpolated
frames, bottom row: layer support maps corresponding to frames 319, 321 and 323.
4
Region-Based Multi-view FRUC by Rigid Body Motion
Model Parameter Interpolation (FRUC-RM)
The rigid object motion can exactly be modeled by using the Euclidean transforma-
tion, [32]:
⎡
⎤
⎡
⎤
X
t
Y
t
Z
t
X
t
−
1
Y
t
−
1
Z
t
−
1
⎣
⎦
=
R
⎣
⎦
+
t
,
(12)
where [
X
t
,
Y
t
,
Z
t
]
T
and [
X
t
−
1
,
Y
t
−
1
,
Z
t
−
1
]
T
represent 3D coordinates of the moving
object between successive frames,
R
denotes a 3
x
3rotationmatrixand
t
is a 3
x
1
translation matrix.
Although only image coordinates of the moving objects are available in video
sequences, unfortunately, the utilization of (12) requires 3D coordinates of these
moving objects. In order to accomplish such a goal, the calibration of the camera
capturing the scene, which defines the projection of 3D world coordinates onto 2D
pixel coordinates, as well as the depth map of the scene should be known [32]. The
following equation relates pixel coordinates with the world coordinates by the help
of a 3
x
4 projection matrix,
P
,as:
