Istanbul 2004
ps, Elsevier.
1995), Using
's Residential
able at
(accessed on
patial data, in
Neural Nets:
ver Academic
ser-scanning -
technique for
e$. Journal of
04.
eport: ISPRS
y, Faculty of
University of
at
t05082003.pd
ost-processing
dings ASPRS
CD-ROM, 14
ulating Urban
, 46, pp. 234-
elf organizing
roc. INNC '90,
> Dordrecht,
tion methods,
ume 3(2), pp.
rhankangas J.
^57. Helsinki
aser altimetry
^ and Remote
2
€
TREE CROWN DETERMINATION USING TERRESTRIAL IMAGING FOR LASER
SCANNED INDIVIDUAL TREE RECOGNITION
U. Pyysalo
Institute of Photogrammetry and Remote Sensing,
Helsinki University of Technology, P.O.Box 1200, 02015 HUT, Finland
ulla.pyysalo(ghut.fi
Commission III, WG III/3
KEY WORDS: Method, Development, Image, Tree crown, Error simulation, Laser scanning
ABSTRACT:
A method to measure tree crown dimensions is introduced in this article. The method is based on terrestrial images, which are
captured from two different directions. Image orientation is registered during the image acquisition and this geometrical information
is utilized in image pre-processing. In pre-processing images are corrected from distortion errors and rectified to plane parallel to tree
trunk. Parameters measured from the images are tree height, crown height and crown widths at several heights. Trunk tilting is
calculated estimating trunk as three-dimensional vector and determining its parameters from trunk projections in image planes.
Special attention is paid to orientation errors, which effects dimension measurement. Errors in ®, & and K and error in scale are
analysed separately. Study aimed at developing method, which would not require special instruments, time consuming tacheometer
measurement or positioning of image capture location.
The method development initiated from the need to collect reference material for laser scanned tree crown study. Laser scanning data
has been successfully used for single tree wise forest inventory. In forest area, first returns are typically reflections from the top of
tree crowns. In addition to pulses reflected from tree crown top, several pulses also reflect from crown sides and inside crown.
Utilising these echoes, information below the crown surface is attained. Reference data of captured trees has been previously
obtained with tacheometer and hypsometer measurements. However, these measurements didn't include information below crown
top. With method suggested in this article that missing information is provided. It was expected that reference material provided with
method would be accurate enough to be utilized as reference material for laser scanning tree crown study.
1. INTRODUCTION
The development of laser scanning has offered new challenges
in characterization of individual (Brandtberg, 1999; Hyyppä
and Inkinen, 1999, Lim et al., 2001; St-Onge and Achaichia,
2001, Næsset and Okland, 2002; Persson et al., 2002). To verify
the quality of geometric 3D tree model, however, the effect of
digital terrain model and the behaviour of the laser beam on
hitting the top of the trees, the effect of slope and the effect of
angular trees have to be known. The quality of digital terrain
model in forested areas has been investigated by Kraus and
Pfeifer (1998) and Hyyppä et al. (2000). It has been shown that
the accuracy of the DEM extraction under similar forest
conditions with Toposys-1 is between 14 and 22 cm (Pyysalo,
2000; Ahokas et al., 2002). Height of trees can be measured
with accuracy better than 1 m (Hyyppá et al. 2001). Hyyppä and
Inkinen (1999) demonstrated the possibility to measure single-
tree-wise information using laser scanning and to adopt the
retrieved parameters (height, crown width, tree species) in forest
inventory calculations. In forested areas the effect of slope and
angular trees has been reported by (Heurich et al. 2003). Yu et
al. (2003) demonstrated that even small changes, such as height
growth of forest, are possible using laser scanner data. The
factors affecting the quality of high-pulse laser scanners for
digital target models have been reported by Ahokas et al.
(2002). Rónnholm et al. (2003) reported the integration of laser
data and terrestrial panoramic digital images inspecting the
behaviour of laser scanner data.
In several studies approach is to process the forest canopy as
surface, which is created from laser scanner data. The forest
canopy surface is constructed as TIN, contour or grid models.
However, laser scanning also provides information below the
crown surface. Vertical structure of forest has been studied be
Friedlaender and Koch (2000) by analysing statistically laser
pulse forest penetration. In Pyysalo and Hyyppà (2002) an other
approach was introduced to describe tree crown using vector
polygons circulating around crown sides and line estimating
shape and location of the trunk.
Collecting large amounts of reference data with traditional ways
is time-consuming and expensive. The tree trunk may be
estimated as straight line or cone and its location may be
measured with positioning instruments, such as tacheometer or
GPS-instrument and tree height and crown height may be
measured using hypsometer. The width of the tree crown is,
however, more difficult to measure. In tacheometer
measurements a prism is moved around the crown sides, not just
near the lowest branches but also to the tips of the top branches.
This is carried out climbing to tree or using some lifting
platform instrument instead. Another method is to project
visually tips of the branches to the ground level and measure
their locations with positioning instrument. Afterwards
elevations of the branch tips are measured separately with
hypsometer. Both of these approaches take lots of effort and are
very time-consuming.