Semantically Enhanced 3D Building Model Reconstruction from Terrestrial Laser-Scanning Data

In recent years, three-dimensional (3D) models have been used in a large variety of applications, and the steadily growing capacity in both quality and quantity is increasing demand. To apply new applications to already existing buildings, reconstructed 3D models need to provide the spatial information and semantic and thematic characteristics of target buildings, which are obtained from uninterpreted geometry data sources (i.e., surveying data). This paper proposes a system to automatically reconstruct semantically enhanced 3D building models from terrestrial laser-scanning (TLS) data by combining the strengths of grammars and the maximum a posteriori (MAP) principle. The context-free grammars and rules are predefined and adopted to generate the candidate building models with alternative modeling possibilities. To address the problem of modeling ambiguities, the MAP principle is used to evaluate and control the reconstruction process for both aspects (i.e., goodness of fit and complexity of selected models). Two building models were reconstructed automatically to demonstrate the validation of the authors' proposed method. (C) 2017 American Society of Civil Engineers.