[Back]

G. Mandlburger, C. Briese:
"Automatic derivation of natural and artificial lineaments from ALS point clouds in floodplains";
Poster: EGU 2009, Vienna; 2009-04-19 - 2009-04-24; in: "Geophysical Research Abstracts", 11 (2009), Paper ID EGU2009-9820-1, 1 pages.

Water flow is one of the most important driving forces in geomorphology and river systems have ever since formed
our landscapes. With increasing urbanisation fertile flood plains were more and more cultivated and the defence
of valuable settlement areas by dikes and dams became an important issue. Today, we are dealing with landscapes
built up by natural as well as man-made artificial forces. In either case the general shape of the terrain can be
portrayed by lineaments representing discontinuities of the terrain slope.
Our contribution, therefore, presents an automatic method for delineating natural and artificial structure
lines based on randomly distributed point data with high density of more than one point/m2. Preferably, the last
echoes of airborne laser scanning (ALS) point clouds are used, since the laser signal is able to penetrate vegetation
through small gaps in the foliage. Alternatively, point clouds from (multi) image matching can be employed, but
poor ground point coverage in vegetated areas is often the limiting factor.
Our approach is divided into three main steps: First, potential 2D start segments are detected by analyzing
the surface curvature in the vicinity of each data point, second, the detailed 3D progression of each structure line
is modelled patch-wise by intersecting surface pairs (e.g. planar patch pairs) based on the detected start segments
and by performing line growing and, finally, post-processing like line cleaning, smoothing and networking is
carried out in a last step.
For the initial detection of start segments a best fitting two dimensional polynomial surface (quadric) is
computed in each data point based on a set of neighbouring points, from which the minimum and maximum
curvature is derived. Patches showing high maximum and low minimum curvatures indicate linear discontinuities
in the surface slope and serve as start segments for the subsequent 3D modelling. Based on the 2D location
and orientation of the start segments, surface patches can be identified as to the left or the right of the structure
line. For each patch pair the intersection line is determined by least squares adjustment. The stochastic model
considers the planimetric accuracy of the start segments, and the vertical measurement errors in the data points.
A robust estimation approach is embedded in the patch adjustment for elimination of off-terrain ALS last
echo points. Starting from an initial patch pair, structure line modelling is continued in forward and backward
direction as long as certain thresholds (e.g. minimum surface intersection angles) are fulfilled. In the final postprocessing
step the resulting line set is cleaned by connecting corresponding line parts, by removing short line
strings of minor relevance, and by thinning the resulting line set with respect to a certain approximation tolerance
in order to reduce the amount of line data. Thus, interactive human verification and editing is limited to a minimum.
In a real-world example structure lines were computed for a section of the river Main (ALS, last echoes, 4
points/m2) demonstrating the high potential of the proposed method with respect to accuracy and completeness.
Terrestrial control measurements have confirmed the high accuracy expectations both in planimetry (<0.4m) and
height (<0.2m).

http://publik.tuwien.ac.at/files/PubDat_175580.pdf