AUTOMATIC RECONSTRUCTION OF SINGLE TREES
FROM TERRESTRIAL LASER SCANNER DATA
Norbert Pfeifer!, Ben Gorte! and Daniel Winterhalder?
1: Section of Photogrammetry and Remote Sensing, TU Delft, Kluyverweg 1, 2629HS, The Netherlands
2: Institute for Forest Growth, Univ. Freiburg, Germany
{n.pfeifer,b.g.h.gorte} @]Ir.tudelft.nl, daniel. winterhalder@iww.uni-freiburg.de
KEY WORDS: Laser scanning, Close Range, Modelling, Forestry, Automation, Measurement, Algorithms
ABSTRACT:
The investigation of single trees in a forest is of ecological and economical interest. One aim is to capture the geometric
aspects of a tree: the length and diameter of the trunk and individual branches, the change of the radius along the branch and similar
measures. These measures can be determined automatically from terrestrial laser scanner data. The conditions for scanning in the
forest, but also the irregular structure and surface of trees aggravate the reconstruction process. The branches of the trees are locally
modelled by circular cylinders. With the radius, the axis direction and the axis position the main parameters of interest are captured.
We describe a set of algorithms for automatically fitting and tracking cylinders along branches and reconstructing the entire tree.
Especially for coniferous trees the computation of an outer hull, giving the extent in different directions and at different heights is an
alternative, as the dense foliage coverage renders a distinction between branches and needles impossible. Examples for the different
reconstructions of trees are presented.
1 INTRODUCTION
The world forest area covers roughly one quarter to the total land
area of the world. Considered this, it is obvious that forests
play an important role in our lives for ecological and economical
reasons. Assessing various forest parameters is performed with
photogrammetric techniques (optical and radar satellite remote
sensing, aerial photography, and airborne laser scanning), but also
with (terrestrial) field surveys. In this paper a new measurement
method is added to the existing ones, offering the possibility for
objectively determining parameters of single trees.
The tree parameters considered in this study are of geometric type,
including the diameter of the trunk and the branches, but also the
angles between different branches and their location, as well as
the crown diameter are to be determined. These parameters are of
interest due to ecological reasons (habitat investigations, studying
growth reactions to wind and other environmental influences, etc.)
and economical reasons (timber volume estimation for wood pro-
duction, detection and quantization of failures during the growth
process, etc.) and can be described in terms of an ‘as-grown’ anal-
ysis. Current measurement methods are either based on human
estimation and experience (e.g. for crown diameters) or performed
with very simple tools (e.g. tape measurements). Generally, it can
be said that there is a lack of automation in the current method-
ology, making it expensive and subjective (i.e. dependent on the
operator).
Airborne laser scanning with its ability to penetrate the tree crown
cover is investigated and applied in forestry (e.g. (Pyysalo and
Hyyppä, 2002)) for measuring tree height and crown diameter, or
forest height respectively, depending on the data density. Satel-
lite laser scanning, combined with full waveform capturing (e.g.
NASA's ICESat mission) offers the possibility to measure the
biomass in forests. Naturally, these methods provide not much
information on a single tree, but their strength lies in providing
overview information on logs or complete forests. In aerial imag-
ing only the upper crown surface can be seen, but automation
(i.e., detecting and analyzing single trees) is low. With aerial
imaging only the parameters visible from above (e.g. crown di-
ameter, health state) can be determined. Ground based imaging
methods, on the other hand, are not suitable due to the irregular
structure of the tree surface (considering image matching), which
would require further manual processing, and due to the often poor
lighting conditions in the forests. With the advent of terrestrial
laser scanning an active measurement technology — independent
of the sun or an additional artificial light source —, capable of
providing millions of points on highly irregular surfaces is now
available for measuring inside forests.
This paper presents a method for automatic reconstruction of
branches, and therefore trees, from terrestrial laser scanning data.
Special consideration has to be given to their irregular structure
and the problems of data recording in forests. In Section 2.1 the
requirements for the reconstruction are stated and in Section 2.2
the reconstruction methodology is presented. A collection of
algorithms for the reconstruction is presented in Section 3. In
Section 4 examples are presented and discussed.
2 SINGLE TREE MODELLING
2.1 Modelling Requirements and Scanning Environment
The tree model we want to reconstruct has to provide information
on 1) the start point and end point of each branch, and ii) the
radii at these points. This captures in a straightforward manner
the essential measures of the tree components (i.c. the branches),
giving sufficient input to the above described tasks in forestry.
As pointed out in the Introduction, an automated reconstruction
of the tree model is only suitable with terrestrial laser scanning
data. Laser scanner data can be acquired in short time (a few
minutes per scan) and provides a dense point cover on the surface
with an accuracy of a few centimeters or even better. In Fig. 1 a
part of one scan (thinned out) is shown. However, point density
decreases with distance from the device, providing less and less
points in the higher parts of the tree. Likewise, shadow effects
from lower branches will generate gaps in the coverage of higher
branches (cf. Fig. 1 and Fig. 2). Additionally, the chances of
hitting a branch with the laser beam drop with its radius. There-
fore, the outer branches and the higher branches will be covered
[nternationa
EEE
Ly
pe
n
"Peas
Y
at
^
H
>
X
e
~
"T
Figure 1: Po
the point cl
10x10x10cn
side of the ir
on the inner
less dense w
described m;
liage of coni
difficulty. A
other difficu
branches mo
can move a
Finally, one |
scans from d
(Besl and M
each other re
Due to shado
only at the tr
region but ro
remain probl
sharp corner
planimetry it
scene, but di
tie-points his
orientation ir
Summing up
Surements ar
reconstructic
data a difficu
Struction met
on individual
of a tree.
2.2 Recon:
The principle
are described
reality in two
circular, but «
