Istanbul 2004
nalyzed by a
travel time of
rocessing the
operating on
roperties (e.g.
t and so less
letection the
vance, only
nning system
mitted pulse.
wer and the
dth 7 without
nsider pulse
oduct of a
ng bandwidth
rm the range
the speed of
(3)
'anner system
geometry in
gth) different
In this paper
id number of
the distance
a description
| area.
| locate the
ons.
the intensity
inge value is
the matched
ca of interest
We integrate
wer value P
(4)
of quantizing
19 the noise
ie number of
e number of
ment for the
ut signal.
higher level
the spatial
ce primitives
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
in the form of regions. Separating the pulse properties range and
power in regions is fundamental to handling the ambiguity of
multiple reflections at the same spatial position. Boundary
pixels are labeled as shared pixels belonging to background and
foreground. Figure 6 illustrates the region definition.
The computed region N consisting of range values is described
by the range image Ay. Then the pulse power values of the
same pulses are used to generate the power image /y. In this
manner we receive a corresponding representation of all
determined regions (N = 1, 2, 3, ...).
The region boundary mainly contains pixels derived from
multiple reflections. The region interior is characterized by
single reflections and fills up the region to the boundary. The
region background is labeled with zero pixel values. Generally
it can be said that nonzero pixels belong to the region and zero
pixels constitute the background.
Region background:
Region boundary :
Figure 6. Region components definition
An iterative region growing algorithm considers the range
properties of all pulses in the spatial neighborhood: if the range
difference of a pulse and the proofing pulse is below a given
threshold, then the pulse is connected and grouped as a new
element to this region. Samples of range and power images as
segmentation results are depicted in Figure 7.
e f g h
Figure 7. Segmentation results (grid mesh width g-275cm,
Gaussian distribution of the beam at the grid line
+26): a)-d) range image Ry=1 Ry=r, Ry=3, Ra,
e)-h) power image /v=1 /v=, Ines, Ines
4.4 Region edges
For physical edge description a data driven region boundary
(edges) approach similar to the algorithm of Besl (1988) is
used. This approach generates a parametric function
representation of the region boundary for higher level
processing. To refine the region boundary the estimation of
edge position and edge orientation from received pulse power
and analysis of the spatial neighborhood is proposed.
First each power image /y is processed with a 4-neighboring
edge following algorithm to determine a parametric function
representation of the region boundary. The algorithm traces the
exterior boundary of the region N in the power image /y(x.)).
As output for each trace, we construct two boundary
parameterization functions xy(s) and ya(s) containing the row
and column coordinates of the connected boundary pixels.
Additionally the boundary range function zy(s) can be found
using the range image.
Figure 8 shows an example of an exterior edge description with
the three parameter boundary functions xy=2(5), Vw=2(5). En=a(5)
of the region N=2 and corresponding power image /v with
overlaid boundary.
© 40— Trt ery BA 2
> | >
S30, S 30
o | o
= | e
& 20 2 20! |
mud > |
$10 210
= | = |
o [m T T o { :
8.0 50 100s +? © 50 100s
a b
$ 195--—
N
S 134: |
2.133; |
d nl
2132
(v
2 131
o E tue Ld
8130. 50 100s
C d
Figure 8. Edge description: parameter functions a) xy-»(s),
b) vw-2(5), €) Zw-2(s), d) power image /y-» with
overlaid outer boundary function
boundary labeled Æ, the region interior / and the region
background B. By assuming a straight edge from the upper left
pixel to the center pixel the line is undetermined without further
information. A sample of possible edges ë, & and & is
depicted.
Figure 9a shows a 3-by-3 pixel neighborhood including a region
él à! ! 1 1
TC moy ED TUA NL r =
& BN a
~ By N 1 1 é2
2. pi 1 1
€ B ' i
TE I ET - -
! ' i |
9 1 1 t
1 1 1 1
zl 1 x 1
oe mee TTT - -
m i 1 i
1 1 1 1
1 1 1 1
D A 1 1
Sp pm Fm EC = + = -
' 1 t t ‘ '
a b
Figure 9. Edge estimation a) sample of edges e;; &; and e;,
b) estimated edge e;
To estimate the edge position and edge orientation the received
pulse power is analyzed for the geometrical shape of the
illuminated surface. With the assumption of a uniform
reflectance property of the surface the target reflectance
depends only on the geometrical shape. Therefore we assume
uniform reflectance and straight object edges. The laser beam
with its radial symmetric Gaussian distribution illuminates an
edge. For integrating along the coordinate we assume the edge
is vertical. Furthermore the beam center distance to the edge e
is given by d (Figure 10).
111