In: Wagner W„ Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
422 
2.2 Pros and Cons 
The acceptability and utilization of true orthophotos should be 
compared by the pros and cons of the process and their 
properties. 
The high investment of the flight, the calculating process and 
the high requirement of the hardware is accompanied by the 
higher automatism and the endproduct. The output is an 
orthophoto without any disposal or occlusion in the image, 
where all visual content will be shown in the correct location. 
Furthermore, the long time needed seam line digitalisation of 
normal orthophotos to correct geometric mistakes is not 
applicable. Uses, like digitalisation or classification are easier 
and more detailed. It does not matter if the images are RGB or 
CIR. In both color spectrums, true orthophotos can be 
generated. 
The nicest secondary effect is that the process of the true 
orthophoto makes a high resolution surface model as a co 
product in the same resolution as the orthophoto. This DSM can 
be used by other applications and for analysis of questions. 
The DSM can be the base for generating a digital elevation 
model. But it must be said that in large closed areas of 
vegetation, there are no points on the ground. In this case the 
generation of the DTM is very difficult and needs other data or 
more algorithms to calculate and interpolate the ground height. 
A semi - automatic process is needed in these cases. With free 
sight to the ground there will be an accuracy and a level of 
detail not reached by normal models made by a laserscanner. 
Mobile objects, like cars or motor bikes will be displayed 
diffused, confusing some users. 
3. AUTOMATIC TREE EXTRACTION AND 
CLASSIFICATION IN FORESTRY 
Especially in the forestry sector, the working with true 
orthophotos has proved itself and is the base for the forestry 
data acquisition in the house of FMM. This includes automatic 
tree extraction, inventory differentiation and volume 
calculation. 
This information must be updated and renewed permanently, 
because the forest industry has to react quickly to exterior 
factors like bug infestations, storm loss and deforestation. A 
mostly automatic process on the base of true orthophotos is a 
maintask of the work of FMM. 
3.1 Datastructure 
For the analysis of forestry information, FMM uses true 
orthophotos with resolution of 12cm and their digital surface 
models. With the characteristic of the aerial camera 
(UltracamX) to make multi spectral images, it is possible to get 
CIR images in addition to RGB images simultaneously. 
The images in the near-infrared range are used for the 
classification of vegetationdata. 
This grid data will be split in 500*500m tiles for faster 
analysis. 
Furthermore a DTM from a laserscan flight are used to 
normalize the surface data. That is affected by the region, where 
the data will be analysed. 
In mountainous regions it is advisable, but in a flat region, a 
normal interpolation can be used. 
3.2 Tree extraction 
Generally the tree extraction in the true orthophoto is easier 
than in normal orthophotos, because the tree have no disposals 
and are in correct position in picture and DSM. 
The combination of DSM and DTM from laserscanning flight 
provides a quality control between both models and as well as a 
standardization of forestareas. 
With that way it can be taken the relative height of the tree 
exactly and fast, which is not unimportant for the further 
development. A DTM is essential, because the consistence of 
the terrain in big forestareas can be catched in the produced 
DSM. 
Based on an objectoriented algorithm, which registered local 
maximum in the DSM, will be detect the height and height 
changes, what from possible treetops can be suggested. 
In addition to the position of the tree, information about the 
characteristics and the species of the trees can be detected. The 
algorithm was developed by FMM in cooperation with some 
universities in Austria and optimized for the high requirements 
of hardware by FMM. High performance the hardware is 
needed, because of the large amounts of data. With an adequate 
tile format for the model, the working process can be controlled 
and accelerated. 380000 trees can be extracted in around 20 
hours. That means an area with around 16 square kilometers. 
After detection of all heights and topforms of the trees, it is 
possible to make a segmentation of the species of trees. In the 
first step it can differ between broadleaf tree and conifer. In a 
second step with the help of the CIR images, it can be analyzed 
and filtered for the species of the tree. Every species has its own 
characteristics. 
With example measurements and available data, an accuracy of 
90 - 95 % of the extraction in matured timber could be proven. 
This statement relates to the mass of the timber in a body and 
the object extraction as comparably. The last 5-10 percent of 
error is a result of not visible trees in the area of vegetation. 
Figure 5. Tree height detection