Regine Brügelmann
(a) Example of laserdata: dike with ascents (b) Automatically extracted vector
breaklines, not yet smoothed
Figure 1: Objective of approach: automatic derivation of terrain breaklines from laserdata
Up to now breaklines are measured photogrammetrically, which is a time consuming task. Deriving breaklines automati-
cally from laserdata by means of image processing algorithms will facilitate the acquisition of this additional morphologic
information. For this task, however, the density of laserpoints must be much higher than the above mentioned one point
9
per 16 m^.
In this contribution, some methods for the extraction of breaklines from laserdata are shortly described. The suitability
of one of these methods based on hypothesis testing has been investigated. For the assessment of the performance of
the algorithm, the automatically extracted breaklines are compared with photogrammetrically measured breaklines. It is
shown that automatic breakline extraction from airborne laserdata in principle is feasible.
2 METHODS FOR BREAKLINE DETECTION
The output of a range image sensor, e.g. a laseraltimeter, is the scene surface geometry in sampled form. It is most
commonly represented as a two-dimensional array of pixels, each pixel representing the range of a sampled point on
the surface from a reference point. The role of segmentation is to extract geometric primitives relevant to higher level
cognitive processes from the pixel-level representation. In the context of range image segmentation, the simplest possible
geometric primitives are continuous surface patches and surface discontinuities. Likewise, most surface segmentation
techniques can be classified as either edge-based or region-based.
The key idea behind region-based range image segmentation is to estimate the surface curvature at each range pixel
and cluster range pixels with homogeneous surface curvature properties. Against that, the basic idea behind edge-based
range image segmentation techniques is to detect significant surface discontinuities and classify them as jump, crease or
curvature edge (see fig. 2 and tab.1). Jump edges can be detected using standard edge operators designed for intensity
images such as the gradient, Sobel, Kirsch, Laplacian of Gaussian, and Canny edge operators (Suk and Bhandarkar, 1992).
Detection of crease and curvature edges, however, requires specialized operators. Once edge pixels have been detected
and classified, they are linked together to form surface discontinuity contours.
Seeking for breaklines in filtered laserdata, where houses, trees and bushes have been removed, we will concentrate on
edge-based segmentation techniques and, moreover, on the detection of curvature edges because jump edges (e.g. in case
of buildings) hardly appear in filtered data. In the following, we will shortly describe some edge-based segmentation
techniques for range images.
2.1 First derivatives and borders of sloped areas
(Gomes-Pereira and Wicherson, 1999) used first derivatives for the automatic extraction of breaklines from digital sur-
faces. Pixels were labelled as 'slope pixel’ or "flat pixel' corresponding to their tilt in x- or y-direction. Breaklines are
then detected as borders between slope and flat areas by checking the 8-neighbourhood of every flat pixel. If at least one
slope pixel is present, the center pixel is classified as breakline pixel. The resulting breaklines were rather fragmented.
2.2 Laplacian and LoG
In (Gomes-Pereira and Janssen, 1999) the Laplacian operator is applied for the breakline detection task. In comparison
with a manually derived reference dataset, much more breaklines have been detected by the automatic procedure. This
110 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.