CLASSIFICATION AND FILTERING OF AIRBORNE TOPOGRAPHIC LIDAR DATA
D. Gajski, T. Fiedler, A. Krtalic
Institute for Photogrammetry, Faculty for Geodesy, University of Zagreb, Kaciceva 26, Zagreb, Croatia —
(dgajski,tfiedler,akrtalic)@geodet.geof.hr
KEY WORDS: LIDAR, Filtering, DTM
ABSTRACT:
Data collection for DEMs by using of the airborne topographic LIDAR thanks his popularity to the high efficiency and high
potential of accuracy. However, there is no way of differentiation between the features of importance from the others, reachable by
LIDAR, if they are pointed by laser beam, too. Therefore, there are the methods of classification and filtering of LIDAR data of most
importance. This paper presents several approaches of classification and filtering of LIDAR data, discussing their characteristics and
suitability to concrete applications.
1. INTRODUCTION
The extraordinary measuring capabilities of airborne
topographic LIDAR systems and their fast evolution since last
ten years leaded to very reliable and economical acceptable
technology for a whole process of capturing the data. Thus is
laser scanning established as a very precise and effective way of
detailed measuring of terrain surface as well as many other
surfaces lied above the terrain (i.e. vegetation canopy surface,
roof-surface of buildings). Although, the laser scanning is full
automatic method of data capturing, it is no selective. So, many
surfaces of different materials reflect enough intensity of
radiation to be registered by LIDAR sensor, if they are pointed
by laser-beam, too (table 1). Therefore, all this points have
same chances to be measured by the system although they
belong to different object classes (fig. 1).
Figure 1. Non-selectivity of data capturing by LIDAR.
100
There are many application areas, in which a surface,
enveloping the terrain with all objects above them, is used
(radio or mobile phone network planning, urban planning,
analysing of pollution spreading in ecology, bio-mass
estimation in forestry). But, the topographic application needs
the description of true topography (the terrain) only. So, there is
a strong need to find a reliable method to perform a basic
classification of gathered dataset into points that belongs to
terrain, and off-terrain points. There are many approaches to
perform this, non-trivial task and some of them are presented in
this paper.
If we intend to extract points from whole dataset, that describe
the terrain surface only, because of interpolation of DTM, then
the off-terrain points can be treated as roughly-wrong
measurements of terrain. Therefore, we can use the adopted
statistical methods for gross-error detection, to filter out such
“outliers” from dataset. This process is called filtering.
The filtering can be done:
- before interpolation of DTM
- together with interpolation of DTM
Several approaches of filtering and classification methods are
present, depends on features of measuring that are taken into
consideration. These features are:
- physical characteristics of received measuring signal
(«first pulse — last pulse« method)
- A statistical characteristic of captured dataset
- morphometric characteristic of gathered data
Some approaches are favourable, that they deal with original
data. Others do some pre-processing to put the data in regular
matrix structure. That can speed up the filtering and
classification drastically, and many algorithms, originated from
image processing, are directly applicable.
2. METHODS OF CLASSIFICATION AND
FILTERING
2.1 First pulse — last pulse
By this method is determined, which of reflected pulses has to
be registered and used to compute coordinates of measured
point. Namely, the divergence of laser beam lie in range 0.3-
2mrad ^
height o:
03-2
reflects
objects 1
reflect l:
reaches
one emi
object (fi
15! retu
20 reti
3rd retu
last retu
Pu
]
It is obvi
to LIDA
these poi
produce :
the off-te
Figure :
Oct
This met
capable 1
develope
returned
importan
2001).
