International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
grid unit, in this paper we call it Density of Projected Points 
(DoPP); 3) to regard DoPP as the basis of objects classification. 
According to the difference of DoPP, sometimes need 
additional  height-information, different objects can be 
distinguished. The operational workflow of data processing is 
shown in Fig. 2. 
  
    
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Range Image 
Y [Theme Range 
Calculating DoPP »|/[Mage 
Object Segmentation Y 
Feature Extraction, Modeling, ... 
  
  
  
  
  
  
  
Fig.2 Workflow of Data Processing 
3. SPATIAL FEATURE AND SEGMENTATION OF 
RANGE IMAGE 
3.1 Spatial Feature Analysis Of Different Objects 
Range images consist of objects such as buildings, ground, 
trees, vehicles, lamp-poles, pedestrians etc. Our research in this 
paper is focused on the object segmentation and feature 
extraction of the important objects such as buildings, ground 
and independent objects such as lamp poles etc. 
As Fig. 1 shows, all targets in the field of view of the scanner 
can reflect laser to get “image-point”. On one hand, data 
collection has somehow blindness or uncertainty, so it is not 
certain to obtain feature of ground objects and terrain, and the 
automated identification of objects is very difficult, On the 
other hand, there has discreteness in the points on the same 
scan line and divergence between adjacent points which makes 
data processing very complex. In the following we analyze the 
spatial feature of different objects. 
3.1.1 Spatial Feature Of Ground Points: Range image as 
shown in Fig. 3 consists of plenty of topographic points. The 
topographic points usually are smoother and the height value 1s 
relatively smaller and the variation in height value is not big. 
The distribution of topographic points on the horizontal plan is 
irregular. The sampling points on each unit area accord with a 
certain rule: There are many points near the scanner. The farer 
distance from the scanner, the less points. 
  
Fig.3 Points cloud of a building 
3.1.2 Spatial Feature Of Independent Object: Independent 
objects such as trees and lamp-poles, with a certain height and 
area range, which are higher than surrounding topographic 
points and the sampling frequency of local units on horizontal 
plan is high. As shown in Fig. 4. 
» Wo 
agua ot 
€ 
Lamp Pole = a P mua 
x Tide Hannes 
u 
Fig.4 Sampling independent objects 
3.1.3 Spatial Feature Of Building Points: Without lost 
universality, we can consider the buildings are higher than 
surrounding terrain and their walls are vertical. There are plenty 
of sampling points from the building surface and the sampling 
frequency of the horizontal outlines of the building is high. And 
on the same scanning line, the deviation between adjacent 
points is very small in X and Y directions and there has an 
approximately vertical direction vector in Z direction. 
3.2 Principles And Method Of DoPP 
According to the spatial feature analysis of points, we propose 
the method of DoPP below (see equation 1). 
ATN(H | D) 
a 
DoPPzin (1) 
Here, the value of DoPP is dependent on the target's height (H). 
distance from the scanner to the target (D), and on the 
resolution of the scanner ( a ). 
To simplify the calculation, firstly we can divide surveying 
region into regular mesh grids, then project all points into 
horizontal plane by equation 2, 
   
Int 
reg 
car 
we 
rea 
san 
of J 
T, 
forr 
Do.