Airborne laserscanning has proven to be a powerful technique for the detection and modeling of man-made objects such as buildings, which form a substantial part of 3-D city models. As an active technique, laserscanning delivers reliable 3D data points without requirements to surface reflectance variations, thus evading a number of problems occurring with techniques based on 2-D imagery. The publication discusses the use of invariant moments applied to laserscanning data for the determination of roof parameters of simple building types. The technique works on the original, irregularly-distributed laserscanner data points, thus avoiding effects caused by an interpolation to a regular grid. Using only first and second order invariant moments, a number of basic parameters of a building (position, orientation, length, width, height, roof type and roof steepness) can be determined as closed solutions from ratios of binary and height-weighted moments of segmented point clouds. Using higher order moments, more complex roof shapes can be modeled as well. By analyzing differences between point cloud and building model in a second processing step outliers can be detected and systematic deviations from the assumed model such as dorms on a roof can be modeled. The technique was applied to a section of a FLI-MAP laserscanner dataset with an average point density of five points per square meter. No a priori information, such as 2-D GIS data, was used. Instead, the dataset was segmented by a the analysis of height texture measures, followed by morphological filtering and connected component labeling. All detected buildings complying with the assumed simple building types could be detected and modeled successfully. Moreover, most dorms with an extension of greater than two square meters could be modeled in the step of model fit analysis. The precision potential for the building parameters is in the order of 0.1-0.2m for the dimensions and 1-2 degrees for orientation and roof steepness.
[1]
Ming-Kuei Hu,et al.
Visual pattern recognition by moment invariants
,
1962,
IRE Trans. Inf. Theory.
[2]
Roland T. Chin,et al.
On Image Analysis by the Methods of Moments
,
1988,
IEEE Trans. Pattern Anal. Mach. Intell..
[3]
Wolfgang Förstner,et al.
Towards automatic building extraction from high-resolution digital elevation models
,
1995
.
[4]
Scott Mason,et al.
PROJECT AMOBE : STRATEGIES , CURRENT STATUS AND FUTURE
,
1996
.
[5]
Fiona Lang,et al.
3D-city modeling with a digital one-eye stereo system
,
1996
.
[6]
Frank Ade,et al.
Project Amobe: Strategies, Current Status And Future Work
,
1996
.
[7]
Ansgar Brunn,et al.
Extracting Buildings from Digital Surface models
,
1997
.
[8]
Claus Brenner,et al.
Generation Of 3D City Models From Airborne Laser Scanning Data
,
1997
.
[9]
George Vosselman,et al.
Two algorithms for extracting building models from raw laser altimetry data
,
1999
.
[10]
H. Maas.
THE POTENTIAL OF HEIGHT TEXTURE MEASURES FOR THE SEGMENTATION OF AIRBORNE LASERSCANNER DATA
,
1999
.
[11]
Hans-Gerd Maas,et al.
Fast determination of parametric house models from dense airborne laserscanner data
,
1999
.