Digital Signal Processing Reference
In-Depth Information
3
Multi-part Fuzzy Projection
Assuming that the homogeneous coordinates of point p in world coordinate system
are
(
XYZ
, its corresponding image coordinates are
(,)
uv
when the cam-
, , , )
WWW
. According to equ. (1), if
(,)
uv
are known, we
era coordinates are
XYZ
, ,
CCC
Z
[13].
could project p onto the space plane with fixed
X
W
u
Y
[
]
W
ZvKRt
=
(1)
C
Z
W
1
1
R is a orthogonal identity matrix with the size of 3×3. It is a rotation parameter. t is a
translation parameter. Both R and t is decided by the world coordinates of the camera.
Matrix K is a intrinsic parameter of the camera.
Fig. 1.
Sketch map of multi-view ground plane warping
In order to decrease the computational costs, a rectangle is used to represent each
foreground blob. In that case, warping a foreground blob onto a space plane has
become the process of computing corresponding values of
(
XY
with the coordi-
, )
WW
nate input of rectangle corners. On the ground plane,
. After warping, a rec-
tangle will become an irregular quadrilateral on the plane. Based on the conventional
assumption that feet area are always on the ground, the irregular quadrilateral areas
belonging to the same target are sure to overlap, and the overlapping area on the
ground plane is related to warped feet area, shown in Fig.1. Camera X and camera Y
are two different views, and there are four irregular quadrilaterals on the ground
plane. The two yellow ones are warped foreground blobs from the image captured by
Z
=
0
W
