Navigation and Tree Measurement
For Your Info
The forestrix project was funded by the Finnish Funding
Agency for Technology and Innovation (TEKES) and participating
companies. The description of the project is based on two
conference publications authored by the Forestrix team ( 4), ( 5).
algorithm handles the distinct tree features in most cases
with sufficient certainty.
While performing the scan depicted in Figure 4, the
scanner has not been level with the ground. In the right-hand side of the figure, a set of large clusters can be seen.
These represent the points where the laser beams have hit
the ground. The algorithm is tuned in a way that these
kinds of outlier points are discarded. The algorithm is quite
picky with the discarding, as was the intention. Only the
most definite features are accepted as trees. This way, some
actual tree features are dropped but now we can be sure
that every passed feature is truly a valid one and its
parameters can be calculated.
Results and Conclusions
The current version of the SLAM software is written in
the Java programming language. The software can read
input data either from data files or from the actual sensors.
The data files are essential for testing the system in a
laboratory environment. The software can connect to the
2D laser scanner and the DGPS receiver. The SLAM
path can be matched to the DGPS path. As a result, the
algorithm also gives the tree positions in absolute map
coordinates. The measured tree diameters are then stored
in the tree map. The resulting absolute accuracy is mostly
limited by the DGPS receiver.
A tree map produced by Forestrix SLAM is shown in
Figure 5. The DGPS path is marked with a narrow line and
the SLAM path is drawn with a bold line. Measured tree
positions are also shown. These positions and diameters
can be compared to hand-measured tree information
provided by METLA (Finnish Forest Research Institute).
During the experimental drive, a total of 277 trees
were observed. The testing environment is depicted in
Figure 6. At a measurement distance of 1 m, the maximum
observed diameter error is 1 mm. The error increases
linearly with distance up to 10 m, where it is 6 mm. After
that, the error increases more rapidly.
The current algorithms implemented in Java run somewhat slower in modern laptop computers than what is required
for real-time applications. Some parts of the presented
SLAM algorithm can be easily tuned for better performance.
Java is a safe language which makes it easy to program
with, but it is not the best programming language perform-ance-wise. Also, interfacing it with exotic hardware such as
2D laser scanners can be difficult. C++ implementation may
have to be considered in later phases of the project.
A forest is a very tough environment for precision
FIGURE 6. One of the experimental environments in a pine forest.
instruments. Luckily, there are 2D laser scanners and DGPS
receivers that are designed for outdoor use. The forest
terrain is often quite rough, which adds additional
challenges for the sensor system. It may be necessary to tilt
the sensor package when the harvester is working in an
inclined position or traversing slopes. More development
work is needed to find better solutions to the association
problem. The current system works well for small loops ( 50
m x 100 m). However, the accumulation of errors may be a
problem for larger loops. Identifying trees may also be a
problem in more dense and cluttered forest environments.
Digital forest imaging has turned out to be a very
challenging research area, but it is essential for future forest
mapping systems which will require features such as tree
species recognition. Digital imaging systems have the ability
to see textures and other tree qualities that laser scanners
cannot perceive. SV
(1) Virtual Finland: Finland at a glance. [Online] [Referenced
on: 8. August 2007.] http://virtual.finland.fi/netcomm/news/
(2) The World Factbook. [Online] Central Intelligence Agency,
16. August 2007. [Referenced on: 31. August 2007.] www.cia.
( 3) forest.fi. [Online] Finnish Forest Association. [Referenced
on: 8. August 2007.] www.forest.fi/smyforest/foresteng.nsf/all
( 4) Tree Measurement in Forest by 2D Laser Scanning. Jutila,
Jaakko; Kannas, Kosti and Visala, Arto. Jaksonville, FL, 2007.
International Symposium on Computational Intelligence in
Robotics and Automation, CIRA.
( 5) Tree Measurement and Simultaneous Localization and
Mapping System for Forest Harvesters. Öhman, Matti & al.
Chamonix, 2007. The 6th International Conference on Field
and Service Robotics.