Information Technology Reference
In-Depth Information
⎡
⎤
⎡
⎤
X
m
Y
m
Z
m
X
t
−
1
Y
t
−
1
Z
t
−
1
⎣
⎦
=
R
m
⎣
⎦
+
t
m
.
(20)
The interpolation frame for any desired view of the multi-view set can then be re-
constructed by re-projecting the 3D points onto the corresponding image plane:
⎡
⎤
x
m
y
m
=
P
desired
X
m
Y
m
Z
m
⎣
⎦
.
(21)
Intensities corresponding to foreground objects are interpolated by using the Red-
Green-Blue (RGB) values of corresponding pixels at time instant
t
1. For interpo-
lating the background regions, the segmentation maps, given in Fig. 10 are exploited
by assigning average RGB values for common background regions and assigning
RGB values from either frame for uncommon background regions. Fig. 12 shows
the resulting interpolated frame of
Akko-Kayo
sequence.
Due to the rounding effects, some parts of the foreground object remain unfilled;
thus, in order to alleviate such problems, the interpolated frame is post-processed
by a 3
x
3 median filter. Fig. 13 illustrates the resulting frame after post processing.
So far, 3D coordinates of the foreground objects, and hence, the corresponding
2D pixel locations at the interpolation frame of time instant
t
−
1
2
are calculated
−
−
using the frame at
t
1 via (20) and (21). In order to increase the quality of the
interpolation, bi-directional filling, which utilizes the intensities of frames at time
instants
t
1and
t
, is employed. For bi-directional interpolation, backward rotation
matrix
R
b
and translation vector
t
b
transforming 3D foreground coordinates from
−
Fig. 12
Interpolated frame using 3D motion information
