The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
455 
3. LIDAR DATA PROCESSING 
LIDAR system uses random commercial software to process the 
data of plane trackings plane attitude ^ laser ranging and the 
swinging angle of laser scanning mirror, and finally, obtains the 
three-dimensional coordinates(X,Y,Z) data of various surveying 
points. The points, namely LIDAR original data, called “point 
cloud”, are three-dimensional discrete dot matrix data without 
attribute, suspending in the air as shown in Figure 1,it is the 
point cloud data in Lu Jia zhui area overlapped ortho-image. 
The LIDAR original point cloud data have following 
characteristics: 
1) Massive data: the data of 20 square kilometers acquired by 
Shanghai Institute of Surveying and Mapping is about 10G data 
(one point every 0.6m). at present, the software and hardware of 
an ordinary computer are unable to handle so magnanimous 
data once, therefore ,data must be blocked for the follow-up 
processing. 
2) Insignificant discrete point: As the figure shows, point cloud 
reflects wholly the shape of buildings and the topography 
distribution characteristics, but in fact, single point cloud has no 
significance. Therefore, firstly, we must conduct the 
classification and recognition of original point cloud in order to 
realize the reconstruction of the three-dimensional city model 
through the operations of feature extraction and the constructed 
surface. 
3) Rich information: the original point cloud information 
includes topography and objects, the latter comprise houses, 
trees, vehicles and power lines and so on. Now, The technology 
is unable to classify and recognize all the information one by 
one from the point cloud, therefore, the present research mainly 
focuses on how to extract interesting information, in the 
meanwhile, the emphasis is also different in different research 
fields. Regarding city, the key to research is the extraction of 
topography and buildings; and regarding forestry, the key is the 
extraction of trees. 
LIDAR original data is pre-processed to produce digital surface 
model (DSM). Through classification and extraction, 
topography and object information related to model is acquired 
to prepare for three-dimensional city model. 
3.1 Data preprocessing 
Data preprocessing includes the deletion of abnormal point, 
coordinate transformation and flight strip combination. 
1) Deletion of abnormal points: in the process of actual flight 
surveying, due to all kinds of factors such as mirror reflection, 
circuit problem of system and obstacle, there is abnormal value 
in LIDAR original data, so filtering the original data must be 
conducted in order to delete those abnormal points that are 
higher than the flight height or lower than the ground. 
2) Coordinate transformation: original point cloud data of 
LIDAR belongs to WGS-84 coordinate system. As far as 
Shanghai is concerned, WGS-84 coordinate system of those 
points should be transformed into local coordinate system, for 
this purpose , firstly, WGS-84 coordinate system is 
transformed into Beijing 1954 coordinate system, and then 
transforms it into local coordinate system. Regarding elevation 
datum, what GPS provides is geodetic height based on the 
surface of ellipsoid. However, in practice, what we need is 
normal height based on the geoid. Elevation datum 
transformation can be achieved by establishing normal height 
model depending on some known control point fitting. 
3) Flight strip combination: when LIDAR system works, due to 
the limitation of flight height and scanning FOV (field of view), 
the plane must fly the multi-strip route of zigzag in order to 
cover a certain area. Moreover, these routes must maintain 
certain degree of overlapping (10%-20%). Therefore the 
original LIDAR data of different flight strips must be merged, 
and put into order according to X direction or Y direction. And 
then the sorted LIDAR data is merged into a whole according to 
a certain order for block extraction and processing. 
3.2 Data classification and extraction 
LIDAR technology has enjoyed a development of more than ten 
years. The questions of hardware technology and system 
integration have been solved very well and now there are many 
mature commercial systems available. But the post-processing 
technology of LIDAR data lags relatively behind the demands, 
whose key problem is classification and extraction of data. 
Many international commercial companies and institutes of 
scientific research are devoted to this aspect and many kinds of 
filtering algorithms and extraction methods are already 
presented. At present, duo to the complexity of the object and 
topography in the objective world, the researches on the 
classification and the extraction of data lay different emphasis 
for different application. Speaking of three-dimensional city 
modeling, the research mainly concentrates on the extraction of 
topography and buildings. 
In the aspect of topography extraction, Axelsson et al. in 
Sweden presented the gradual enhancement algorithm based on 
TIN (Triangular Irregular Network), which first chooses the 
initial ground point to construct the initial TIN, and second set 
parameter values such as the angle of iterative, the distance of 
iterative, the angle of tilt and so on, and third add the points of 
DSM to TIN step by step through iterative repeatedly, and 
finally, realize topography extraction; Franz Rottensteiner et al. 
in Austria used the lamination robust interpolation algorithm 
based on the non-uniform error distribution function to obtain 
DTM (Digital Terrain Model); Vosselman et al. in Delft 
University of Dutch presented morphology filtering method to 
separate the topography points from the non-topography points 
in DSM, which was improved and optimized later by many 
people and from which many other algorithms evolved; Wack 
and Wimmer in Austria presented a method based on grid 
grading to obtain DTM from DSM; Sithole et al.in Dutch 
compared all the algorithms mentioned above and chose many 
regions with different characteristics for experiment. Their 
results showed that these algorithms are effective for simple 
regions. But for the complex regions, especially the cities, the 
results aren’t very satisfactory and need manual intervention or 
further processing. 
In the aspect of building extraction, Alhartthy et al. of Purdue 
University in America designed a evolution-filtering method to 
extract the three-dimensional information of a building; 
Vosselman et al .of Delft University in Dutch used the three- 
dimensional Hough transform to extract the roof information 
from DSM and the outline from the information of the plane. 
Regarding extremely crowded DSM point cloud, they also 
applied Delaunay triangle processing method to obtain the 
information of buildings; Michel et al. of Ohio University in 
America extracted the outside shape of buildings by the region-