Sei sulla pagina 1di 3

A flow to generate DEM from Lidar data

Lai Xudong Zheng Xuedong


129 luoyu Road, Wuhan 430079, China Spatial Information Technology Application Monograph
School of Remote Sensing and Information Engineering, Wuhan, China
Wuhan University
Wuhan, China Jiang Junwei
lxdwh@163.com China railway siyuan survey and design group co., ltd
Wuhan, China

Abstract—To consume less time and less memory, this paper footprint is on non-ground objects. And we must eliminate it.
presents a flow to generate DEM from Lidar data. Firstly the Based on this principle, much different filtering arithmetic has
flow labels non-ground points based on multi-echo information. been presented.
Secondly by the relativity among neighbor points, the flow labels
many other non-ground points. After these two up-to-down steps, Volsselman designed a specific kernel function using such
many non-ground points have been eliminated, which has theory in 2000, and generated DEM by mathematical
reduced the amount of the Lidar points successful. At last, the morphology operation [1].Zhang Dong has proposed a
flow use filtering arithmetic based on scan lines to get DEM. This threshold segment method based on the wave crest of range
paper comes to a conclusion that both the flow and algorithm are image. As it still uses a certain value to estimate every image
feasible by using them to disposing laboratory data. element, it can’t reflect the real terrain impersonally. The
results will be worse especially in areas where landform
Keywords- Lidar; Filter; DEM; flow change is great [2]. Through order scan along the scan lines of
LIDAR data, Shan marked ground point and object point by
I. INTRODUCTION judging whether the different of gradient and elevation between
neighbor points exceed threshold [3]. Hyun S. Lee improved
Nowadays, Lidar (light detection and ranging, Lidar)
linear prediction method, assembling linear prediction and
technology has become an important remote sensing
adaptation processing technique to generate elevation model,
technology to acquire of spatial information, which includes
obtained refined DEM [4]. There is some other filtering
light detection and ranging technology, computer technology,
arithmetic which is use mult-kinds data, such as intensive data,
satellite positioning device (GPS), inertial navigation device
digital photo data, GIS data, etc[5,6].
(IMS). With the rapid development of relevant technology, the
application of Lidar technology in the field of remote sensing is All these filtering arithmetic mentioned above are difficulty
more and more extensive. to apply in project, the reasons mostly is the Lidar data is too
magnitude to filtering it in a short time. Today there are mainly
Lidar data is useful for rapidly obtaining large-scale 3D
two filtering arithmetic have been used in business software.
terrain data and generating digital products such as DEM
One is TerraSolid; it is based on the TIN densification. Firstly
(Digital Elevation Model). The DEM can be used in many
it partition the region to some grids each of which is 60 meters
fields, such as, in urban plan, in digital city, in the
measurement and calculation of earthwork, in management of h60 meters or more. Secondly, it selects one low point in
drainage area, etc. Since it can fast and veraciously generate every grid. Then it builds a TIN of the low points. At last, it
DEM, Lidar data is becoming a very important way in adds nearby points to TIN if the points meet the rules. The
surveying and mapping. other is Scop++; it is based on Least Squares Interpolation.
Firstly, it uses all points to interpolation with unit weights.
Secondly it analyses of the residuals. Then it weights the
II. PRESENT STATUSES AND ANALYZING measurements with an asymmetric weight function. At last it
The Lidar data usually gives footprints not all on the real interpolation with new weights. All these filtering arithmetic
terrain surface, some on artificial buildings (roofs, chimneys, can get good results in most condition, but they are all time-
towers, etc.) and some on vegetation (trees, bushes, grass, etc.). consuming.
In order to generate DEM we must eliminate off-terrain points
and this is filtering of Lidar data. III. FLOW DESCRIPTIONGS
The most classic principle of generate DEM from Lidar The flow has several steps:
data is that the Lidar points’ height will change in short
distance if there are two kinds of objects. That is, the elevation A. Use multi-echo label non-ground points
should be continuous and the topographic fracture lines (steep
Unlike the TerraSolid filtering arithmetic, this flow uses an
banks, cliffs, etc.) are less. So, if suddenly changing is taken
up-to-down tactic to filter the Lidar data. According to the
place to the adjacent footprints, it is probable that the higher

978-1-4244-4994-1/09/$25.00 ©2009 IEEE


multi-echo information we can get the non-ground points and
label them. Those points which are not the last returns of multi-
echo are affirmatively non-ground points. By this way, we can
reduce the amount of the Lidar data.

