Digital Signal Processing Reference
In-Depth Information
The proposed method in this paper is done on the 3D space. The detection method
is mainly based on the following prior knowledge: elevation relative to ground sur-
face of ground targets compared with that of buildings or trees is much lower; distri-
bution of points of the same ground target is relatively compact while different
ground targets are relatively far away from each other; ground targets has a sufficient
flatness areas.
In this study, the sections are arranged as follows: (1) In section 2, we first remove
the outliers from the data set. Then filter the points to separate the ground points from
non-ground points. The ground points are later used to estimate the ground surface
from which the relative height of non-ground points can be obtained to remove trees
and high part of buildings. (2)In section 3, 3D region growing algorithm base on vec-
tor consistency is employed to detect and represent suspicious targets; (3)In section 4,
experiment and conclusion.
2
Segment
2.1
Outlier Points Removal
The outlier is a point away from the landscape, such as systematic errors, noise, or
birds. The outlier is usually classified as high outliers and low outliers. High outlier
occurs when the laser beam hit the birds, low flying aircraft and noise points whose
removal is relatively easy. The low outlier is generally caused due to multipath effects
and other factors.
Assumed k nearest neighbours of LiDAR point
p
(
x
,
y
,
z
)
is, respectively
0
0
0
0
pxyz p
and
(, , .
kkkk
p
..
p
(
x
,
y
,
z
)
for
…
together form the point set
1
1
1
1
1
2
k
{ }
{ }
I
can be fitted by the least squares method.
A point will be determined to be a outlier if the point to fitting plane distance is
greater than a given threshold. The method can not only filter out the high outlier and
low outlier, can effectively filter out for a few outliers close to each other.
I
.Then the fitting plane of point set
2.2
Point Cloud Filtering
Point Cloud filtering is the process that removes the non-ground points from point
cloud. The main purpose of the point cloud filtering in the subject of remote sensing
is to filter out the non-ground points and generates DEM by the ground point, but in
our proposed method, point cloud filtering is used to extract targets of interest by the
non-ground points. ETEW (Elevation Threshold with Expanding the Window Filter)
which is a point cloud filtering algorithm based on the slope change is adopted in this
paper.
The abbreviated steps of ETEW are as follows: (1) data set is divided into an array
of square cells. In each cells, only the elevation of the lowest point is retained. For the
next iteration the cells are increased in size and the minimum elevation in each cell is
determined. (2) each point in each cell minus the height of the minimum point in this
cell. If the height difference is greater than the threshold, then those points are classi-
fied as non ground points. (3)The process is repeated with the cells and thresholds
increasing in size until no points from the previous iteration are discarded.
