“…Among these studies, model-driven methods assume a building is an assembly of roof primitives (e.g., gable roof and hipped roof), which and whose topology are predefined in a model library (Tarsha-Kurdi et al, 2007). To extract roof primitives from LiDAR point clouds, techniques such as invariant moments (Maas and Vosselman, 1999), graph matching (Oude Elberink and Vosselman, 2009;Verma et al, 2006;Xiong et al, 2014), Support Vector Machine (SVM) (Henn et al, 2013;Satari et al, 2012), RANdom SAmple Consensus (RANSAC) (Henn et al, 2013) and Reversible Jump Markov Chain Monte Carlo (RJMCMC) (Huang et al, 2013) are used. However, these approaches tend to fail when reconstructing complex roof shapes.…”