Interactive semantics-driven reconstruction methodology from Point Clouds

Abstract : 3D point cloud acquisition has become a common practice in the AEC Industry (Architecture, Engineering and Construction). With the emergence of BIM (Building Information Modeling), more and more industrials are interested in creating an "as-built" CAD model of their facilities. The transition from point cloud to a digital model is currently an extremely tedious task due to the large volume of data and the uniqueness of each scanned environment. Many researches are directed towards an automatic extraction of structural and piping elements of increasing complexity. The result however, is a geometry devoid of meaning, or a semantic classification of specific objects in very limited number. In this article, we highlight the context and industrial needs for a generic object reconstruction method and we present the state-of-the art of manual and automatic methods. We then propose a semi-automatic reconstruction method combining semantization with 3D digitization of objects. We review the most recent methods for each operation. Finally, we identify and discuss the technological gaps that need to be addressed in our future research.