This paper presents our approach to improve the 3D-MRDB population techniques in the context of buildings (i.e. to define several geometric representations, different levels of details, of a geographical object). After having discussed our motivations, we present an overview of existing researches relating to the extraction of buildingsÂ’ geometries. A description of the multi-scale pattern concept, developed in our working group to support the extraction of simplified geometries, is also provided. We present afterwards our system approach to improve the detailed geometry extraction automation (through parametric models) while facilitating the extraction of simplified geometries (through multi-scale pattern). Our system architecture, implementing the Instance Driven SASS (Instance Driven Selection of Algorithms Setting and Sources) concept based on a priori knowledge is then described as well as the associated concepts. This semi-automatic approach requests the operator to introduce a priori knowledge; the sources, the algorithms and the parameters are then automatically selected according to the context. Finally, our methodology, aiming at implementing and validating our system, and the progress report of the project are described in the last part of this paper.