Geography Reference
In-Depth Information
set of grids (C++/Python). Each grid has different resolution to lower hardware
requirements. Next, the raw LIDAR scan is analysed from the point of view of the
slope factor and is used further to classify interpolated grids into groups. Classified
grids are processed by 3D scripting language to form 3D polygonal digital terrain
model. The procedure was described in detail by Hovad et al. (
2013
).
Resulting DTM is used in the end of the whole procedure together with other
partial models (vegetation and clouds) to form a resulting photo-realistic scientific
model of the surface based on the real-world values scanned by LIDAR.
Data Filtering and the Tree Recognition
Authors use simple method to extract the needed information from the LIDAR data.
Basic vegetation and ground recognition is based on the scattering or gradual
reflection of the LIDAR laser ray from the targeted objects. The ray, which is
wholly returned from the first reflection, is classified as a ground. In some cases, the
ray is partially reflected and returned more than once. In these occurrences, objects
are classified as vegetation because the ray propagates down through the treetops,
which increases the reflected counter. The last reflected pulse is with a high
probability the ground as well. There are enough points available within the
LIDAR data to calculate minimums or maximums for the area of interest to increase
precision and exclude those rays that do not reach the ground entirely.
It is possible to filter raw LIDAR data to select only the points that are reflected
first and leave behind the rest of the point cloud, which includes points scattered
through the treetops. The scattered points represent the
inner layer
. The result of the
selection (firstly reflected points) creates a
cover layer—
the top layer describing
ground and the highest occurrences of treetops. As it was mentioned before, the
irregularly distributed points form an
inner vegetation layer
which is later used to
create the base for trees. As the next step, this irregular point cloud is transformed
into the point grid by means of the maximum function that aggregates points in the
given area together. An appropriate spatial resolution of the raster must be chosen at
this moment too. One meter resolution is used in this case. Resulting grid introduces
some uncertainty. This uncertainty is a sparse in the given square area of each grid
cell. In this case, each tree has a 1 meter spatial tolerance of a measuring accuracy.
This spread is sufficient in the case of reconstructing large vegetation areas like
forests. Also, it reduces the point cloud and simplifies the next data processing (see
Fig.
1a
).
The point grid cover layer includes treetops along with terrain. Data have a form
of a matrix, which can be filtered by a customized filter to obtain
treetops layer
.
Simultaneously, vertical or/and horizontal elevation is computed as a difference
between two adjacent points (see Fig.
1b
).
The above mentioned filter is quite simple, written by authors in the Python 2.7.
The code is object oriented and easy to read because of the Python features. The
main idea is to use Hash-table (dictionary—key value pair) to store and process the
