The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
423 
3.1 Internal accuracy of evaluating 
In data extraction and processing, internal accuracy of the 
estimates used to evaluate the quality of the data. Although this 
criteria is not within the very tight precision, but to a certain 
extent, we can see that the accuracy of the data and the 
processing of the data quality. Such as the integrality of laser 
sampling and data packet. 
3.2 Classification of point cloud 
Above ground objects are problematic for tie point observations. 
Due to the angle of incidence, foliage and multiple echoes, 
vegetation is rarely observed in the same way from two 
different strips. To avoid poor tie points around vegetation or 
man made objects, the data should be filtered to extract the 
ground. 
Although it is usually quite simple for a human operator to 
identify what is ground, it is not practical to manually edit large 
amounts of LiDAR data. There are several methods of 
automated filtering to choose from including morphological, 
slope based filters and least square estimators. The results of 
these systems vary by terrain types (urban, steppe, mountainous) 
and the density of the LiDAR data. The systems can be 
compared by effectiveness of building/vegetation removal, 
speed of algorithm and smoothness of the derived surface. 
3.3 Wiping off abnormal points from ground data 
In order to wipe off abnormal points from ground data, the filter 
based in entropy of range image is put forward by author. 
The Airborne LiDAR data is a high-density point sampling of 
the terrain. These data are processed either as TIN elevation 
surface model or interpolated into a regular elevation surface 
model (DSM). The LiDAR DEM data can be also converted 
into a raster image. In addition LiDAR system captures also the 
intensity of the response signal corresponding to the LiDAR 
ground point. The intensity is a function of the reflectivity of 
the ground material and the intensity changes form a 
georeferenced grey level image like output. Figure 3 shows 
examples of range (left) and intensity (right) images generated 
from a LiDAR point cloud. 
Figure 3. Range image (left) and intensity image (right) 
generated from a LiDAR point cloud 
3.3.1 Entropy: Entropy is the quantitative measure of 
disorder in a system. The concept comes out of 
thermodynamics, which deals with the transfer of heat energy 
within a system. 
In image, the entropy is the measurement of information. So in 
a digital image, there are n gray values, gl, g2, ..., g n . The 
probability of gi is pj. The entropy of this digital image is 
defined as below: 
n 
H[P] = H [p\, p n ] = -¿X Pi log Pi (5) 
;=1 
Where The probability pi is approximately equal to: 
Where f t is frequency of gi 
N is the sum of pixels in image 
As proved, the most entropy exists in image where distributing 
homogeneously of gray value. 
The entropy of part of image is measurement of information 
locally. It denotes whether existing feature in part of image. 
3.3.2 Procedure: 
a) Calculating the entropy of entire image as E. 
b) Exporting raster image of elevation from ground 
LiDAR data after classification. 
c) In unit area (the size is defined by experience), 
calculating the entropy as e. If e > E, considering there are 
abnormal points. 
d) In the area where e > E, calculating average AVR, 
standard deviation S of elevation. For per point, 
calculatingUH = abs(H i — A VR) , where H j is the 
elevation of this point. IfD H > S , considering this point 
is abnormal and wiping off from ground data. 
3.4 Bore-sight calibration 
Bore-sight error is the angular misalignment between the laser 
sensor unit and IMU. Unlike a photographic image, a bore-sight 
error affects each observation and cannot be removed by 
applying a simple affine transformation to the entire strip. 
The systematic errors of bore-sight misalignment and scanner 
torsion error can be expressed in a parametric form. To solve 
for the unknown values, standard least-squares techniques can 
be used to determine the parameters in redundant data. 
In procedure of adjustment, a method of patch matching is 
needed to improve accuracy. 
3.4.1 Patch Matching: Conjugate planar patches in 
overlapping strips are supposed to be coplanar, regardless of the 
flying direction or any other parameters, unless there are biases 
affecting the data. Using this fact, planar patches can be used to 
qualify or detect biases in the LiDAR datasets. 
In a common area for two overlapping strips, that is represented 
the same physical surface for conjugate patches. So the angle of 
normal is the least for conjugate patches. This could be 
accomplished through an automatic process as described below: 
e) Generating a Triangulated Irregular Network (TIN) 
for overlapping areas. 
f) Choose one patch in left strip, matching the conjugate 
patch in right strip. The condition of marching successfully 
is that the angle between the first patch and the potential