The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
248 
multiple profiles, which can weaken the impact of special 
circumstances. Secondly, as to the inconsistency condition of 
highway width or water surface width, the same number of 
samplings from the start line is adopted to get uniform sequence. 
Thirdly, after the positions related with the start point, the 
termination point, the highest elevation and the lowest elevation 
are sought, constant-value function is applied for fitting the 
neighbour points of given number. Then the nearest point on the 
fitting line replace the above four calculated points as the actual 
positions used in determination functions. 
Although the above functions utilize the common regulars, the 
real-world has some anomalies. The artificial river probably has 
a cross interface with natural river. One side-bank of artificial 
rivers sometimes makes use of natural terrain slope. The special 
situations also should be considered in the following work. 
3. EXPERIMENTS AND ANALYSIS 
The experiments will use real LiDAR ranging data to testify the 
algorithm proposed above. And some related analyses are made 
for further improvements, which aims at becoming one module 
of practical software. 
3.1 Data 
This paper applies Abbeville district’s LiDAR ranging data in 
America to validate the proposed algorithm, at the same time 
rivers’ distribution from the same area’s high-resolution RS 
image is applied and extracted as comparison template. 
Figure 5 and 6 illustrates the characteristic distribution of the 
chosen area. LiDAR samplings has 5 times grid-spacing than 
the spatial resolution of RS image (lm), which needs data 
compression for comparison. 
The figures also show that LiDAR images are generally more 
homogeneous than RS images. This avails of automatic process 
when extracting information. And this also constitutes the 
foundation of whole algorithm. 
Figure 5. Typical LiDAR image for experiments 
Figure 6. RS image of the same area for comparison 
3.2 Results 
After edge extraction, skeleton generation and PFF recognition, 
as showed in figure 7, 8 and 9, the areas related with artificial 
river, natural river and highway can be determined. The results 
indicate that although some long strips of terrain are picked out 
by edge extraction, PFF can recognize them. 
Figure 8. LiDAR image’s skeleton generation result based on 
figure 7.