TREE MODELING WITH DSM DATA

Keonsoo Park, Jehoon Park, Choi Ji-Hoon, Sun-Jeong Kim and Chang-Geun Song

Department of Computer Engineering, Hallym University, Gangwon-do, Korea

Keywords: LiDAR, Tree, Forest, DSM, DEM.

Abstract: This study aims to resolve the problem of not being able to directly examine forests or each individual tree

of a forest. In order to get the specific information on actual trees. Such as their locations, heights and the

number of trees. We used an aerial photograph that is 4096x4096 pixels. And process the DSM/DEM data

with a raw 16 bit-‘unsigned short’ data value. Through the collected information, we might model trees and

a forest.

1 INTRODUCTION

Tree modeling refers to acquiring important

information on trees such as location, height, width

and the number of trees. LiDAR (Light Detection

And Ranging) is a space information acquiring

technology where many laser pulses (70 KHz) are

shot from a plane and the reflected laser pulses are

used to acquire high definition height information of

the surface to create an accurate model of the laser.

It is a new measurement technology that can be used

to acquire high quality 3D digital data. Tree

modeling with LiDAR data will increase efficiency

when designing golf courses with lots of trees. This

study introduces a solution to not being able to

directly examine or explore forests and each

individual trees by using an 4096X4096 aerial

photograph and processed DSM/DEM data with a

raw 16 bit-‘unsigned short’ data value a tree

modeling method.

2 PREVIOUS STUDIES

Tree modeling has previously been studied in a

variety of different ways in the field of remote

sensing. First, there was the study where a sectional

maximum filter was applied to a certain band and

the satellite image resolution value was used to

predict the central point of trees. There was also the

study where an aerial photograph and LiDAR data

were converged and a division method was used to

estimate the height of each tree. Although both these

methods provided adequate solutions, they failed to

provide a way to isolate and explore individual trees

within the data collected.

3 METHOD AND RESULTS

Figure 1 shows the process for tree modeling used in

this study. Previously, three parameters were

produced for modeling purposes location, height and

the number of trees. We expect to be able to enhance

the existing modeling techniques to be able to

differentiate the kinds of trees such as broadleaf or

needle leaf trees. Our method consists of 4 major

stages; First, extraction of areas where there are

predicted to be trees; second, calculation of eigen

value; third, elimination of errors in division; fourth,

division of basin.

Figure 1: Tree modelling process.

189

Park K., Park J., Ji-Hoon C., Kim S. and Song C..

TREE MODELING WITH DSM DATA.

DOI: 10.5220/0003840401890192

In Proceedings of the International Conference on Computer Graphics Theory and Applications (GRAPP-2012), pages 189-192

ISBN: 978-989-8565-02-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

The tree modeling uses DSM input data that

consists of a raw data value that is a 16 bit –

‘unsigned short’ value. The entered DSM data is

used to establish an image pyramid, as shown in

Figure 2.

Figure 2: Image pyramid.

If an image pyramid as shown above is used,

areas where trees are expected to be found can be

sought out in advance from a smaller image. Then,

when the smaller image is used on a higher level,

areas that have already been processed will not go

through the process of finding areas with trees. This

process will result in increased processing speeds

and enhanced image data.

Figure 3: Difference of Gaussian.

Areas expected to have trees are found using a

DoG (Difference of Gaussian) filter with curve

characteristics as shown in Figure 3. From the image

that goes through DoG filtering we then looks for a

sectional maximum value or minimum value. The

image is a 16 bit height map, so exceptionally higher

or lower values than the surrounding areas are

extracted for areas with trees.

The top left block of Figure 4 shows part of a

color aerial image, and the top right is the DSM data

with a 16 bit height value that will be used for tree

modeling. The DSM data is used to calculate a

normal vector and create a visual presentation, and

the screen shot shown on the bottom left block is

gained. If a DoG filter is applied to the DSM data,

an image as shown on the bottom right appears. The

bright area in the center is the sectional minimum

value at [-65535, 0], and can be evaluated as the

location of trees. In addition, a simplified box filter

as shown in Figure 5 can be used instead of a

Gaussian filter for increased speed.

Figure 4: Visual presentation of areas expected to have

trees.

Figure 5: DoB (Difference of Box) results.

The location of trees is predicted by using the

sectional maximum or minimum values of an image

after it has gone through the DoG or DoB filters, so

creating a visual presentation of the predictions leads

to the image shown below in Figure 6.

Figure 6: Sectional maximum/minimum values.

The red areas in Figure 6 are areas predicted to

have trees. It shows both groups and individual

trees. The area is classified to be a group area when

the red area is greater than a predefined width. After

this processed, a watershed algorithm is applied to

divide it into individual trees. If the divided area is

less than a predefined minimal width, it is classified

as a single tree, and the process of finding out what

type of tree begins.

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

190

Figure 7: Classification of single tree and group trees.

4 CONCLUSIONS

An aerial photograph 4096x4096 in size and

processed DSM/DEM data with raw data value that

is a 16 bit-‘unsigned short’ value were used in this

study for tree modeling. This study used a

4096X4096 photo, processed DSM/DEM data, and a

raw 16 bit-‘unsigned short’ data value to create the

tree model, The process involves extracting areas

expected to have trees, and applying a basin division

algorithm is to extract results for tree modeling, but

stopped before identifying the type of tree. The

proposed process is largely divided into 4 steps- :

extraction of areas expected to have trees, eigen-

value calculation, elimination of errors in division,

and division of basin. The information on trees

gained through this process is used to output Figure

8 as a result.

The four processes proposed in this study are

used to gain important elements for tree modeling,

with the exception of identifying the type of trees.

Figure 8 has a random tree type because a specific

tree type analysis was not conducted. Tree modeling

may be improved in the future through studies on

identifying the type of trees.

ACKNOWLEDGEMENTS

This research was financially supported by the

Ministry of Education, Science Technology (MEST)

and National Research Foundation of Korea(NRF)

through the Human Resource Training Project for

Regional Innovation.

Figure 8: Results output.

REFERENCES

M. A. Wulder, J. C. White, K. O. Niemann, T. Nelson,

2004. Comparison of airborne and satellite high spatial

resolution data for the identification of individual trees

with sectional maxima filtering. Taylor & Francis vol

25.

Juan C. Suarez, Carlos Ontiveros, Steve Smith, Stewart

Snape, 2005. Use of airborne LiDAR and aerial

photography in the estimation of individual tree

heights in forestry. Computers & Geosciences 31.

Timothy L. haithcoat, Wenbo Song and James D. Hipple,

2001. Building Footprint Extraction and 3-D

Reconstruction from LIDAR Data. IEEE/ISPRS Joint

Workshop in Remote Sensing and Data Fusion over

Urban Areas.

Jean Cousty, Gilles Bertrand, Laurent Najman, and Michel

Couprie, 2010. Watershed cuts: thinnings, shortest-

path forests and topological watersheds. IEEE

Transactions on Pattern Analysis and Machine

Intelligence.

TREE MODELING WITH DSM DATA

191

Geun-sang Lee, Deok-gu Koh, Hyo-seok Chae, Young-ho

Shin, 2003. “A Comparative Study on DEM Errors

According to Resolution of Aerial Laser Measurement

Material,” Korean Geographical Information Society

Journal.

Open Source Computer Vision Library Reference Manual,

Intel., sourceforge.net/projects/opencv/.

Imagirae Takashi. DirectX 9 Shader Programming. Seoul:

Hanbit Publishing.

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

192