3D Building Model Reconstruction from Multi-view Aerial Imagery and Lidar Data

A novel approach by integrating multi-view aerial imagery and lidar data is proposed to reconstruct 3D building models with accurate geometric position and fine details. First, a new algorithm is introduced for determination of principal orientations of a building, thus benefiting to improve the correctness and robustness of boundary segment extraction in aerial imagery. A new dynamic selection strategy based on lidar point density analysis and K-means clustering is then proposed to identify boundary segments from non-boundary segments. Second, 3D boundary segments are determined by incorporating lidar data and the 2D segments extracted from multi-view imagery. Finally, a new strategy for 3D building model reconstruction including automatic recovery of lost boundaries and robust reconstruction of rooftop patches is introduced. The experimental results indicate that the proposed approach can provide high quality 3D models with high-correctness, high-completeness, and good geometric accuracy. Introduction Accurate and timely updated 3D building model data are very important for urban planning, communication, transportation, tourism, and many other applications. Aerial photogrammetry has been and still is one of the preferred ways to obtain three-dimensional information of the Earth’s surface (Brenner, 2005), and it appears to provide the economic means to acquire truly three-dimensional city data (Foerstner, 1999). Being very well understood and delivering accurate results, its major drawback is that the current state of automation in 3D building model reconstruction is still surprisingly low (Suveg and Vosselman, 2004). Airborne lidar technology has also been an important way for the derivation of 3D building models. A laser scanning system is able to provide data of directly measured three-dimensional points, which is beneficial to improving the automation level in the 3D reconstruction processes. However, the quality of the building models derived from lidar data is restricted by the ground resolution of lidar data. In general, lidar data has reached a one-meter level of spatial resolution, but it is still much lower than the aerial imagery regularly used in photogrammetry with a centimeter level of spatial resolution. It is difficult to obtain building models PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Feb r ua r y 2011 125 Liang Cheng, Manchun Li, and Yongxue Liu are with the Department of Geographical Information Science, Nanjing University, Nanjing, Jiangsu, 210093, China ([email protected]). Jianya Gong is with the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan, Hubei, 430079, China. Photogrammetric Engineering & Remote Sensing Vol. 77, No. 2, February 2011, pp. 125–139. 0099-1112/11/7702–0125/$3.00/0 © 2011 American Society for Photogrammetry and Remote Sensing 3D Building Model Reconstruction from Multi-view Aerial Imagery and Lidar Data Liang Cheng, Jianya Gong, Manchun Li, and Yongxue Liu with highly precise geometric position and fine details only using lidar data. It is clear that photogrammetry and lidar technology are fairly complementary for 3D building reconstruction, and their integration can lead to more accurate and complete products, and a higher automation level of processes (Ackermann, 1999; Baltsavias, 1999). Some typical research associated with 3D building reconstruction by integrating imagery and lidar data have been reviewed in the following section; however, for most of the existing approaches, it is difficult to determine highly accurate and detailed building models. Because most of the existing approaches did not strive for reconstruction of highly accurate 3D building boundaries, their common idea is to determine approximate building boundaries from lidar data, then to refine them using image information. Its main limitation is the unstable quality of the approximate boundaries determined by lidar data. The quality of these boundaries is significantly influenced by the quality of lidar data filtering processing and irregular point spacing. Meanwhile, the process of robust reconstruction of building roof patches based on lidar data is still very difficult. Therefore, a novel approach for the reconstruction of 3D building models with precise geometric position and fine details by integrating multi-view aerial imagery and lidar data is proposed. It consists of three main steps: 2D building boundary extraction, 3D building boundary determination, and 3D building model reconstruction. Related Works The research of 3D building reconstruction from imagery and lidar data has been increasingly focused. Schenk and Csatho (2002) proposed a method on the establishment of a common reference frame between lidar data and aerial imagery by utilizing sensor-invariant features, such as breaklines and surface patches for a richer and more complete surface description. Mcintosh and Krupnik (2002) merged edges detected and matched in aerial images and the digital surface model produced from airborne lidar data to facilitate the generation of an accurate surface model, which provides a better representation of surface discontinuities. Rottensteiner and Briese (2003) detected building regions and roof planes from lidar data, and then grouped these roof planes to create polyhedral building models, in which the shapes of the roof plane boundaries were improved by an overall adjustment integrating aerial imagery information. Nakagawa