3, 2012
ns present the
‚ed pulse. The
ons of all the
. The form of
ferent studies
ion capability
meters (width
ride additional
09) show that
rch effort to
s.
se a modified
| include the
fferent terrain
ith traditional
bh.
je issue. This
er of different
| made in this
atics research
with current
ated filtering
cene. For this
of different
classification
e of different
ective of this
cated on the
ed on another
sthodology is
rocess and a
e proposed
1 preparation
ing segments
triangulation
traction, and,
d the ground
tly using the
ie detection-
d in the area
nerated. This
einal data in
ons must be
med value to
according to
oss of spatial
nce the grid
are available
n any point.
n the ground
; only simple
Since there
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
are positions without any original data inside, it is necessary to
perform an interpolation process (distance interpolation from
neighboring positions). The whole process to obtain the regular
grid is shown in Figure 1, and it is designed to obtain digital
surface model of the scene in raster format (DSM).
Figure 1. Regular grid generation process. Upper: original data;
Middle: Initial regular grid generation —red pixels
representing no data grids-; Lower: Final regular
grid.
2.2 Segmentation process
The segmentation process aims with grouping cells in regions
according to a similarity criterion. For this process, the different
segments or regions presented in the scene must be generated,
using a region growing process considering a binary space,
where all the pixels will belong to a one of the two possible
classes (ground or not-ground).
This methodological propose introduces the “planar point”
concept, where a given point defines a local plane. For this
consideration, it is necessary to obtain the plane define for each
point and its neighbourhood. The accuracy of the adjustment is
used to establish whether the point have a local behaviour as flat
or not, generating for this purpose an image of the scene
assigning to each position the precision adjustment centered at
that position.
Figure 2. Segmentation process. Upper left: MDS; Upper right:
accuracy of local planes adjustment; Lower left:
binary image; Lower right: segmentation.
For the generation of the binarized space from the previous
image, it is necessary to introduce also the "barrier point"
concept or point of geometric discontinuity. This concept refers
to a point where there is an abrupt change in the geometric
continuity defined at the scene of points corresponding to the
outer edges and interiors of buildings, vegetation zones, bridges
or other structures, etc., where precision adjustment is low and
it should behave as a barrier in the later region growing process.
Once the threshold indicated to differentiate the "planar points"
from the "barrier points", it is possible to generate the binarized
image which will be the segmentation by region growing
procedure. This region growing process ends with the image
segmentation and it obtains the different segments present in the
scene. The results of the process of segmentation can be seen in
Figure 2.
2.3 Progressive densification process
Parallel with the segmentation process, another process oriented
to the labeling of low points that belongs to the ground will be
developed though a progressive triangulation densification
according to a methodology similar that the proposed by
Axelsson (2000). This process starts from an initial
triangulation created from the low-points of the DSM.
The process will be a densification process of the triangulation
based in the use a double threshold algorithm (angle and
distance) that will incorporate this triangulation ground points
with the required density in each case. In this way, it is possible
to obtain a significant representation of the positions
corresponding to the ground that could be a approximation of
DTM in TIN structure.
153