B. Reduce point number by check neighbor points


Judging the neighbor points of the labeled points, if they
have the same altitude or their distance is shortly, label these
points as non-ground points too. This step is the key of the
flow.
There are several kernels in this step. Firstly is how to
define “neighbor”, we can detect the original Lidar data to get
the density of points (suppose it is D). As we know, most
building is less than 60 meters, so we can define “neighbor” as
60 metershD, so we can check the points with the number of
60 metershD before and follow the labeled point. Secondly is
how to check the neighbor points. Some arithmetic builds TIN
or other index (k-d tree, etc.) to search neighbor points, these
will cost much time and the searching of data need much time
too. Because the Lidar data are scattered, it is difficult to index
it in a simple way. In this test, we check the neighbor along the
scan lines of Lidar data. Since the Lidar System gather data by
line-scan mode, the Lidar data are stored in the order along the
scan lines. So we can get neighbor points along the scan lines
quickly. Thirdly is how to define the same altitude. We
suppose that in a small area, ground and objects are continuous. Figure 1. the flow diagram
If the height difference between the neighbor points and the
labeled point is less than a given threshold, we can look them
as a same class. Finally is why we define a short distance. This IV. EXPERIMENT AND RESULT
way is to label the wall points. When Lidar scan the ground The follows are the experimental results of Lidar data by
there are some footprints reflected by wall. The character of using the flow proposed above. Fig. 1 is the original data; Fig.
wall points is scatter in 3D space but compact in 2D space, and 2 is the result of label multi-echo points; Fig. 3 is the labeled
they have only one echo. So we can define a short distance, if non-ground points; Fig. 4 is the ground points after filtering.
the distance between a point and the labeled point is less than
the defined distance and only have one echo, we look it as wall
points.
After this step, we can reduce the amount of the Lidar
points about 10-40 percent.

C. Label ground points


From the above steps we reduce the amount of the Lidar
points. In this step, we can use some adult filtering arithmetic.
To avoid build index, we use the arithmetic based on scan
lines, so that we can save much time and memory. The process
is alone the scan lines, get the first unlabeled point, and label it
as ground point. Then compare its value to the next point’s
value, if the difference is less than given threshold, label it as
ground point too. Continue this, until the pending points has
been labeled or no points. Do this procedure again from the
contrary order of Lidar data. At last, get the intersection of Figure 2. Figure2. original data
above two results. That is the finally result.
Figure 5. ground points after filtering

V. EPILOGUES
Figure 3. multi-echo points(yellow color) To consume less time and less memory, this paper presents
a flow to generate Dem from Lidar data. By this flow, we need
not build index to the scatter points. The test proves this flow is
feasible.

ACKNOWLEDGMENT
This work is supported by The National High Technology
Research and Development Program of China
(2006AA122101) and The Research Fund for the Doctoral
Program of Higher Education (20060486041).

REFERENCES
[1] Vosselman G., 2000. Slope based filtering of laser altimetry data.
International Archives of the Photogrammetry, Remote Sensing and
Spatial Information Science. 33(B3/2), pp. 935 -942.
[2] Zhang Dong, 2005. 3-D Reconstruction of Buildings Based on Lidar
Data and Aerial Image. Wuhan: Wuhan University (in chinese)
[3] Shan J, A., 2005. Sampath. Urban DEM generation from raw LIDAR
data: A labeling algorithm and its performance. Photogrammetric
Figure 4. labeled non-ground points engineering and remote sensing, Journal of the American Society for
Photogrammetry and Remote Sensing. 71(2), pp. 217-226.
[4] Lee H S, Younan N H., 2003. DTM extraction of Lidar returns via
adaptive processing. IEEE Transactions on Geoscience and Remote
Sensing. 41(9), pp. 2063-2069
[5] Elaksher, Ahmed F, 2008. Fusion of hyper spectral images and lidar-
based dems for coastal mapping. Optics and Lasers in Engineering.
46(7), pp. 493-498
[6] Papasaika, Haris; Poli, Daniela; Baltsavias, Emmanuel, 2009. Fusion of
digital elevation models from various data sources. Proceedings of the
International Conference on Advanced Geographic Information Systems
and Web Services, GEOWS 2009, pp. 117-122

Potrebbero piacerti anche