HIGH
edu.cn
jp
ard Hall,
es for urban
val. The cast
ial image is
determine if
ative tracing
f the traced
statistics of
mean of the
je reference
ty mapping.
right region
Zhou, Qin,
model was
‘and, 2003)
lysis and
del is given
? corners in
y the same
ile with the
raphics as a
/ with a 3D
if the light
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
between a pair of points is occluded (Paglierroni D. W. and
Petersen S. M., 1994)( Paglieroni D. W., 1997)( Paglieroni D. W.,
1999)( Bittner J.). In our work, it is used to decide the shadows in
the projected image. Shadow restoration commonly is
accomplished by histogram approaches, or homomorphologic
filtering (Castleman K.R., 1998 ). These methods adjust the
intensity for each pixel in the image, instead of local processing
for the shadow area. A local processing is fulfilling the shadow
area by another image. To do this a pair of images is needed. In
our work digital image processing technique is used to remove
the shadows in the original image directly.
The image used in our research is the aerial photo acquired by
ADS40, which is a line scanning sensor. Another data is the
digital surface model (DSM) of the same region. We developed a
system to automatically detect, segment, and remove the
shadows of the buildings. It first computes the space coordinate
of a shadow from DSM by photogrammetriy and then projects it
to the image plane. By integrated shadow detection, the shadow
area in the image is segmented and labeled. At last the intensities
of the shadow area are restored. In shadow computation and ray
tracing, we proposed a building contour driven model. It is based
on partial parameter plane transform (PPPT). In shadow removal
we developed a method called companion area intensity
mapping (CAIM). Experiments show that the system can
precisely detect the shadow area and restore the brightness of the
shadow to a natural visual effect.
2. SHADOW DETECTION AND SEGMENTATION
The photo scanned by ADS40 is rectified through level 0 and
level 1 rectification to create the pseudo-orthoimage which is
used in our research. We have studied three approaches to detect
and segment the shadows in the image.
The first one is to detect and segment shadows only using image
analysis. The results are correct in most cases. However, because
of the complicity of the urban circumstance, there may be some
factors effecting the detection of the shadow. For example, the
high reflectivity ground, the glass wall of the building, make
shadow somewhere bright, and cause their intensities close to the
non-shadow area. These shadows may not be detected by image
analysis technology. Besides, the threshold of the segmentation
is difficult to decide. Thus, the segmentation of the shadow is not
803
reliable. The second one is to compute the shadow location in the
RGB image by photogrammetric engineering, using camera
model and digital surface model (DSM). If the DSM had been as
precise as the image, the result would have been perfect.
However, for the time being, this is not practice. The basic
locations of the buildings in DSM have no problem, but the
resolution is lower than the image. In addition, DSM is short of
details of the buildings. From our experiments, using this method
singly, the shadows detected in the image have some errors.
Based on the above factors, the last one is to integrate the above
two methods to adopt their advantages and abundant their
shortcomings. In detail, since the locating of method 2 is reliable,
the cast shadows are first computed by method 2, and based on it
the shadow area and its corresponding bright area are segmented
and labeled by image analysis. This strategy ensures correct
shadow locating, and no false shadow or losing shadow happens.
Meanwhile, the details of the shadow area shape persist.
2.1 Coordinate Of The Shadow In 3D Space
a
= | Line Extraction
=a
L/ ADS.
L
n
»
"nuuummamuumuumut
Fore-end Shadow
Detection
Back Tracing of the
Same-Section Shadows
PI ILEEEIEIELERIEEEELEEEEES
Figure 1 Flow chart of space shadow detection
The ADS40 model, DSM are used in this stage to figure out the
space coordinate of a shadow. The shadow we care refers to that
casts by a building. After all shadows are computed, they have to
be decided if they are visible or not in the image. The algorithm
of shadow coordinate in local space rectangular (LSR)
system(ADS40 Information Kit) can be described by the flow
chart in Figure 1. The LSR system is the object space under
WGS84 used for photogrammetric processing.
The altitudes of the sun can be represented by the zenith angle
and the altitude angle. The two angles are independent obviously.
We first rotate the DSM by an angle equal to the zenith, making
the zenith angle is equivalent to point from the left to the right
horizontally for the rotated DSM. Then we can compute the
