Integrating building footprints and LiDAR elevation data to classify roof structures and visualise buildings Cici Alexander a, * , Sarah Smith-Voysey b , Claire Jarvis a , Kevin Tansey a a Department of Geography, University of Leicester, University Road, Leicester, LE1 7RH, UK b Ordnance Survey, Romsey Road, Southampton SO16 4GU, UK article info Keywords: Building footprints LiDAR 3D Visualisation Roof type TIN abstract Three-dimensional urban models are increasingly needed for applications as varied as urban planning and design, microclimate investigation and tourism. Light Detection And Ranging (LiDAR) data are con- sidered to be highly suitable for the three-dimensional reconstruction of urban features such as buildings. Ongoing research is determining how best to integrate LiDAR elevation data with already available vec- tor-based data. This paper reports research on combining building footprints and LiDAR to visualise an urban area (Portbury near Bristol, England) with an emphasis on representing buildings in a GIS environ- ment. The main emphasis here is on retaining a vector model that is suitable for representing regular man-made structures. A major difference between this and earlier work is that before visualisation, this work classifies roof types of buildings as either flat or pitched. We compared LiDAR data at three point densities in terms of successful building type detection and visualisation: 1 (low), 16 (medium) and 40 (high) points per m 2 . There are important data acquisition cost issues at each of these resolutions. High density LiDAR yielded the highest overall accuracy of building type detection and proved useful for iden- tifying roof features, yet lower densities proved more useful for revealing overall roof morphology. Ó 2009 Elsevier Ltd. All rights reserved. 1. Introduction Urban areas are continuously changing due to construction and extension of features such as buildings (Morgan & Habib, 2002). Three-dimensional (3D) urban models are increasingly needed for various applications such as urban planning and design, pre- visualisation of new developments and utility-service planning, microclimate investigation, telecommunication, noise simulation (Zhou, Song, Simmers, & Cheng, 2004), urban regeneration, tourism and visualisation of the urban environment (Baltsavias & Gruen, 2003; Vosselman, 2002). Accurate roof morphologies are required for applications such as rain–runoff modelling (Smith, 2003), as well as those employed by the telecommunications industry. De- tailed rain–runoff models require information about roof shape, roof slope gradient, and roof surface material, as well as artificial drainage networks (such as gutters) that form part of the roof. The accurate extraction of such information from remotely sensed images is likely to open up a range of potential GIS applications (Smith, 2003). Light Detection And Ranging (LiDAR) is a relatively new remote sensing technique that is revolutionising topographic terrain map- ping. LiDAR is an active system that uses laser technology to reflect pulses of light from an aerial sensor to the ground surface (see Lillesand, Kiefer, & Chipman, 2004). The specific use of LiDAR data for urban modelling and visualisation has received increased atten- tion in recent years, primarily due to closer integration with Global Positioning System technologies (Palmer & Shan, 2002). LiDAR data is very suitable for 3D reconstruction of urban features such as buildings (Alharthy & Bethel, 2002; Verma, Kumar, & Hsu, 2006; Zeng, 2008; Zeng, Lai, Li, Mao, & Liu, 2008). While national map- ping organisations have had an historical emphasis on quality two-dimensional (2D) products, in many countries including Great Britain the potential for extending mapping capability to capture the 3D characteristics of objects is of great interest. In this context, airborne LiDAR is being investigated as a potential data source for providing roof morphology and height information for integration with traditional 2D products (Holland, 2002). From a review of the relevant literature, it is clear to us that the main problems in creating a 3D urban visualisation from LiDAR data are in the detection of building edges and in the interpolation of heights. In the absence of building footprint data, building boundaries have been approximated from LiDAR data (Alharthy & Bethel, 2002; Cho, Jwa, Chang, & Lee, 2004) or digitised from aer- ial photographs (Palmer & Shan, 2002). Steed, Spinello, Croxford, and Milton (2004) and Tse, Dakowicz, Gold, and Kidner (2005) used building boundaries from OS Landline Ò , an Ordnance Survey topographic vector product now replaced by OS MasterMap, to create an urban model. However, these studies did not consider 0198-9715/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.compenvurbsys.2009.01.009 * Corresponding author. Tel.: +44 0 116 252 5148; fax: +44 0 116 252 3854. E-mail addresses: ca90@le.ac.uk (C. Alexander), Sarah.Smith@ordnancesurvey. co.uk (S. Smith-Voysey), chj2@le.ac.uk (C. Jarvis), kjt7@le.ac.uk (K. Tansey). Computers, Environment and Urban Systems 33 (2009) 285–292 Contents lists available at ScienceDirect Computers, Environment and Urban Systems journal homepage: www.elsevier.com/locate/compenvurbsys