Information Technology Reference
In-Depth Information
ROI of the left image and the initial ROI depth image do not match correctly on
the region of ROI boundaries. The main reason of the mismatch is the slight incor-
rectness of the camera calibration. We solve the mismatch problem using image
segmentation for the left image. In order to correctly detect ROI of the left image,
we overlap the color-segmented left image onto the initial ROI depth image. Then,
we measure the color segment set for ROI from color segments of the left image by
n
(
A
(
s
))
⎧
1
if
i
≥
0
.
⎪
⎨
R
(
s
)
=
(5)
n
(
s
)
i
i
⎪
⎩
0
otherwise
where
R
(
s
i
) indicates whether the
i
th
color segment
s
i
of the color segmented left
image is included in ROI of the left image or not. When
R
(
s
i
) is 1, the correspond-
ing color segment is included in the color segment set for ROI. The term of
n
(
s
i
) is
the total count of pixels in
s
i
, and
n
(
A
(
s
i
)) is the total count of pixels on the region
of initial ROI depth image
A
(
s
i
) that is matched with the region of
s
i
. Figure 5(a)
and Figure 5(b) show the left image and its color segment set for ROI,
respectively.
(b) Color segment set for ROI
(a) Left image
(d) ROI enhanced depth image
(c) ROI depth image
Fig. 5
ROI enhanced depth image generation
After ROI detection, we refine the initial ROI depth image from the color seg-
ment set by eliminating outside pixels on the former with comparison to the letter.
Then, we fill holes in the ROI depth image with the pixels generated by linearly
interpolating with their neighboring pixels [17]. The hole-filling algorithm is per-
formed by the unit of a color segment applying
