Two algorithms for extracting building models from raw laser altimetry data
Introduction
The automation of the generation of 3-D city models as required by many users of geographic information systems has become a major focus of photogrammetric research in the past few years. Starting with 2-D image processing techniques, researchers did soon turn towards 3-D approaches like grouping features matched in multiple images. Both semi-automatic (e.g., Lang and Förstner, 1996) and fully automatic photogrammetric approaches (e.g., Henricsson et al., 1996) have been developed. The automatic extraction of parametric and prismatic building models from dense digital elevation models generated by photogrammetric techniques or airborne laser scanning has been shown by Weidner and Förstner (1995).
Due to its advantages as an active technique for reliable 3-D point determination without requirements towards surface reflectance variations and time consuming error-prone matching techniques, airborne laser scanning has meanwhile become a rather important source of information for the generation of 3-D city models. Although the point densities delivered by most systems in standard operation mode are still too small (often in the order of 1 point/10 m2), sufficiently dense datasets can be acquired with several systems today.
A number of authors has shown approaches for the generation of 3-D building models mainly or solely based on laser altimetry data: Haala and Brenner (1997)extract planar roof primitives from dense laser altimetry data (TopoSys sensor, 4 points/m2) by a planar segmentation algorithm, using additional ground plan information for gaining knowledge on topological relations between roof planes. Lemmens et al. (1997)show the fusion of laser-altimeter data with a topographical database to derive heights for roof-less cube type building primitives. Hug and Wehr (1997)show the detection and segmentation of houses from ScaLARS height and reflectivity data based on morphological filtering with successive progressive local histogram analysis; in addition, they use the laser reflectivity measure for separating buildings from vegetation. Haala et al. (1998)derive parameters for 3-D CAD models of basic building primitives by least-squares adjustment minimising the distance between a laser scanning digital surface model and corresponding points on a building primitive; the boundaries of buildings are derived from ground plans. The implementation is limited to four standard building primitives and combinations of those. Further refinement has to be performed interactively.
In the following, we show two new approaches for the automatic derivation of building models from laser altimetry data. The first approach is based on the analysis of invariant moments of point clouds: closed solutions for the parameters of standard gable roof house types are derived from 0th, 1st and 2nd order moments, and asymmetries such as dorms on the roof are detected and modelled. The second approach is a data driven technique based on the intersection of planar faces in triangulated point. Common to both approaches is the fact that they use the original 3-D data points rather than data interpolated to a regular grid, thus avoiding unwanted effects caused by interpolation. Moreover, they can be applied to laser altimetry data without the need of any further source of data, such as 2-D GIS data. As an option, such data may be used to strengthen the technique or to warrant consistency with available information.
Section snippets
Segmentation
With either interpolated or original data, the laser altimetry data has to be segmented as a first processing step. In our approach, segmentation means only the extraction of a point cloud describing a building from the laser altimetry data. Such a segmentation of laser altimetry data may be obtained in several ways.
⋅ In many cases, available 2-D GIS information can be used as reliable and usually sufficiently accurate source of information—with the advantage of data integrity but the
Derivation of house model parameters from invariant moments
The analysis of invariant moments has been used in image processing for a long time. Early publications go back to the 1960s (e.g., Hu, 1962).
In the continuous domain, moments are defined as:with i,j the order of the moment and f(x,y) the continuous image function.
Airborne laser scanning delivers discrete, irregularly distributed 2 1/2-D point data with the height H as a function of planimetric coordinates X and Y. Height-weighted moments can be computed by
House model reconstruction by intersection of planar faces
More generically, polyhedral models can describe most buildings. Therefore, planar faces describe the surface of a house roof. Because of the high density of laser scanner data, these faces can be recognised clearly and the parameters of the planes can be estimated accurately. The outlines of the faces of a roof are more difficult to determine. If the roof surface is continuous (e.g., there are no height jumps on the roof), the edges between adjacent faces can be reconstructed by intersection
Practical results
The techniques shown above were applied to a FLI-MAP laser altimetry dataset containing 51 buildings. FLI-MAP (Fugro-Inpark, see, e.g., Pottle, 1998) is a helicopter-based laser scanning system with 8 kHz sampling rate. It acquires 40 profiles/s with 200 points/profile. Range measurement is limited to first-pulse measurement at 20–200 m distance, thus providing a maximum strip width of 200 m at a scan width of 60°. Due to these system parameters, the point density is usually rather large (more
Conclusions
Dense laser altimetry datasets with a point density of 1 point/m2 or higher depict a very valuable source of data for the automatic generation of 3-D city models. Based on the computation of invariant moments, closed solutions can be formulated for the determination of the parameters of simple building models, yielding a precision of 0.1–0.2 m for the building dimensions and 1–2° for the building orientation and the slope of roofs. Going beyond primitive house models, techniques based on the
Acknowledgements
The authors would like to thank the Survey Department of Rijkswaterstaat in Delft for providing the FLI-MAP laser scanner data used in this study.
References (13)
- et al.
Towards automatic building reconstruction from high resolution digital elevation models
ISPRS Journal of Photogrammetry and Remote Sensing
(1995) - et al.
Algorithms for the reduction of the number of points required to represent a digitized line or its caricature
Canadian Geographer
(1973) - Haala, N., Brenner, C., 1997. Generation of 3D city models from airborne laser scanning data. Proc. 3rd EARSEL Workshop...
- et al.
3D urban GIS from laser altimeter and 2D map data
IAPRS
(1998) - et al.
Project Amobe: strategies, current status, and future work
IAPRS
(1996) Visual pattern recognition by moment invariants
IRE Transactions on Information Theory
(1962)
Cited by (409)
Building outline extraction using adaptive tracing alpha shapes and contextual topological optimization from airborne LiDAR
2024, Automation in ConstructionUse of UAV-based photogrammetry products for semi-automatic detection and classification of asphalt road damage in landslide-affected areas
2021, Engineering GeologyCitation Excerpt :In Phase III, the dense point cloud is cropped to extract the road surface and remove vegetation and traffic barriers. The pavement point cloud is then processed to obtain i) the deviation of each 3D point from a reference planar surface, and ii) its covariance matrix and geometric features (Maas and Vosselman, 1999). First, to identify hotspots on the road pavement where anomalies are expected, a reference road surface is constructed by progressively fitting a 6-by-1-m plane perpendicular to the road central line along the roadway (Fig. 3a).
Contribution of Colour Information to the Classification of 3D Points from Aerial Images
2024, Arabian Journal for Science and EngineeringComplex Roof Structure Reconstruction by 3D Primitive Fitting from Point Clouds
2023, Journal of Geo-Information ScienceSurface-volume-bottom joint-filtering algorithm for Airborne LiDAR bathymetric point cloud
2023, Cehui Xuebao/Acta Geodaetica et Cartographica Sinica