|, 9-11 Nov. 1999 
Cascades of Oregon. Rem. 
ing, D.J. (In Prep), PAR 
rmined from airborne lidar 
itum measurements. Rem. 
rding, D.J. (In review), 
10py height characteristics 
1), The structure of natural 
1iglas-fir forests in Oregon 
yervice general technical 
west Research Station., 
son of spectral reflectance 
canopy level for selected 
n., 33:79:93. 
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999 
Interpolation of high quality ground models from laser scanner data in forested areas 
N. Pfeifer!, T. Reiter?, C. Briese!, W. Rieger? 
1: Institute of Photogrammetry and Remote Sensing, Vienna University of Technology 
2: Institute of Surveying, Remote Sensing and Land Information, University of Agricultural Sciences Vienna 
np@ipf.tuwien.ac.at 
WG 111/5 and WG 11/2 
November 1999 
KEY WORDS: Laser Scanner, digital terrain model, interpolation, linear prediction, robust method 
ABSTRACT 
Air-borne laser scanning is an applicable method to derive digital terrain models (DTMs) in wooded terrain. A considerable 
amount of the laser points are reflections on the tree tops (vegetation points). Thus, special filtering algorithms are required 
to obtain the ground surface. Earlier, we proposed to use iterative linear prediction. We review existing methods and compare 
them to our approach. A list of advantages and disadvantages of our method is presented, but this list has also validity for 
laser scanner data processing in general. The quality of the DT Ms derived from laser scanner data and accuracy investigations 
are presented for two examples. 
1 Introduction 
The applications of DTMs are well known. Also for forested 
areas it can be of interest to have a DTM of high quality. 
Until now, this was not possible. Terrestrial tacheometry 
on one hand, is too expensive and takes too much time for 
recording the ground surface in a forest. Photogrammetry on 
the other hand, can (depending on visibility) only provide a 
surface through the tree tops. With the emerging of airborne 
laser scanning systems, the chance is given to obtain high 
quality DTMs in forested areas. 
The laser beam from an air-borne laser scanner system can 
penetrate the tree tops and therefore a signal can be received, 
which originates from the ground surface. Of course not all 
laser points originate from reflections on the ground. Depend- 
ing on the forest structure, time of flying (season) and tree 
type the penetration rate (i.e. the portion of ground points) 
can range from almost 10096 to 096 [Rieger et al., 1999b]. 
Thus, laser scanner systems provide a point cloud, some of 
the points are ground points, others are so-called vegetation 
points. The aim of this paper is to demonstrate the applica- 
bility of laser scanner data for the derivation of high quality 
DTMs in forested areas. It shall also be made clear, that it 
is worth using a sophisticated filtering method for this end. 
In this paper we will first present a review of the methods 
for laser scanner data evaluation, including the classification 
and interpolation algorithm we developed. A short compari- 
son of the algorithms will be given. Following is a section on 
the performance of our algorithm. This section also includes 
passages valid for many different laser scanner filtering algo- 
rithms. In section 4 two examples will be presented in more 
detail. Results of accuracy investigations will be mentioned 
too. We conclude by presenting an outlook for our research 
activities. 
2 Classification and interpolation of laser scanner data 
2.1 Review of methods 
At the Institute of Photogrammetry at the Stuttgart Uni- 
versity the so-called morphological operator ‘opening’ has 
been applied for the task of evaluating laser scanner 
data [Kilian et al., 1996]. A window is moved over the data 
set. The lowest point in the window is considered to be a 
ground point. All points within a certain height bandwidth 
above this point are considered to be ground points as well. 
They are given a certain weight depending on the window 
size. This is repeated for several window sizes. The last 
step is the surface interpolation (approximation) under the 
consideration of the weights. 
The TerraScan (a module of TerraModeller) filtering method 
is described in [TerraScan, 1999]. An XY-grid is laid over 
the whole data set. The size of this grid has to be specified. 
The lowest point of each mesh is considered to be a ground 
point. These points are triangulated which gives a first repre- 
sentation of the surface. The final surface is build iteratively 
by adding points to this triangulation. Points are accepted or 
rejected according to certain criterion. One criteria measures 
the height of a candidate point above the present surface, 
an other measures the angle between the surfaces with and 
without the candidate point. If certain threshold values are 
reached a point is inserted into the triangulation. 
At the Institute of Photogrammetry and Remote Sensing at 
the University of Karlsruhe an other filtering method has 
been developed [Hansen and Vogtle, 1999]. They propose 
two methods. The first method proceeds by always taking 
the lowest point of a window which is systematically moved 
over the area of interest. For the second method the convex 
hull of the point cloud has to be derived first. The lower 
part’ of the convex hull (in the form of a triangulation) is the 
first approximation of the terrain. Next, points are included 
into the ground surface, if they match certain criteria which 
measure the distance of a candidate point from the present 
surface. 
Generally, the approaches to evaluate laser scanner data can 
be categorised in two ways. 
e The algorithms in the first group perform only a clas- 
sification. A surface model can be derived on the basis 
of the classification, i.e. as the last step. 
e The algorithms in the other group derive a surface 
model. Classification of the points is done with re- 
spect to this surface.