International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
t
QUALITY CONTROL ISSUES OF AIRBORNE LASER RANGING DATA AND ACCURACY STUDY IN AN URBAN AREA
T. Schenkl, B. Csathó?, D.C. Lee!
! Department of Civil and Environmental Engineering and Geodetic Science
?Byrd Polar Research Center
The Ohio State University
schenk.2@osu.edu, csatho.1 @osu.edu, lee.100@osu.edu
KEY WORDS: Photogrammetry, Laser Ranging, Error Analysis, Surface Reconstruction, Segmentation, Fusion
ABSTRACT
The major factors affecting the accuracy of Airborne Laser Scanning (ALS) systems are the errors in ranging, in the
position of the laser firing point, and in the attitude of the laser beam. Since the derivation of a precise theoretical
accuracy model is rather complicated, accuracy estimates are mostly obtained by comparing Digital Elevation Models
(DEMs) derived from laser scanning with reference DEMs. This technique works well on flat or gently sloping terrain,
however it is not suitable for complex 3D terrain. For example, small displacements of the laser footprints can cause
large elevation errors around tall buildings. Large range errors can occur when the elevation or brightness varies
within the footprint, such as along the boundaries of buildings, or around trees. Moreover, other non-sensor related
factors, such as the point distribution, the post-processing algorithms, and the extracted features also effect the
accuracy. We first present a general quality control scheme, followed by analyzing the accuracy of ALS over urban
terrain. The analysis is performed low altitude aerial photographs. First, a reference DEM was measured from the
aerial photographs on the analytical plotter. To facilitate the visual comparison between the reference DEM and the
surface points from ALS, both the reference DEM and the laser points were back-projected onto the aerial stereo
photographs. The differences between the two surface descriptions were obtained in the classical way by comparing
the laser points with the reference DEM. This standard approach is not suitable in areas of large elevation changes. A
better way to compare the accuracy is to extract features and to compare them. We compare planar surface patches
and 3-D lines obtained from intersecting planes.
1 Introduction
Airborne laser ranging, with its high accuracy potential
and dense sampling, is a technology successfully used
in an ever increasing range of applications. Originally
employed for ice sheet monitoring, DEM derivation in
forested areas, and biomass computations, new applica-
tions, for example in mapping urban areas are explored.
The information that constitutes the result of an applica-
tion, e.g. a DTM including breaklines, or buildings for a
city model, is not explicitly available in the raw laser data
set; rather, it must be extracted. The question of how
well features can be extracted and how accurate they are
is a quality control aspect.
There is no redundant information available when com-
puting the 3-D positions of individual laser points.
Hence, no explicit quality measure exists at the outset;
we rely on the assumption (based on experience) that the
points are good. Another characteristic of raw laser data
is its random distribution with respect to object bound-
aries. It would be sheer coincidence if a laser shot had
hit the boundary of an object to be mapped; even if it
had we would not know because laser points carry no
information about objects.
We present a third argument in support of developing a
general quality control scheme for airborne laser rang-
ing data and derived features. Imagine we have aerial
imagery and laser ranging data of the same scene. Im-
agery is immediately accessible by our visual system for
analysis and interpretation. However, this is notthe case
for the laser data for humans do not have a sensory sys-
tem that would response to range data—we cannot di-
rectly interpret raw laser points and quickly decide if the
data make sense. A transformation into a more suitable
representation is required.
The next section elucidates quality control issues related
to airborne laser ranging. It begins with an assessment
of raw laser points. As the abstraction level increases,
the quality of extracted features is increasingly influ-
enced by the post-processing algorithms. Apart from
the data, it is important to include the algorithms into
the quality control.
We have performed several experiments in the Ocean
City test site, established by ISPRS Working Group IlI/5
(Csathó et al. (1998)). Surfaces obtained from laser
scanning systems are compared with photogrammetric
measurements, carried out on analytical plotters and
softcopy workstations. This paper describes the experi-
ments, reports the results, presents an analysis and out-
lines future research.
2 Background
Raw laser data is hardly ever used as an end result. Usu-
ally, information is extracted during various post pro-
cessing steps. Fig. 1 depicts the major processing steps
(see also Schenk (1999a)).
Quality control should be conducted on every stage, be-
ginning with an assessment of the raw laser points, in-
cluding blunder detection, to an analysis of extracted
features and derived surface properties. Such analysis
may range from simple plausibility checks to thorough
error studies. We distinguish between qualitative and
quantitative methods.