(XXIX-B3, 2012
d when the amount of
d, the wall surface
S Way, an exact wall
CEDURE
ht Parameters: The
digital camera used to
ible 1, and the flight
e shown in Table 2.
680 [Pixel]
2 [mm]
itions
ate 600 [m]
ate 66.7 [96]
ate 66.7 [96]
]
ters
ple of the multi-view
n Figure 5.
> Experiments
| complex image with
s: from numbers | to
osed technique can
t are taken as a result
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
42 Experimental Procedure
42.1 Generation of Voxel-Images: The wall polygons,
which are obtained by the extension method with the line
segment of the building footprint created in generating the
voxel-image, are superimposed on independent rectification
images. Moreover, the result for interpolation processing of the
wall surface images in a voxel-image will be illustrated. This
shows that the generation of voxel-images is functioning
correctly geometrically. In the experiment, the building
polygon data of the existing map was used to specify the
processing region.
422 Extraction of rooftop and ground by horizontal
plane matching: The division method of the voxel-image and
the difference in the searching direction illustrate that the
search for the rooftop and ground is possible. The variation in
the pattern of the correlation coefficient in the image matching
is illustrated in Figure 6, and the validity of the matching
processing is proved.
42.3 Division matching for wall direction compensation:
Here wall division matching in the object space is carried out,
and it verifies that wall matching, which also considers
direction compensation, is performed appropriately. The
position obtained by manual plotting is compared with the
result of the proposed technique, and the position accuracy is
evaluated quantitatively.
5. RESULTS AND DISCUSSION
5.1 Generation of Voxel-Image
We performed a domain extension that made the baseline on
the roof footprint line segment a part of the outline on the
south. The partial voxel-image that is generated from the
multi-view images is shown in Figure 6. The spatial vertical
line is taken from the two-point vertex of a baseline of the roof
footprint line segment, shown as the black solid line in the
figure, toward the vanishing point. The square area at the top
of the partial voxel-image, as shown by the red frame in the
figure, is decided to surround this baseline. The independent
rectification image can be calculated by the simultaneous
magnification conversion of the image centering on the
vanishing point with the multi-view images. The clipping
image is created with the magnification conversion according
to an altitude level (example: every 10 cm from 0 to 300 m),
and it accumulates into the object space as a 3D partial voxel-
image. According to this process, a five-dimensional partial
voxel-image of the object-space coordinates XYZ (three
dimensions), image numbers (one dimension), and a color band
(one dimension) is generated.
Next, the interpolation of the wall surface image into a voxel-
Image is shown in Figure 7. This figure of the wall surface
shows the cross-sectional image, which is the sliced partial
voxel-image, at the initial position to obtain from the baseline.
Image numbers 7 through 11 showing the upper part of the
building were used to extract an undistorted image of the right
wall. In the images of the lower half (numbers 1 to 6), shown in
Figure 7, the right wall surface has become invisible because of
the building’s own occlusion. In this process, we are able to
extract the wall surface texture from the multi-view images
using the character of the independent rectification images that
can be calculated by the simultaneous magnification conversion
Figure 6. Generation of Voxel-Image
#1
Figure 7. Interpolation in Local Voxel-Image
of the images centering on the vanishing point with the multi-
view images. It is possible that the right projected conversion of
the multi-view images enables extraction by the arbitrary slice
plane in the object-space coordinate system.
5.2 Extraction of rooftop and ground by horizontal plane
matching
Plane extraction by object-space searching is performed for the
rooftop and the ground using the partial voxel-image of the
nearby roof footprint line segment. The result is shown in
Figure 8. The graph of the correlation value for the multi-image
matching is illustrated in Figure 8 for the rooftop and ground.
The blue frame in Figure 8 shows the results of the multi-image
matching using all the partial voxel-images by the inside of the
building, as explained in Figure 3.
