Extracting Accurate Building Information from Off-Nadir VHR Images

Abstract

GIScience 2016 Short Paper Proceedings Extracting Accurate Building Information from Off-Nadir VHR Images A. Suliman, Y. Zhang, R. Al-Tahir Department of Geodesy and Geomatics Engineering, University of New Brunswick, Canada, E3B 5A3 Email: {a.suliman; yunzhang; riad.altahir}@unb.ca 1. Introduction Since buildings are the most prominent class in urban areas, updated building information plays an important role in urban planning and management applications. The most cost- effective and broadly available geo-data for mapping building information is the very high resolution (VHR) satellite images. Thus, building detection in VHR images has been an active research area for the remote sensing community during the last two decades (Doxani, Karantzalos, and Tsakiri-Strati 2015; Khosravi, Momeni, and Rahnemoonfar 2014). VHR optical images are the 2D perspective projection of the real surface. Hence, they lack the elevation information. Since buildings are elevated objects, image and elevation data co-registration is required for accurate and reliable detection results. However, several problems are introduced when such data types are integrated (Dowman 2004). Elevation data are typically generated by the photogrammetric approaches and LiDAR (light detection and ranging) technology. Both of these sources provide the height information of the tops of surfaces such as buildings or trees. Hence, they result in digital surface models (DSMs). DSMs are generated as an orthographic projection (nadir view), while VHR images are usually acquired as perspective projections with across-track/along- track angle (off-nadir view). This projection difference makes accurate co-registration very difficult to achieve. The co-registering methods of image and elevation data are extensively reviewed in our previous work (Suliman and Zhang 2015). In that work, we introduced the line-of-sight DSM (LoS-DSM) solution to overcome most of the limitations in the reviewed methods and provide pixel-by-pixel co-registration. In our recent work, some modifications have been made to the original algorithm where an improved efficient algorithm based on disparity information has been developed (Suliman and Zhang 2016). In this study, we demonstrate the applicability of the recently developed disparity-based LoS-DSM solution for building detection in off-nadir VHR satellite images using stereo- based elevation data. The elevation data generated by the improved algorithm are of minimized terrain-relief effects. Hence, the need for calculating the aboveground building heights should be reduced in the dense urban areas that are characterized by moderate terrain- relief variations. This paper is organized as follows: the proposed method is briefly described in Section 2, the optical data and the achieved results are provided in Section 3, and finally the conclusions are drawn in Section 4. 2. The Proposed Method The method proposed is flowcharted in Figure 1. The method has four steps that are described as follows: