Digital Signal Processing Reference
In-Depth Information
point belongs to. It is know that the relative elevation to ground surface of ground
targets is low, usually less than 4 meters, while that of trees and buildings is relatively
high. According to this prior knowledge, a elevation threshold is set to get rid of a lot
of trees points and building points. Unluckily, this method has a flaw, that is, the low
part of trees and buildings will be also included.
Laser radar echo information can also be used to filter out some of the objects.
Pulses with multiple echos and with large elevation difference between first echo and
last echo are very likely to be reflected from the trees or the edge of the buildings.
Fig. 3.
Result of objects removal
We can see from Fig 3 that almost civilian vehicles have been retained correctly.
However, edge of the buildings and tree with low elevation are also mistaken as po-
tential ground targets points which will inevitably bring adverse effect to the detection
performance.
3
Detection of the Ground Targets
3.1
3D Region Growing Based on Vector Consistency
One of the basic problems of Region growing algorithm is to obtain the initial seed
points. The initial seed points are extracted in the undivided original point cloud.
First, the point cloud space is divided into a number of cells. Each cell is assigned as a
value, that is, maximum absolute elevation difference which is the difference between
the highest elevation and the lowest elevation in the cell. A cell will be marked as
seed cell if the maximum absolute elevation difference of the cell is larger than a
given thresold
thr
T
, consequently, the point with the highest elevation of the seed
cell will be chosen as a initial seed point.
thr
T
has a large impact on detection
performance. On this parameter, we will be in the experimental section for further
analysis. Ground targets usually have a sufficient flatness plane. Points on the plane
will produce high vector consistency. Region growing criteria is determined by this
property.
Suppose that
s
is a seed point,
p
,
p
...
p
are the nearest neighbors of
s
, re-
n
and
1
2
k
,
21
.
If
p
is not a seed point and satisfies condition (1), then A is added to the seed
region.
n
,
n
...
n
are the unit normal vector of
s
and
p
p
...
p
spectively.
s
1
2
k
k
