Processing of Airborne Laser Data and Images
Arttu Soininen
Software Developer Hannu Korpela,
Marketing and Sales Terrasolid Ltd., Finland
info@terrasolid.fi
The workflow for processing airborne laser data (LiDAR) and airborne images may be divided into six major
steps: Initial setup, Calibrating data, Classifying points, Processing images, Validating
positioning and Creating delivery products.
1.INITIAL SETUP
The initial setup involves importing all
the necessary raw data into the processing
software, applying coordinate
transformations, organizing the data,
throwing out unnecessary information
and checking
that the project
area has
been covered.
The number
of points in a
laser survey
project may be
anything in
the range from
5 million to 50
billion. A large
data set needs
to be divided
into smaller,
more manageable
geographical
blocks.
About 5-20
million points
is suitable
block size which will fit in random
access memory and still leave room for
information that various processing
routines need to build internally.
Many of the processing steps can be
executed as batch processes without
human supervision. These automatic
tasks can be speeded up by distributing
the computation to several computers
on the network. No matter how sophisticated
the automatic routines are, the
human operator always has to do visual
checking of the results and fix problem
locations interactively. The ability
to view the results of each step in a fully
3D environment is the key to producing
accurate models. Laser data needs
to be viewed in top views, cross section
views and in views with freely selected
3D rotation using different colouring
modes: colour by class, intensity value,
elevation or flight pass.
To ensure proper classification, the
operator needs to be able to see the
laser data overlaid with orthophotos or
individual airborne images.
2.CALIBRATING DATA
All laser scanner owners calibrate
their instrument but this is not enough
to achieve accurate positioning. The
calibration parameters need to checked
for each flight session. In a sense,
the laser scanner owners are constantly
fine tuning the calibration of the
sensor. Calibration is based on comparing
the laser data produced by different
flight passes which overlap each other.
To make this task possible, each project
flight session must include some flight
passes which overlap other flight
passes. To increase the amount of comparison
area, one normally flies some
crossing flight passes for calibration
purposes.
Aerial sites will have side overlap
between parallel flight passes automatically.
Corridor sites are more
demanding when it comes to mission
planning to ensure that the data can be
internally checked and calibrated. It is
not enough to fly a corridor object in
one direction only. One should add
crossing flight passes at regular intervals
(5 - 20 km).
The calibration is normally based on
surface to surface matching of the different
flight passes. As preparation
step, one has to classify ground in each
flight passes separately to remove the
noise that vegetation would bring into
the comparison. This classification can
often be done as an automatic batch
process. The most common, basic
matching steps are:
Solve misalignment angles between laser scanner and IMU together with
scanner mirror scale. This step can be done using only some selected blocks from the project.
Solve dZ correction for all flight lines.It is very common that some flight passes are a few centimeters too high and
some a few centimers too low.
Only when the matching of flight passes is complete, should one continue with rest of the processing steps.
3.CLASSIFYING POINTS
Classifying points is a task where we
try to determine what type of object
each laser point is a reflection from.
This task is often the step which consumes
the most operator time. Even
though automatic routines will do over
90% of the work, this will still leave
millions of points where the human
operator has to make classification
decisions.
The survey flight will often produce
data which is not needed in the final
product. The operator will want to classify
these points out of the active data
set. Points may be excluded because
they:
are outside the project area
are from the overlap area where points from another flight pass will be kept
are lower positional quality due to weather conditions or some other reason
The level of classification detail varies
greatly from one project to another. In
many projects, the only delivery product
is a ground model and perhaps contour
drawings generated from the
model. In those cases, 5-8 classes is all
that is needed.
Low vegetation, Medium vegetation
and High vegetation classes will not
mean that the object is necessarily vegetation.
Points in these classes will
include hits on other surface objects as
well: cars, trains, lampposts, wires etc.
Some engineering type projects may
have more than 50 classes into which
points need to be classified. The more
detailed the classification, the more
operator time is required.
Usually classification is based on first
running automatic routines and then
performing interactive editing of the
results. The interactive editing is a step
where the user needs to be able to view
the data set in flexible ways and to
view other information sources at the
same: orthophotos, existing vector
maps, old surface models etc.
Orthophotos are particularly important
to ensure proper handling of the
laser data. Images are essential to
understanding what the laser scanner
has captured.