Elsevier

Computers & Geosciences

Volume 33, Issue 8, August 2007, Pages 1076-1087
Computers & Geosciences

Segmentation and object-based classification for the extraction of the building class from LIDAR DEMs

https://doi.org/10.1016/j.cageo.2006.11.012Get rights and content

Abstract

A new method is presented for the extraction of a class for buildings from light detection and ranging (LIDAR) digital elevation models (DEMs) on the basis of geomorphometric segmentation principles. First, seed cells and region growing criteria are specified. Then an object partition framework is defined on the basis of region growing segmentation. Size filtering is applied to objects and connected components labelling identifies background and foreground objects that are parametrically represented on the basis of elevation and slope. K-means classification reveals a set of clusters. The interpretation of the spatial distribution of clusters assisted by the interpretation of cluster centroids, allows for the identification of the building class, as well as building sub-classes with different geomorphometric characteristics.

Introduction

High-resolution (spacing is less than 2 m) digital elevation models (DEMs) derived by airborne laser altimetry (LIDAR: light detection and ranging) are available for selected areas.1, 2 The development of airborne laser scanning goes back to the 1970s (Irish and Lillycrop, 1999). By emitting a laser pulse and precisely measuring the return time to the source, the range can be calculated using the value for the speed of light.3 Elevations can be derived rapidly and at high resolution from LIDAR in comparison to manual reconstruction from photogrammetric techniques (time consuming, not a cost-effective solution).

The LIDAR DEMs are extremely valuable in the field of earth sciences since elevation data from different flights might be compared to determine the patterns and magnitudes of coastal change4 (erosion, over wash, etc.) and the loss of buildings and infrastructure.5 LIDAR DEMs are used in tectonic studies for fault recognition (Harding and Berghoff, 2000) and topographic change mapping.6 Additionally, airborne laser altimetry data were also used in an attempt to examine and model the surface morphology of landslides (Glenn et al., 2006).

The building class extraction from high-resolution urban DEMs (spacing <2 m) is of primary importance in many applications, including urban planning, telecommunication network planning and vehicle navigation which are of increasing importance in urban areas (Kokkas, 2005). Nevertheless, the building class mapping is of great importance in the earth science field, for earthquake damage assessment. For example the difference DEM with 5 m spacing derived from pre- and post-earthquake aerial images revealed successfully the collapsed buildings, caused by the 1999 Izmit earthquake in Turkey (Turker and Cetinkaya, 2005). The critical element of detecting earthquake-induced heights changes through the DEM differencing method, was deciding where to place the boundaries between change and no-change pixels (Turker and Cetinkaya, 2005). Thus, if a building class was detected from the pre-earthquake DEM then the elevation differences caused by city ruins over the non-building class (roads, city parks) could be omitted.

Methods for building detection and reconstruction from laser altimeter data are applied to the point cloud (irregularly spaced points) and either directly derive the surface parameters in a parameter space by clustering the point cloud or segment a point cloud based on criteria like proximity of points or similarity of locally estimated surfaces (Vosselman et al., 2004; Heuel et al., 2000). Kokkas and Dowman (2006) introduced a method that employs a semi-automated technique for generating the building hypothesis by fusing LIDAR data with stereo matched points extracted from the air–photograph stereo model. The roof reconstruction is achieved by implementing a least squares-plane fitting algorithm on the LIDAR point cloud and subsequently neighbouring planes are merged using Boolean operations for the generation of solid features.

On the other hand, quantitative techniques have been developed in order to automate the interpretation of terrain features from DEMs and various geomorphometric parameters were developed in an attempt to characterize the landscape (Miliaresis and Kokkas, 2004). Geomorphometric segmentation extracted mountains developed on different base levels (Miliaresis, 2001a) and fluvial landforms developed along mountain fronts (Miliaresis, 2001b) from moderate resolution DEMs. The technique is summarized as a region growing segmentation algorithm that uses seeds (e.g. the ridge cells) and growing criteria (usually based on slope and elevation). Whereas geomorphic segmentation defines both abstractions of landforms and an object partition framework of the landscape, the parametric representation of landscape objects on the basis of their spatial three-dimensional arrangement (Miliaresis and Argialas, 2002) allows the recognition of landscape objects) that have similar and contrasting ranges of characteristics, leading to a terrain classification scheme (Miliaresis and Illiopoulou, 2004) according to geomorphologic principles and understanding (Miliaresis et al., 2005).

The aim of the current research effort is to design a new method for extraction of the building class from LIDAR DEMs on the basis of the geomorphometric segmentation principles. Thus, an object partition framework of the LIDAR DEM will be defined. Finally, objects representation on the basis of geomorphometric parameters combined by object classification is expected to allow the extraction and mapping of the building class.

Section snippets

Methodology

First, the study area and its hypsometric characteristics are introduced. Then seeds cells and region growing criteria are defined. Finally, region growing segmentation is performed and thus, an object partition framework is defined. Connected components labelling identifies background and foreground objects that are parametrically represented on the basis of elevation and slope and classified.

Conclusion

Geomorphometric region growing segmentation combined by median filtering for identifying seed cells, connected components labelling, size filtering and object labelling, object parametric representation on the basis of slope and elevation attributes and classification was proved capable of delineating the building class within the study area. The interpretation of cluster centroids allowed the identification building sub-classes with different geomorphometric characteristics that are associated

Acknowledgements

The authors are grateful for, and this paper was greatly benefited from, the thorough and evaluations of the two anonymous reviewers. The authors would like to thank NPA Group for providing the LIDAR data and express their gratitude for the financial support offered by the Greek Scholarship Foundation.

References (26)

  • S. Heuel et al.

    Topological and geometrical reasoning in 3D grouping for reconstructing polyhedral surfaces

    International Archives of Photogrammetry and Remote Sensing

    (2000)
  • Kokkas, N., 2005. City modeling and building reconstruction with Socet Set v.5.2. BAE Systems, Customer presentation in...
  • Kokkas, N., Dowman, I., 2006. Fusion of airborne optical and LIDAR data for automated building reconstruction. In:...
  • Cited by (137)

    • Modelling of buildings from aerial LiDAR point clouds using TINs and label maps

      2019, ISPRS Journal of Photogrammetry and Remote Sensing
      Citation Excerpt :

      The whole process can be executed in 3D or on a 2D support plane. 2D support grids for 3D point processing have been introduced in the past (Miliaresis and Kokkas, 2007; Meng et al., 2009; Lafarge and Mallet, 2012; Sun and Salvaggio, 2013; Chen et al., 2014; Li et al., 2015). These methods project 3D points to grid cells and carry out further processing; other approaches work directly on the 3D point cloud (Overby et al., 2004; Dorninger and Nothegger, 2007; Verma et al., 2006; Oude Elberink and Vosselman, 2009; Xiong et al., 2015; Nan and Wonka, 2017).

    • Tracking the removal of buildings in rust belt cities with open-source geospatial data

      2018, International Journal of Applied Earth Observation and Geoinformation
    View all citing articles on Scopus

    Code available from server at http://www.iamg.org/CGEditor/index.htm.

    View full text